AI StocksLeaderboardCapexExposureFactsMembership

ALLFACTSAI · The Intelligence Club

The AI boom
decoded
for investors

Daily verified AI facts. Investing intelligence. Model leaderboards. And by email, free: The Daily Receipts every weekday, The Weekly Brief every Saturday.

Get them free▶ Watch the trailer
AI data centres worldwide reached a power capacity comparable to New York State at peak demand in 2025, which Stanford's AI Index puts at 29.6 gigawattsStanford HAI, AI Index 2026 · 2025Frontier AI models are catching up fast to benchmarks meant to resist them. A year ago they scored under 10% on Humanity's Last Exam, a 2,700-question test designed to be hard for AI and favorable to human experts. Now accuracy has reached 38.3% on Humanity's Last Exam, up from under 10% a year earlierStanford AI Index 2026, citing CAIS · 2026 AI Index ReportThe information sector — telecom, media, tech services — led US corporate AI hiring in 2025. Lightcast data cited in Stanford's 2026 AI Index found AI skills required in 13.2% of the sector's job postings, up from 7.8% in 2024Lightcast, via Stanford AI Index · 2025 data, pub. 2026Applied Materials sells the deposition, etch and inspection machines chipmakers must install before a single AI chip exists, so its results run a step ahead of the chip cycle. Its third fiscal quarter, ended 26 July 2026, was the biggest in the company's history: Applied Materials booked a record $9.12bn of revenue, up 25% year over yearApplied Materials 8-K (SEC) · Quarter ended 26 July 2026AMD's Data Center segment, the one selling Instinct GPUs and EPYC server chips, took $16.6 billion of the company's record $34.6 billion of revenue in 2025 — 48% of all AMD revenueAMD FY2025 results · year ended 27 Dec 2025Amkor Technology, the outsourced chip packaging and test house scaling up advanced packaging for AI processors, reported record second-quarter 2026 net sales of $1.90 billion, up 26% year-over-year $1.90 billion in record Q2 net sales, +26% YoYAmkor Technology · Q2 2026Amazon's cash capital expenditure across the whole of 2025 came to $131.81 billion in one yearAmazon 10-K, SEC EDGAR · FY2025Amazon's most recently reported quarterly cash capital expenditure reached $54.21 billionAmazon, SEC EDGAR (XBRL filings) · Q2 2026Amazon's cash capital expenditure in the first quarter of 2026 was $44.2 billion in a single quarterAmazon 10-Q, SEC EDGAR · Q1 2026Arista Networks, whose switches carry AI training and inference traffic inside hyperscaler data centres, reported total revenue for the quarter to 30 June 2026 of $3.036 billion, its first quarter above $3 billion and up 37.7% year-over-yearArista Networks · Q2 2026Applied Digital had signed long-term hyperscaler leases for 1,410 megawatts of AI data-centre capacity across five campuses as of May 31, 2026, worth approximately $36 billion in contracted revenue over the initial 15-year lease terms 1.41 gigawatts of AI data-centre capacity leased for $36 billionApplied Digital Corp. · as of May 31, 2026Arm's total royalty revenue grew 11% year-over-year to $671 million in its fiscal fourth quarter of 2026, and the company said its data centre royalty revenue within that more than doubled year-over-year on rising Arm-based server chip deployments by hyperscalers Arm data-centre royalties more than doubled YoYArm Holdings · Q4 FY2026Every advanced AI chip needs packaging and testing before it ships, and ASE Technology, the world's largest outsourced semiconductor packaging and test house, is racing to add capacity for it. Equipment spending in its packaging and test segment jumped in the quarter to 30 June 2026, from $942 million a year earlier to $1.646 billion in ASE's advanced-packaging equipment spending, up 74.7% year-over-yearASE Technology Holding · Q2 2026Every advanced AI chip is printed on an ASML machine, and the company says AI demand has its customers accelerating their expansion plans. On 15 July 2026 ASML said it would lift its 2026 low-NA EUV capacity of around 65 systems, for 2027, by 30% more EUV machines a yearASML Q2 2026 results · 15 July 2026Astera Labs makes the connectivity chips that link AI accelerators inside data-centre racks, and demand pushed its revenue for the quarter ended 30 June 2026 up 104% year-on-year to $392.4 millionAstera Labs 10-Q (SEC) · Q2 2026Broadcom reported $20 billion of AI revenue in its 2025 financial year, out of $63.89 billion in total — 31.3% of everything it soldBroadcom FY2025 Form 10-K · FY2025Broadcom's entire 2025 financial year produced $20 billion of AI revenue. In the single quarter to 3 May 2026, its AI semiconductor revenue — custom accelerators and AI networking — reached $10.8 billion in one quarter, up 143% year on yearBroadcom · Q2 FY2026 (to 3 May 2026)Baidu's AI Cloud Infra revenue grew 50% year-over-year in the quarter to 30 June 2026, powered by GPU Cloud demand up 283%, and the company poured capital into keeping pace. Group-wide capital expenditure nearly tripled year-over-year to RMB11.39 billion ($1.68 billion), Baidu's Q2 2026 capital expenditure, up 199.7% year-over-yearBaidu · Q2 2026BLS's 2024-34 employment projections show AI adoption fueling outsized growth in the occupations that build and secure it: data scientist jobs are projected to grow 33.5%, more than ten times the 3.1% projected growth rate for all US occupations combined 33.5% projected growth in US data scientist jobs through 2034U.S. Bureau of Labor Statistics · 2024-34 projectionsCadence Design Systems, whose software chipmakers use to design AI accelerators, reported $1.584 billion of revenue in its second quarter of 2026, up 24.2% from $1.275 billion a year earlier, with quarter-end backlog reaching a record $8.1 billion as CEO Anirudh Devgan cited accelerating demand for AI-driven design tools Cadence's Q2 2026 revenue: $1.584 billion, up 24.2% year-over-yearCadence Design Systems · Q2 2026Long-dated nuclear supply is being locked up now for demand that only arrives at the end of the decade. In the quarter to 30 June 2026 Constellation Energy signed power purchase agreements running 15 to 20 years with investment-grade customers, beginning between 2029 and 2032: Constellation signed another 920MW of long-term nuclear supply, none of it starting before 2029Constellation Energy 8-K (SEC) · Q2 202644% of US firms using AI report it augmenting workers' existing tasks rather than replacing them, versus just 5% reporting AI substituting tasks outright, according to Census Bureau research on business AI adoption 44%U.S. Census Bureau · 2026 BTOS AI supplementThe Census Bureau's Business Trends and Outlook Survey found AI adoption climbing fastest among big employers: as of May 2026, 37% of firms with 250+ employees used AI vs 32% for 100-249 employees and under 20% for the smallest firmsUS Census Bureau · May 202655% of U.S. workers say they have used AI on the job for at least one of 11 listed tasks, and among those who used it in the past week, 31% said it saved them one to two hours of work, per the Census Bureau's March 2026 Household Trends and Outlook Pulse Survey 55% of U.S. workers have used AI on the job, Census Bureau survey findsU.S. Census Bureau · March 2026 HTOPSNearly 750 executives in the Duke/Federal Reserve CFO Survey expect AI to barely touch total headcount — aggregate employment down less than 0.4% in 2026. What shifts is the mix, over the next three years: routine clerical work down more than 2 percentage points of employment shareNBER w34984, The CFO Survey · 2026Q1 survey waveChatGPT was estimated to have reached 100 million monthly users roughly two months after launchUBS, reported by The Register · February 2023The total computing power of the stock of AI chips is growing about 3.4× per yearEpoch AI · 2026A UC Berkeley, Sao Paulo and Stanford study of 141 countries found that national Claude usage rises 12x for every 0.10-point increase in a country's AI exposure score, based on Anthropic's Economic Index data for February 2026 Claude usage rises 12x per +0.10 rise in national AI exposureUC Berkeley/Stanford study · arXiv, Aug 2026Epoch AI asked LLM judges how long an experienced engineer would need, without AI, to reproduce each merged pull request from the 41 collaborators on OpenAI's public Codex repository. In the second quarter of 2026, work beyond what anyone could finish in a single unassisted day accounted for 8% of contributor-days, up from 2% a year earlierEpoch AI · Q2 2026Coherent's Datacenter & Communications segment, the optical transceivers that link GPUs inside AI data centers, brought in $1.615 billion in the quarter to 30 June 2026, up 58.6% from $1.018 billion a year earlier as demand for 800G and 1.6T transceivers accelerated Coherent's AI-datacenter segment revenue: $1.615 billion, up 58.6% year-over-yearCoherent Corp · Q4 FY2026 (June 2026)Coherent's Datacenter & Communications segment, which supplies optical transceivers for AI data centers, generated $1.62 billion in revenue in the quarter ended June 30, 2026, up 58.6% from a year earlier $1.62 billionCoherent Corp. · Q4 FY2026CoreWeave rents AI computing power by the hour, and demand is not the problem: first-quarter 2026 revenue rose 112% year on year to $2.08 billion. The company still reported a $740 million net loss for the quarterCoreWeave · Q1 2026Core Scientific has contracted about 1.1 gigawatts of leased data-center power capacity to AI and HPC customers, including CoreWeave and AMD, representing more than $24 billion of potential revenue, the company said in its second-quarter 2026 results Core Scientific's leased AI data-center capacity: 1.1 gigawatts, over $24 billion in contractsCore Scientific · Q2 2026Credo Technology makes the active electrical cables and optical DSPs that connect AI accelerators inside hyperscaler racks, and demand for that interconnect gear pushed its revenue for the quarter ended 26 April 2026 up 157.0% year-over-year to $437.0 million in quarterly revenueCredo Technology · Q4 FY2026Salesforce brands its whole product line around AI agents and reports separately what customers actually pay for them. In the quarter to 30 April 2026, on total revenue of $11.13 billion, that came to $1.2 billion of Agentforce annual recurring revenue, up 205% year on yearSalesforce Q1 FY2027 results · Quarter to 30 April 2026Salesforce brands its whole product line around AI agents and reports separately what customers actually pay for them. In the quarter to 31 July 2026, Agentforce annual recurring revenue exceeded $1.5 billion in Agentforce ARR, up over 240% year-over-yearSalesforce Q2 FY2027 results · Q2 FY2027Salesforce's quarter to 31 July 2026 brought total revenue of $11.3 billion, up 11% year-over-year, with subscription and support revenue of $10.8 billion providing the bulk of the growth $11.3 billion in Q2 FY2027 revenueSalesforce · Q2 FY2027CoreWeave rents out AI computing capacity, and building that capacity is capital-intensive. Its first-quarter 2026 cash flow statement shows capital expenditure on property and equipment, including capitalised internal-use software, jumped to $7.7 billion in Q1 2026 capital spending, a 5.5x jump from $1.4 billion a year earlierCoreWeave · Q1 2026Dell's quarter to 28 May 2026 was the one where AI servers took over the business: total revenue hit a record $43.8 billion, up 88% year on year, and of that, $16.1 billion came from AI servers aloneDell Technologies 8-K · Q1 FY2027, ended 28 May 2026Data-center landlords are racing to keep pace with AI and cloud demand. Digital Realty's second-quarter 2026 financial supplement to investors disclosed a global construction pipeline of Digital Realty's 1,402 megawatts of IT capacity under construction, 54% already pre-leasedDigital Realty Q2 2026 supplement · Q2 2026As cloud providers shifted from AI training to inference, conventional DRAM contract prices jumped 93% to 98% in a single quarter — and on TrendForce's count of the first quarter of 2026, global DRAM revenue hit $97 billion, up 81%TrendForce · Q1 2026Epoch AI's Capabilities Index scores frontier models on one aggregate scale, and fits a break in the trend at April 2024, when reasoning models and reinforcement learning took over frontier training. Before the break the frontier gained 8.3 index points a year; after it, 15.5 points a year, 1.85 times as fastEpoch AI · December 2025NBER asked 69 economists what the US economy would look like in 2050 if AI capabilities advance rapidly. Their median answer was 3.5% annual GDP growth, alongside a labour force participation rate falling from today's 62% to 55.0% by 2050, the level before women entered the workforce en masseNBER working paper 35046 · April 2026AI data centres are pushing US electricity generation to record highs. Total output from the country's power sector in the first half of 2026 rose 1.8% versus the first half of 2025, adding 37 billion kilowatt-hoursEIA, Short-Term Energy Outlook · Aug 2026 STEOEIA modelled what would happen if data-centre power demand grew faster than forecast. Its February 2026 baseline had US coal generation falling 9.3% between 2025 and 2027; in the higher-demand scenario, coal plants run on their existing spare capacity and the decline shrinks to a 5.0% fall in US coal generation instead of 9.3%EIA Short-Term Energy Outlook analysis · March 2026EIA's Annual Energy Outlook 2026 estimates that data-centre servers alone, before any cooling or other building load, already accounted for 7% of all US commercial-sector electricity in 2025US Energy Information Administration · AEO2026, May 2026On 3 August 2026 the governor of Texas paused new data-centre development while projects already under review are assessed. In the same month's Short-Term Energy Outlook the EIA cut its forecast for Texas electricity load growth in 2027 to 6% growth in Texas load, down from 14% forecast a month earlierEIA Short-Term Energy Outlook · August 2026Texas paused new data-centre development on 3 August 2026 while it reviews the projects already in the queue, and the EIA cut its forecast for the state in response: Texas electricity load growing 6% in 2027, not the 14% forecast a month earlierEIA Short-Term Energy Outlook · August 2026AI inference is consuming NAND flash faster than suppliers can allocate it. Enterprise SSD contract prices rose roughly 80% in the first quarter of 2026, and on TrendForce's count the top five brands' combined revenue hit a record $18.46 billion — enterprise SSD revenue rose 86.1% in a single quarterTrendForce · Q1 2026The 84 AI data centers tracked by Epoch AI have a combined real-world power footprint of 16.9 gigawatts as of August 2026, more than New York City's 11 GW peak demand 16.9 GWEpoch AI · August 2026Epoch AI tested three commercial AI detectors — Pangram, GPTZero, Originality.ai — pinned to their June 2026 versions. Against plain machine-generated text they catch nearly everything, but shown five samples of a specific author's writing and told to imitate the style, AI detectors missed 13% of style-imitated AI passages, versus under 1% of plain AI textEpoch AI · Jul 2026Epoch AI rebuilt the cost structure of three AI companies — Anthropic, MiniMax and Z.ai — from IPO filings and reporting. Staff, despite the industry's headline pay packages, take under a quarter of spending. Research and inference compute together take 54% to 62% of an AI company's costsEpoch AI · 2025 accountsEpoch AI fits a trend through the models whose context windows ranked among the ten longest on their release date, and finds that since mid-2023 the frontier context window has been getting about 30× longer every yearEpoch AI · since mid-2023Frontier models have started finding software flaws at scale. Counting the high- and critical-severity CVEs disclosed by 21 major vendors — Microsoft, Google, Apple and Cisco among them — Epoch AI puts July 2026 at about 2,500 serious CVEs in July 2026, five times the monthly record set before April 2026Epoch AI · July 2026Epoch AI tracked 333 compute estimates for notable AI models released between 2010 and May 2024, and found that training compute for the frontier — the top 10 models by compute at release — has been growing each year to reach 5.3x the prior year's levelEpoch AI · 2010–May 2024 dataEpoch AI audited its FrontierMath benchmark after OpenAI flagged suspiciously high scores, and its 12 June 2026 v2 release corrected errors in 42% of the original problems, leaving a verified set of 338 problems across four tiers 42% of FrontierMath's original problems contained errorsEpoch AI · June 2026 (v2 release)AI capability is getting cheaper to reach, not just bigger. Epoch AI tracked what it cost in tokens to hit the same score — about 27% accuracy on its FrontierMath benchmark — first with o4-mini in April 2025, then with GPT-5.2 in December 2025, and found AI capability got roughly 3x cheaper to reach in eight months, part of a ~5-10x per year cost-decline trendEpoch AI · Feb 2026Epoch AI and Ipsos asked 469 US workers who had used AI for work in the previous seven days whose subscription they were on. In computer, engineering and science jobs 67% use a tool their employer provides — everywhere else the free tier wins: 46% of US workers using AI at work are on a free plan, against 37% given a tool by their employerEpoch AI / Ipsos · Fieldwork 10-19 July 2026Epoch AI and Ipsos asked 1,106 employed US adults in July 2026 whether AI had taken over work they used to hand to another person. Reporting that AI now handles at least one task they previously passed to a coworker or contractor: one in five US workersEpoch AI / Ipsos survey · July 2026Eaton's Electrical Global segment, which sells the power distribution and switching gear AI data centres depend on, ended the second quarter of 2026 with total backlog up 103% year-over-year Eaton Electrical Global backlog up 103% year-over-yearEaton · Q2 2026Asked what three years of AI had actually changed inside their own firms, nearly 6,000 senior executives across the US, UK, Germany and Australia mostly reported nothing: nine in ten saw no impact on employment or productivityNBER working paper 34836 · February 2026Of the notable AI models released in 2025, Stanford's AI Index counts over 90% as coming from industryStanford HAI, AI Index 2026 · 2025Across 42 notable language models, Epoch AI estimates frontier training costs have risen from roughly $2 million for GPT-3 to up to nearly $390 million per runEpoch AI · 2024Epoch AI's Open Problems benchmark collects research mathematics problems that have resisted serious attempts by professional mathematicians. Expanded to 50 problems on 31 July 2026, the count two weeks later — with the newest AI solution still marked provisional — stood at 4 solved by AI, 1 by a human, 45 still openEpoch AI FrontierMath · 17 August 2026Alphabet told investors in its second-quarter 2026 results that Gemini models now process 22 billion API tokens a minute, and that the Gemini app has reached 950 million monthly active usersAlphabet 8-K · Q2 2026In a randomised field experiment across 66 firms and 7,137 knowledge workers, a generative AI tool built into everyday work apps saved the people who used it two hours a week on email, with no detected change in their tasksNBER working paper 33795 · revised Nov 2025Mentions of generative AI skills in US AI job postings grew 111% from 2024 to 2025, even as their share of total AI postings slipped, as employer demand pivots toward agentic-AI skills, per Lightcast data cited in Stanford's AI Index 111%Stanford HAI · 2025Across 5,179 customer-support agents, a generative AI assistant raised issues resolved per hour by 14% on average — and by 34% for the least experienced staffBrynjolfsson, Li & Raymond, NBER · April 2023GE Vernova builds the gas turbines and grid equipment that electricity supply runs on, and AI data centres are now one of its largest sources of new demand. Its gas equipment under contract rose from 100 GW to 116 GW during the second quarter of 2026, and data-centre orders through the first half of the year reached over $5 billion of data-centre orders, more than double the whole of 2025GE Vernova Q2 2026 results · First half of 2026Alphabet's most recently reported quarterly cash spending on property and equipment reached $44.92 billionAlphabet, SEC EDGAR (XBRL filings) · Q2 2026Alphabet's cash capital expenditure in the first quarter of 2026 reached $35.67 billion in a single quarterAlphabet 10-Q, SEC EDGAR · Q1 2026Alphabet's quarterly cash capital expenditure has roughly tripled in two years, from $12.01 billion in the first quarter of 2024 to $35.67 billion in the first quarter of 2026Alphabet 10-Q, SEC EDGAR · Q1 2024 - Q1 2026Epoch AI estimates the single largest owner of AI computing power worldwide is not a chipmaker or a cloud reseller but Google, which held, measured in H100-equivalents as of the fourth quarter of 2025, about one quarter of all global AI compute, mostly from its own custom TPU chips rather than Nvidia GPUsEpoch AI · Q4 2025Epoch AI's Capabilities Index scores language models on one aggregate scale, and nothing since has held first place as long as GPT-4 did after its March 2023 release — roughly a year. The second-longest lead, by OpenAI's o1, lasted a little over three months, less than a third of GPT-4's run at the topEpoch AI Capabilities Index · July 2026Traditional computing powers like Germany, Japan and France now play marginal roles in AI infrastructure. Epoch AI's tracking of global GPU cluster performance in May 2025 put China second at 15%, behind a United States hosting about three-quarters of the world's GPU cluster performanceEpoch AI · May 2025High-bandwidth memory is taking over the memory industry's capacity: TrendForce expects HBM to account, among the top three DRAM suppliers by the end of 2026, for 22% of DRAM wafer input, up from 18% a year earlierTrendForce · June 2026 projectionAmazon, Google, Meta, Microsoft and Oracle together held, measured in H100-equivalents as of Q4 2025, an estimated 71% of the world's AI computeEpoch AI · Q4 2025Epoch AI models the world's AI inference capacity, at fixed model size and context length, as growing 3.4× a year — but it puts token demand growth at around 10× a year, three times faster than supplyEpoch AI · May 2026Intel's Data Center and AI segment, the unit selling Xeon server chips and Gaudi/Crescent Island AI accelerators, reported revenue for the quarter ended 27 June 2026 up 59% year-over-year to $6.3 billion in quarterly segment revenueIntel Corporation · Q2 2026IREN, the bitcoin miner turned AI cloud operator, said its AI Cloud Services revenue grew roughly 8x year-over-year to $128.8 million in the fiscal year ended June 30, 2026, and the company now has $4 billion of ARR contracted for 2026 GPU capacity, with $1 billion already operatingIREN FY26 results release · FY2026IREN, a bitcoin miner turned AI cloud operator, told investors in its quarter ended 31 March 2026 that it now holds executed grid connection agreements across seven data-centre sites in Texas, Oklahoma and British Columbia for a combined 4,510 megawatts of AI data-centre power capacityIREN Ltd · Q3 FY2026 10-QEpoch AI tracks the largest AI data centre in the world by computing capacity, and the record has changed hands repeatedly since August 2024 — Colossus, Anthropic and Amazon's New Carlisle, Microsoft's Fairwater Atlanta, Meta's Prometheus. Each time, the bar has risen at the same pace: a doubling every seven monthsEpoch AI · June 2026The largest AI data centre in Epoch AI's tracker, Colossus 2 in Memphis, Tennessee, holds a computing capacity of 1,112,000 H100-equivalent AI chipsEpoch AI · AI data centers tracker, August 2026Nvidia's accelerators have pushed thermal design power from 700W for the H100 to over 1,000W for the B200 and B300, and air can no longer carry the heat away. TrendForce expects liquid cooling in server racks to reach 47% of server racks by 2026TrendForce · November 2025 forecast for 2026Lam Research, whose machines etch and deposit the layers inside AI chips, reported fiscal 2026 revenue of $23.23 billion — up 26% from $18.44 billion in fiscal 2025 — as CEO Tim Archer credited AI-driven demand for record margins and EPS $23.23 billion in fiscal 2026 revenueLam Research · FY2026 (ended June 28, 2026)A McKinsey survey of employers cited in Stanford's 2026 AI Index found anticipated AI-driven workforce cuts concentrated in service operations, supply chain and software engineering, while almost half of organizations expect little or no change overall. Still, one-third of employers expect AI to shrink their workforce in the year aheadMcKinsey survey (Stanford AI Index 2026) · 2025 surveyMeta's purchases of property and equipment nearly doubled in 2025, from $37.3bn to $69.7bn — 34.7% of its revenueMeta FY2025 Form 10-K · FY2025Meta's cash capital expenditure went the other way in the first quarter of 2026 — from $21.38 billion the quarter before, a fall to $19.0 billionMeta 10-Q, SEC EDGAR · Q1 2026Meta's most recently reported quarterly purchases of property and equipment reached $30.12 billionMeta, SEC EDGAR (XBRL filings) · Q2 2026In a randomised trial, 16 experienced open-source developers believed AI tools had made them 20% faster. Measured against issues they finished without AI, they were 19% slowerMETR · July 2025Micron breaks out the memory it sells into AI data centres as its own business unit. In the quarter to 28 May 2026 that unit took in $11.52 billion, against $1.53 billion in the same quarter a year before — Micron's core data centre revenue grew 7.5× in a yearMicron 8-K, SEC · quarter to 28 May 2026Monolithic Power Systems makes the chips that step grid power down to the voltages AI GPUs need. Its Enterprise Data segment, which sells into AI servers, grew far faster than the rest of the company in the quarter to 30 June 2026, when total revenue hit a record $980.6 million and Enterprise Data segment revenue reached $380.6 million in AI server chip revenue, up 164.3% year-over-yearMonolithic Power Systems · Q2 2026Marvell designs the custom accelerators, optics and switching silicon that wire AI clouds together, and that business has crowded out everything else it sells. In the quarter to 2 May 2026, data centre revenue of $1.83 billion out of $2.42 billion total came to 76% of Marvell's revenueMarvell Technology · Q1 FY2027 (to 2 May 2026)Marvell's record quarter to 2 May 2026 looks very different depending on which profit line you read. Non-GAAP net income was $718.0 million. Under GAAP — after $207.6 million of stock compensation, $225.2 million of amortisation on acquired intangibles and a $250.7 million contingent-consideration revaluation — net income was $34.5 million, under 5% of the adjusted figureMarvell Technology · Q1 FY2027 (to 2 May 2026)Marvell's revenue hit a record $2.74 billion in the quarter to 1 August 2026, up 37% year-on-year, led by an AI data-centre business that grew even faster — accelerating to 46% year-on-year, to $2.17 billion in Data Center revenueMarvell Technology, Inc. · quarter ended 1 August 2026Marvell's data infrastructure chip business set a fiscal second-quarter revenue record in the three months to 1 August 2026, with net revenue up 37% year-over-year to $2.739 billion, as Data Center segment revenue growth accelerated to 46% $2.739 billion in record Q2 revenueMarvell Technology · Q2 FY2027Microsoft told investors in April 2026 that its AI business had reached an annual run rate of $37 billionMicrosoft FY2026 Q3 earnings call · April 2026Microsoft's cash capital expenditure rose by more than half between the third and fourth quarters of 2025, reaching $29.88 billion in a single quarterMicrosoft 10-Q, via SEC XBRL · Q4 2025Microsoft's most recently reported quarterly cash spending on property and equipment reached $35.80 billionMicrosoft, SEC EDGAR (XBRL filings) · Q2 2026Microsoft's cash spending on property and equipment reached, in the first quarter of 2026 alone, $30.88 billionMicrosoft 10-Q, via SEC XBRL · Q1 2026A randomised online experiment gave 1,174 adults aged 25 to 45 an incentivised business problem-solving task, some with a generative-AI assistant and some without. Unaided, the higher-education group beat the lower-education group by 0.548 standard deviations. With AI the gap fell to 0.139 — about three-quarters of the education gap closedNBER Working Paper 34851 · May 2026 revisionMenaka Hampole and co-authors linked resume data to firms' AI-application adoption to isolate AI's true effect on jobs, not just correlation. In their preferred instrumental-variables specification, a one-standard-deviation rise in an occupation's AI exposure cuts its within-firm employment share by 14%NBER · Working Paper 33509, 2025Nebius rents out AI cloud capacity and is scaling from a small base. Revenue in the second quarter of 2026 reached $582.3 million against $105.1 million in the same quarter of 2025 — a 454% year-on-year rise in revenueNebius Group · Q2 2026In July 2025, Nvidia became the first US company in history to reach a $4 trillion market capitalisationAl Jazeera · July 2025Nvidia's Data Center segment brought in $193.7bn of the $215.9bn it made in its 2026 financial year — 89.7% of the company's revenueNvidia FY2026 Form 10-K · FY2026, ended 25 January 2026Nvidia's Data Center segment generated $89.0 billion in revenue in its fiscal second quarter ended July 26, 2026, up 117% from a year earlier and now 92% of total company revenue $89.0 billionNvidia · Q2 FY2027Nvidia reported revenue of $96.2 billion for its fiscal second quarter ended July 26, 2026, up 106% from a year earlier and 18% from the prior quarter $96.2 billionNvidia · Q2 FY2027Nvidia spent $18.5 billion on research and development in its 2026 financial year — three times its $6.04 billion of capital expenditureNvidia FY2026 Form 10-K · FY2026Onto Innovation's inspection systems qualify the 2.5D packaging and high-bandwidth memory built into every AI accelerator, and that demand pushed its Specialty Devices and Advanced Packaging segment to an all-time high. Total company revenue in the quarter ended 30 June 2026 reached $343.1 million in quarterly revenue, up 35.3% year-over-yearOnto Innovation · Q2 2026Across January to May 2026, Epoch AI's capabilities index puts the best open-weight models behind the frontier closed models by an average of four monthsEpoch AI · Jan-May 2026Oracle's remaining performance obligations — cloud work it has signed but not yet delivered — stood, when its financial year closed on 31 May 2026, at $638 billion of contracted revenue, up 363%Oracle · Q4 FY2026, ended 31 May 2026Oracle's cash capital expenditures jumped from $21.2 billion in fiscal 2025 to $55.7 billion in fiscal 2026, an increase the company attributes directly to the expansion of its data centres $55.7 billion in Oracle FY2026 capex, up from $21.2 billionOracle · FY2026Stanford's AI Index reports that organisational AI adoption kept climbing in 2025, reaching 88% of surveyed organisationsStanford HAI AI Index · 2025 figureAI chip performance per dollar has been improving by approximately 1.37× per yearEpoch AI · 2026PJM's Dominion zone in Virginia serves the largest concentration of data centres in the world, and their power draw shows up hardest on the coldest nights. Winter peak electricity demand there reached 25,413 megawatts in the 2025–26 winter — 45% higher than the 2019–20 winterEIA · May 2026Palantir positions its whole product line as an AI platform but reports no AI revenue at all — its 2025 annual report splits revenue 54% government, 46% commercialPalantir FY2025 Form 10-K · FY2025Scale is not the only driver of AI progress. Epoch AI measures how much training compute it takes to reach a fixed level of language-model performance, and finds that requirement falling fast: pre-training compute efficiency doubles every 7.6 monthsEpoch AI · Feb 2026Quanta Services builds the transmission lines, substations and high-voltage switchgear that connect new AI data centres to the grid, and it told investors that rising demand from data centres, electrification and grid modernisation is driving orders. Its total backlog reached a record $53.4 billion, Quanta Services' record grid-construction backlog as of 30 June 2026Quanta Services · Q2 2026Samsung is racing to keep pace with HBM4 demand for Nvidia's next GPU generation, and TrendForce reports the company plans to lift its HBM wafer output from about 170,000 wafers a month to roughly 250,000 wafers a month by the end of 2026, a jump of about 47%TrendForce · Dec 2025US cloud providers signed multi-year supply agreements that stop memory makers raising their prices, so TrendForce expects the third-quarter 2026 increase to fall on everyone else: server DRAM contract prices up another 13–18% in one quarterTrendForce · 9 July 2026ServiceNow sells AI as an add-on to the workflow software companies already run, and reports separately what customers pay for it. Agentic deployments rose ninefold in nine months, and in the second quarter of 2026 ServiceNow AI reached $1 billion in annual contract valueServiceNow Q2 2026 results · Q2 2026Soaring AI-server memory prices lifted SK hynix to 79.32 trillion won of revenue and 60.54 trillion won of operating profit in the second quarter of 2026 — an operating margin of 76%SK hynix · Q2 2026SpaceX now reports an AI segment alongside Space and Connectivity. In the quarter to 30 June 2026, advertising on X brought in $367 million — less than a year earlier — while the segment's AI solutions and infrastructure line, covering Grok API access, subscriptions, data licensing and cloud, reached $2.19 billion, seven times its level a year earlierSpaceX 10-Q · Q2 2026SpaceX listed on Nasdaq in June 2026, and its first quarterly report as a public company discloses a metric no rocket company used to publish: installed GPUs multiplied by their all-in power draw, before any cooling or facility overhead. At 30 June 2026 that came to 1.4 gigawatts of installed GPU power draw, 3.5× a year earlierSpaceX 10-Q · 30 June 2026Super Micro Computer assembles and ships AI servers rather than designing the chips inside them. For the financial year ended 30 June 2026 it reported net sales of $39.1 billion, up from $22.0 billion a year earlierSuper Micro Computer 8-K · FY ended 30 June 2026AI data centres need firm, round-the-clock power, and nuclear plants are being locked up years in advance to supply it. In June 2025, independent power producer Talen Energy expanded its agreement with Amazon Web Services to deliver up to 1,920 megawatts of carbon-free nuclear power from the Susquehanna plant through 2042Talen Energy · June 2025Teradyne, whose systems test semiconductor chips before they ship, reported quarterly revenue of $1.33 billion for the quarter ended June 28, 2026, up 104% from $652 million a year earlier, with $1.12 billion of that revenue in its Semiconductor Test segment Teradyne's revenue up 104% year-over-year to $1.33 billionTeradyne Inc. · Q2 2026Tesla's own quarterly filing attributes its capital spending mainly to global AI and operational infrastructure. Across the first half of 2026 that spending more than doubled against the same six months of 2025, from $3.89 billion to $8.28 billion of capital expenditureTesla Form 10-Q · Q2 2026Tesla reports its AI training compute the way it reports factory capacity — in installed megawatts. At the end of the second quarter of 2026 both Texas clusters, Cortex 1 and Cortex 2, were in production, and the compute installed there had more than doubled in six months, to over 205 MW of AI training capacityTesla Q2 2026 Update · Q2 2026Tesla's driver-assistance software is one of the few AI products consumers pay for directly, and the company discloses the subscriber count every quarter. At the end of the second quarter of 2026 it reported 1.48 million active FSD subscriptions, up 56% in a yearTesla Q2 2026 Update · Q2 2026Compute used to train frontier language models has grown roughly 5× per year since 2020Epoch AI · 2026Orders for Nvidia rack-scale platforms and in-house cloud ASICs came in stronger than expected, so in August 2026 TrendForce revised its forecast for this year's global AI server shipments up from 28% growth to nearly 31% growth year-on-yearTrendForce · August 2026TrendForce forecasts that global foundry industry revenue will grow 24.8% year-over-year to approximately $218.8 billion in 2026, with TSMC posting the largest increase of around 32% Global foundry revenue forecast to grow 24.8% to $218.8 billionTrendForce · published March 2026AI servers have repriced memory faster than any forecast could keep up with. In a single quarter TrendForce raised its estimate for 2026 global memory market revenue from $551.6 billion to $889.3 billion — 61% above its own forecast from three months earlierTrendForce · May 2026 forecastTSMC's high performance computing platform, the category that carries AI accelerators, accounted for 58% of its 2025 revenueTSMC 4Q25 management report · 2025In the first quarter of 2026, TSMC's share of revenue across the world's ten largest foundries reached 72%TrendForce · Q1 2026Stanford's AI Index found the United States trailing most of the world's advanced economies on population-level AI adoption. Citing Microsoft AI Economy Institute telemetry for the second half of 2025, the index ranked the US 24th out of the top 30 economies worldwide, with an AI usage rate of 28.3% of Americans, still in the bottom half of the global rankingStanford AI Index (Microsoft data) · H2 2025Across US businesses surveyed between December 2025 and May 2026, AI use ran between 17% and 20% — and even among firms with 250 or more employees, only 37% reported using AIUS Census Bureau, BTOS · December 2025 – May 2026Counting the notable AI models released in 2025, Stanford's AI Index puts the United States on 59, against China's 35Stanford HAI, AI Index 2026 · 2025US data centres used 176 terawatt-hours of electricity in 2023 — on Berkeley Lab's measurement, 4.4% of the country's totalLawrence Berkeley National Laboratory · 2023Berkeley Lab projects that by 2028 US data centres will draw between 6.7% and 12% of all US electricityLawrence Berkeley National Laboratory · Projection to 2028Generative AI reached 53% adoption in three years, faster than the personal computer or the internet — but unevenly: Singapore sits at 61% and the UAE at 64%, while Stanford's AI Index finds the US is 24th for generative AI adoption, at 28.3%Stanford AI Index · 2026 reportEIA forecasts US electricity use rising 1% in 2026 and 3% in 2027, driven largely by large computing facilities including data centres — which would be the first four-year run of demand growth since 2007EIA Short-Term Energy Outlook · Jan 2026Vertiv, which builds data-centre power and cooling, ended 2025 with an order backlog of $15.0 billion — up 109% in a yearVertiv Q4 2025 results release · Q4 2025Vistra signed 20-year power purchase agreements with Meta Platforms in January 2026 to supply a total of 2,609 MW of carbon-free power and capacity from its Perry, Davis-Besse and Beaver Valley nuclear plants in PJM, with full delivery of operating capacity by 2027 and uprate capacity by 2034 2,609 MW of nuclear power under 20-year Vistra-Meta PPAsVistra Corp. · January 2026Stanford's AI Index, citing Brynjolfsson et al., finds AI's labour-market effect landing on hiring pipelines rather than on broad job losses. By September 2025, employment for US software developers aged 22 to 25 had fallen close to 20% from its 2022 peakStanford AI Index · 2026 reportAI data centres worldwide reached a power capacity comparable to New York State at peak demand in 2025, which Stanford's AI Index puts at 29.6 gigawattsStanford HAI, AI Index 2026 · 2025Frontier AI models are catching up fast to benchmarks meant to resist them. A year ago they scored under 10% on Humanity's Last Exam, a 2,700-question test designed to be hard for AI and favorable to human experts. Now accuracy has reached 38.3% on Humanity's Last Exam, up from under 10% a year earlierStanford AI Index 2026, citing CAIS · 2026 AI Index ReportThe information sector — telecom, media, tech services — led US corporate AI hiring in 2025. Lightcast data cited in Stanford's 2026 AI Index found AI skills required in 13.2% of the sector's job postings, up from 7.8% in 2024Lightcast, via Stanford AI Index · 2025 data, pub. 2026Applied Materials sells the deposition, etch and inspection machines chipmakers must install before a single AI chip exists, so its results run a step ahead of the chip cycle. Its third fiscal quarter, ended 26 July 2026, was the biggest in the company's history: Applied Materials booked a record $9.12bn of revenue, up 25% year over yearApplied Materials 8-K (SEC) · Quarter ended 26 July 2026AMD's Data Center segment, the one selling Instinct GPUs and EPYC server chips, took $16.6 billion of the company's record $34.6 billion of revenue in 2025 — 48% of all AMD revenueAMD FY2025 results · year ended 27 Dec 2025Amkor Technology, the outsourced chip packaging and test house scaling up advanced packaging for AI processors, reported record second-quarter 2026 net sales of $1.90 billion, up 26% year-over-year $1.90 billion in record Q2 net sales, +26% YoYAmkor Technology · Q2 2026Amazon's cash capital expenditure across the whole of 2025 came to $131.81 billion in one yearAmazon 10-K, SEC EDGAR · FY2025Amazon's most recently reported quarterly cash capital expenditure reached $54.21 billionAmazon, SEC EDGAR (XBRL filings) · Q2 2026Amazon's cash capital expenditure in the first quarter of 2026 was $44.2 billion in a single quarterAmazon 10-Q, SEC EDGAR · Q1 2026Arista Networks, whose switches carry AI training and inference traffic inside hyperscaler data centres, reported total revenue for the quarter to 30 June 2026 of $3.036 billion, its first quarter above $3 billion and up 37.7% year-over-yearArista Networks · Q2 2026Applied Digital had signed long-term hyperscaler leases for 1,410 megawatts of AI data-centre capacity across five campuses as of May 31, 2026, worth approximately $36 billion in contracted revenue over the initial 15-year lease terms 1.41 gigawatts of AI data-centre capacity leased for $36 billionApplied Digital Corp. · as of May 31, 2026Arm's total royalty revenue grew 11% year-over-year to $671 million in its fiscal fourth quarter of 2026, and the company said its data centre royalty revenue within that more than doubled year-over-year on rising Arm-based server chip deployments by hyperscalers Arm data-centre royalties more than doubled YoYArm Holdings · Q4 FY2026Every advanced AI chip needs packaging and testing before it ships, and ASE Technology, the world's largest outsourced semiconductor packaging and test house, is racing to add capacity for it. Equipment spending in its packaging and test segment jumped in the quarter to 30 June 2026, from $942 million a year earlier to $1.646 billion in ASE's advanced-packaging equipment spending, up 74.7% year-over-yearASE Technology Holding · Q2 2026Every advanced AI chip is printed on an ASML machine, and the company says AI demand has its customers accelerating their expansion plans. On 15 July 2026 ASML said it would lift its 2026 low-NA EUV capacity of around 65 systems, for 2027, by 30% more EUV machines a yearASML Q2 2026 results · 15 July 2026Astera Labs makes the connectivity chips that link AI accelerators inside data-centre racks, and demand pushed its revenue for the quarter ended 30 June 2026 up 104% year-on-year to $392.4 millionAstera Labs 10-Q (SEC) · Q2 2026Broadcom reported $20 billion of AI revenue in its 2025 financial year, out of $63.89 billion in total — 31.3% of everything it soldBroadcom FY2025 Form 10-K · FY2025Broadcom's entire 2025 financial year produced $20 billion of AI revenue. In the single quarter to 3 May 2026, its AI semiconductor revenue — custom accelerators and AI networking — reached $10.8 billion in one quarter, up 143% year on yearBroadcom · Q2 FY2026 (to 3 May 2026)Baidu's AI Cloud Infra revenue grew 50% year-over-year in the quarter to 30 June 2026, powered by GPU Cloud demand up 283%, and the company poured capital into keeping pace. Group-wide capital expenditure nearly tripled year-over-year to RMB11.39 billion ($1.68 billion), Baidu's Q2 2026 capital expenditure, up 199.7% year-over-yearBaidu · Q2 2026BLS's 2024-34 employment projections show AI adoption fueling outsized growth in the occupations that build and secure it: data scientist jobs are projected to grow 33.5%, more than ten times the 3.1% projected growth rate for all US occupations combined 33.5% projected growth in US data scientist jobs through 2034U.S. Bureau of Labor Statistics · 2024-34 projectionsCadence Design Systems, whose software chipmakers use to design AI accelerators, reported $1.584 billion of revenue in its second quarter of 2026, up 24.2% from $1.275 billion a year earlier, with quarter-end backlog reaching a record $8.1 billion as CEO Anirudh Devgan cited accelerating demand for AI-driven design tools Cadence's Q2 2026 revenue: $1.584 billion, up 24.2% year-over-yearCadence Design Systems · Q2 2026Long-dated nuclear supply is being locked up now for demand that only arrives at the end of the decade. In the quarter to 30 June 2026 Constellation Energy signed power purchase agreements running 15 to 20 years with investment-grade customers, beginning between 2029 and 2032: Constellation signed another 920MW of long-term nuclear supply, none of it starting before 2029Constellation Energy 8-K (SEC) · Q2 202644% of US firms using AI report it augmenting workers' existing tasks rather than replacing them, versus just 5% reporting AI substituting tasks outright, according to Census Bureau research on business AI adoption 44%U.S. Census Bureau · 2026 BTOS AI supplementThe Census Bureau's Business Trends and Outlook Survey found AI adoption climbing fastest among big employers: as of May 2026, 37% of firms with 250+ employees used AI vs 32% for 100-249 employees and under 20% for the smallest firmsUS Census Bureau · May 202655% of U.S. workers say they have used AI on the job for at least one of 11 listed tasks, and among those who used it in the past week, 31% said it saved them one to two hours of work, per the Census Bureau's March 2026 Household Trends and Outlook Pulse Survey 55% of U.S. workers have used AI on the job, Census Bureau survey findsU.S. Census Bureau · March 2026 HTOPSNearly 750 executives in the Duke/Federal Reserve CFO Survey expect AI to barely touch total headcount — aggregate employment down less than 0.4% in 2026. What shifts is the mix, over the next three years: routine clerical work down more than 2 percentage points of employment shareNBER w34984, The CFO Survey · 2026Q1 survey waveChatGPT was estimated to have reached 100 million monthly users roughly two months after launchUBS, reported by The Register · February 2023The total computing power of the stock of AI chips is growing about 3.4× per yearEpoch AI · 2026A UC Berkeley, Sao Paulo and Stanford study of 141 countries found that national Claude usage rises 12x for every 0.10-point increase in a country's AI exposure score, based on Anthropic's Economic Index data for February 2026 Claude usage rises 12x per +0.10 rise in national AI exposureUC Berkeley/Stanford study · arXiv, Aug 2026Epoch AI asked LLM judges how long an experienced engineer would need, without AI, to reproduce each merged pull request from the 41 collaborators on OpenAI's public Codex repository. In the second quarter of 2026, work beyond what anyone could finish in a single unassisted day accounted for 8% of contributor-days, up from 2% a year earlierEpoch AI · Q2 2026Coherent's Datacenter & Communications segment, the optical transceivers that link GPUs inside AI data centers, brought in $1.615 billion in the quarter to 30 June 2026, up 58.6% from $1.018 billion a year earlier as demand for 800G and 1.6T transceivers accelerated Coherent's AI-datacenter segment revenue: $1.615 billion, up 58.6% year-over-yearCoherent Corp · Q4 FY2026 (June 2026)Coherent's Datacenter & Communications segment, which supplies optical transceivers for AI data centers, generated $1.62 billion in revenue in the quarter ended June 30, 2026, up 58.6% from a year earlier $1.62 billionCoherent Corp. · Q4 FY2026CoreWeave rents AI computing power by the hour, and demand is not the problem: first-quarter 2026 revenue rose 112% year on year to $2.08 billion. The company still reported a $740 million net loss for the quarterCoreWeave · Q1 2026Core Scientific has contracted about 1.1 gigawatts of leased data-center power capacity to AI and HPC customers, including CoreWeave and AMD, representing more than $24 billion of potential revenue, the company said in its second-quarter 2026 results Core Scientific's leased AI data-center capacity: 1.1 gigawatts, over $24 billion in contractsCore Scientific · Q2 2026Credo Technology makes the active electrical cables and optical DSPs that connect AI accelerators inside hyperscaler racks, and demand for that interconnect gear pushed its revenue for the quarter ended 26 April 2026 up 157.0% year-over-year to $437.0 million in quarterly revenueCredo Technology · Q4 FY2026Salesforce brands its whole product line around AI agents and reports separately what customers actually pay for them. In the quarter to 30 April 2026, on total revenue of $11.13 billion, that came to $1.2 billion of Agentforce annual recurring revenue, up 205% year on yearSalesforce Q1 FY2027 results · Quarter to 30 April 2026Salesforce brands its whole product line around AI agents and reports separately what customers actually pay for them. In the quarter to 31 July 2026, Agentforce annual recurring revenue exceeded $1.5 billion in Agentforce ARR, up over 240% year-over-yearSalesforce Q2 FY2027 results · Q2 FY2027Salesforce's quarter to 31 July 2026 brought total revenue of $11.3 billion, up 11% year-over-year, with subscription and support revenue of $10.8 billion providing the bulk of the growth $11.3 billion in Q2 FY2027 revenueSalesforce · Q2 FY2027CoreWeave rents out AI computing capacity, and building that capacity is capital-intensive. Its first-quarter 2026 cash flow statement shows capital expenditure on property and equipment, including capitalised internal-use software, jumped to $7.7 billion in Q1 2026 capital spending, a 5.5x jump from $1.4 billion a year earlierCoreWeave · Q1 2026Dell's quarter to 28 May 2026 was the one where AI servers took over the business: total revenue hit a record $43.8 billion, up 88% year on year, and of that, $16.1 billion came from AI servers aloneDell Technologies 8-K · Q1 FY2027, ended 28 May 2026Data-center landlords are racing to keep pace with AI and cloud demand. Digital Realty's second-quarter 2026 financial supplement to investors disclosed a global construction pipeline of Digital Realty's 1,402 megawatts of IT capacity under construction, 54% already pre-leasedDigital Realty Q2 2026 supplement · Q2 2026As cloud providers shifted from AI training to inference, conventional DRAM contract prices jumped 93% to 98% in a single quarter — and on TrendForce's count of the first quarter of 2026, global DRAM revenue hit $97 billion, up 81%TrendForce · Q1 2026Epoch AI's Capabilities Index scores frontier models on one aggregate scale, and fits a break in the trend at April 2024, when reasoning models and reinforcement learning took over frontier training. Before the break the frontier gained 8.3 index points a year; after it, 15.5 points a year, 1.85 times as fastEpoch AI · December 2025NBER asked 69 economists what the US economy would look like in 2050 if AI capabilities advance rapidly. Their median answer was 3.5% annual GDP growth, alongside a labour force participation rate falling from today's 62% to 55.0% by 2050, the level before women entered the workforce en masseNBER working paper 35046 · April 2026AI data centres are pushing US electricity generation to record highs. Total output from the country's power sector in the first half of 2026 rose 1.8% versus the first half of 2025, adding 37 billion kilowatt-hoursEIA, Short-Term Energy Outlook · Aug 2026 STEOEIA modelled what would happen if data-centre power demand grew faster than forecast. Its February 2026 baseline had US coal generation falling 9.3% between 2025 and 2027; in the higher-demand scenario, coal plants run on their existing spare capacity and the decline shrinks to a 5.0% fall in US coal generation instead of 9.3%EIA Short-Term Energy Outlook analysis · March 2026EIA's Annual Energy Outlook 2026 estimates that data-centre servers alone, before any cooling or other building load, already accounted for 7% of all US commercial-sector electricity in 2025US Energy Information Administration · AEO2026, May 2026On 3 August 2026 the governor of Texas paused new data-centre development while projects already under review are assessed. In the same month's Short-Term Energy Outlook the EIA cut its forecast for Texas electricity load growth in 2027 to 6% growth in Texas load, down from 14% forecast a month earlierEIA Short-Term Energy Outlook · August 2026Texas paused new data-centre development on 3 August 2026 while it reviews the projects already in the queue, and the EIA cut its forecast for the state in response: Texas electricity load growing 6% in 2027, not the 14% forecast a month earlierEIA Short-Term Energy Outlook · August 2026AI inference is consuming NAND flash faster than suppliers can allocate it. Enterprise SSD contract prices rose roughly 80% in the first quarter of 2026, and on TrendForce's count the top five brands' combined revenue hit a record $18.46 billion — enterprise SSD revenue rose 86.1% in a single quarterTrendForce · Q1 2026The 84 AI data centers tracked by Epoch AI have a combined real-world power footprint of 16.9 gigawatts as of August 2026, more than New York City's 11 GW peak demand 16.9 GWEpoch AI · August 2026Epoch AI tested three commercial AI detectors — Pangram, GPTZero, Originality.ai — pinned to their June 2026 versions. Against plain machine-generated text they catch nearly everything, but shown five samples of a specific author's writing and told to imitate the style, AI detectors missed 13% of style-imitated AI passages, versus under 1% of plain AI textEpoch AI · Jul 2026Epoch AI rebuilt the cost structure of three AI companies — Anthropic, MiniMax and Z.ai — from IPO filings and reporting. Staff, despite the industry's headline pay packages, take under a quarter of spending. Research and inference compute together take 54% to 62% of an AI company's costsEpoch AI · 2025 accountsEpoch AI fits a trend through the models whose context windows ranked among the ten longest on their release date, and finds that since mid-2023 the frontier context window has been getting about 30× longer every yearEpoch AI · since mid-2023Frontier models have started finding software flaws at scale. Counting the high- and critical-severity CVEs disclosed by 21 major vendors — Microsoft, Google, Apple and Cisco among them — Epoch AI puts July 2026 at about 2,500 serious CVEs in July 2026, five times the monthly record set before April 2026Epoch AI · July 2026Epoch AI tracked 333 compute estimates for notable AI models released between 2010 and May 2024, and found that training compute for the frontier — the top 10 models by compute at release — has been growing each year to reach 5.3x the prior year's levelEpoch AI · 2010–May 2024 dataEpoch AI audited its FrontierMath benchmark after OpenAI flagged suspiciously high scores, and its 12 June 2026 v2 release corrected errors in 42% of the original problems, leaving a verified set of 338 problems across four tiers 42% of FrontierMath's original problems contained errorsEpoch AI · June 2026 (v2 release)AI capability is getting cheaper to reach, not just bigger. Epoch AI tracked what it cost in tokens to hit the same score — about 27% accuracy on its FrontierMath benchmark — first with o4-mini in April 2025, then with GPT-5.2 in December 2025, and found AI capability got roughly 3x cheaper to reach in eight months, part of a ~5-10x per year cost-decline trendEpoch AI · Feb 2026Epoch AI and Ipsos asked 469 US workers who had used AI for work in the previous seven days whose subscription they were on. In computer, engineering and science jobs 67% use a tool their employer provides — everywhere else the free tier wins: 46% of US workers using AI at work are on a free plan, against 37% given a tool by their employerEpoch AI / Ipsos · Fieldwork 10-19 July 2026Epoch AI and Ipsos asked 1,106 employed US adults in July 2026 whether AI had taken over work they used to hand to another person. Reporting that AI now handles at least one task they previously passed to a coworker or contractor: one in five US workersEpoch AI / Ipsos survey · July 2026Eaton's Electrical Global segment, which sells the power distribution and switching gear AI data centres depend on, ended the second quarter of 2026 with total backlog up 103% year-over-year Eaton Electrical Global backlog up 103% year-over-yearEaton · Q2 2026Asked what three years of AI had actually changed inside their own firms, nearly 6,000 senior executives across the US, UK, Germany and Australia mostly reported nothing: nine in ten saw no impact on employment or productivityNBER working paper 34836 · February 2026Of the notable AI models released in 2025, Stanford's AI Index counts over 90% as coming from industryStanford HAI, AI Index 2026 · 2025Across 42 notable language models, Epoch AI estimates frontier training costs have risen from roughly $2 million for GPT-3 to up to nearly $390 million per runEpoch AI · 2024Epoch AI's Open Problems benchmark collects research mathematics problems that have resisted serious attempts by professional mathematicians. Expanded to 50 problems on 31 July 2026, the count two weeks later — with the newest AI solution still marked provisional — stood at 4 solved by AI, 1 by a human, 45 still openEpoch AI FrontierMath · 17 August 2026Alphabet told investors in its second-quarter 2026 results that Gemini models now process 22 billion API tokens a minute, and that the Gemini app has reached 950 million monthly active usersAlphabet 8-K · Q2 2026In a randomised field experiment across 66 firms and 7,137 knowledge workers, a generative AI tool built into everyday work apps saved the people who used it two hours a week on email, with no detected change in their tasksNBER working paper 33795 · revised Nov 2025Mentions of generative AI skills in US AI job postings grew 111% from 2024 to 2025, even as their share of total AI postings slipped, as employer demand pivots toward agentic-AI skills, per Lightcast data cited in Stanford's AI Index 111%Stanford HAI · 2025Across 5,179 customer-support agents, a generative AI assistant raised issues resolved per hour by 14% on average — and by 34% for the least experienced staffBrynjolfsson, Li & Raymond, NBER · April 2023GE Vernova builds the gas turbines and grid equipment that electricity supply runs on, and AI data centres are now one of its largest sources of new demand. Its gas equipment under contract rose from 100 GW to 116 GW during the second quarter of 2026, and data-centre orders through the first half of the year reached over $5 billion of data-centre orders, more than double the whole of 2025GE Vernova Q2 2026 results · First half of 2026Alphabet's most recently reported quarterly cash spending on property and equipment reached $44.92 billionAlphabet, SEC EDGAR (XBRL filings) · Q2 2026Alphabet's cash capital expenditure in the first quarter of 2026 reached $35.67 billion in a single quarterAlphabet 10-Q, SEC EDGAR · Q1 2026Alphabet's quarterly cash capital expenditure has roughly tripled in two years, from $12.01 billion in the first quarter of 2024 to $35.67 billion in the first quarter of 2026Alphabet 10-Q, SEC EDGAR · Q1 2024 - Q1 2026Epoch AI estimates the single largest owner of AI computing power worldwide is not a chipmaker or a cloud reseller but Google, which held, measured in H100-equivalents as of the fourth quarter of 2025, about one quarter of all global AI compute, mostly from its own custom TPU chips rather than Nvidia GPUsEpoch AI · Q4 2025Epoch AI's Capabilities Index scores language models on one aggregate scale, and nothing since has held first place as long as GPT-4 did after its March 2023 release — roughly a year. The second-longest lead, by OpenAI's o1, lasted a little over three months, less than a third of GPT-4's run at the topEpoch AI Capabilities Index · July 2026Traditional computing powers like Germany, Japan and France now play marginal roles in AI infrastructure. Epoch AI's tracking of global GPU cluster performance in May 2025 put China second at 15%, behind a United States hosting about three-quarters of the world's GPU cluster performanceEpoch AI · May 2025High-bandwidth memory is taking over the memory industry's capacity: TrendForce expects HBM to account, among the top three DRAM suppliers by the end of 2026, for 22% of DRAM wafer input, up from 18% a year earlierTrendForce · June 2026 projectionAmazon, Google, Meta, Microsoft and Oracle together held, measured in H100-equivalents as of Q4 2025, an estimated 71% of the world's AI computeEpoch AI · Q4 2025Epoch AI models the world's AI inference capacity, at fixed model size and context length, as growing 3.4× a year — but it puts token demand growth at around 10× a year, three times faster than supplyEpoch AI · May 2026Intel's Data Center and AI segment, the unit selling Xeon server chips and Gaudi/Crescent Island AI accelerators, reported revenue for the quarter ended 27 June 2026 up 59% year-over-year to $6.3 billion in quarterly segment revenueIntel Corporation · Q2 2026IREN, the bitcoin miner turned AI cloud operator, said its AI Cloud Services revenue grew roughly 8x year-over-year to $128.8 million in the fiscal year ended June 30, 2026, and the company now has $4 billion of ARR contracted for 2026 GPU capacity, with $1 billion already operatingIREN FY26 results release · FY2026IREN, a bitcoin miner turned AI cloud operator, told investors in its quarter ended 31 March 2026 that it now holds executed grid connection agreements across seven data-centre sites in Texas, Oklahoma and British Columbia for a combined 4,510 megawatts of AI data-centre power capacityIREN Ltd · Q3 FY2026 10-QEpoch AI tracks the largest AI data centre in the world by computing capacity, and the record has changed hands repeatedly since August 2024 — Colossus, Anthropic and Amazon's New Carlisle, Microsoft's Fairwater Atlanta, Meta's Prometheus. Each time, the bar has risen at the same pace: a doubling every seven monthsEpoch AI · June 2026The largest AI data centre in Epoch AI's tracker, Colossus 2 in Memphis, Tennessee, holds a computing capacity of 1,112,000 H100-equivalent AI chipsEpoch AI · AI data centers tracker, August 2026Nvidia's accelerators have pushed thermal design power from 700W for the H100 to over 1,000W for the B200 and B300, and air can no longer carry the heat away. TrendForce expects liquid cooling in server racks to reach 47% of server racks by 2026TrendForce · November 2025 forecast for 2026Lam Research, whose machines etch and deposit the layers inside AI chips, reported fiscal 2026 revenue of $23.23 billion — up 26% from $18.44 billion in fiscal 2025 — as CEO Tim Archer credited AI-driven demand for record margins and EPS $23.23 billion in fiscal 2026 revenueLam Research · FY2026 (ended June 28, 2026)A McKinsey survey of employers cited in Stanford's 2026 AI Index found anticipated AI-driven workforce cuts concentrated in service operations, supply chain and software engineering, while almost half of organizations expect little or no change overall. Still, one-third of employers expect AI to shrink their workforce in the year aheadMcKinsey survey (Stanford AI Index 2026) · 2025 surveyMeta's purchases of property and equipment nearly doubled in 2025, from $37.3bn to $69.7bn — 34.7% of its revenueMeta FY2025 Form 10-K · FY2025Meta's cash capital expenditure went the other way in the first quarter of 2026 — from $21.38 billion the quarter before, a fall to $19.0 billionMeta 10-Q, SEC EDGAR · Q1 2026Meta's most recently reported quarterly purchases of property and equipment reached $30.12 billionMeta, SEC EDGAR (XBRL filings) · Q2 2026In a randomised trial, 16 experienced open-source developers believed AI tools had made them 20% faster. Measured against issues they finished without AI, they were 19% slowerMETR · July 2025Micron breaks out the memory it sells into AI data centres as its own business unit. In the quarter to 28 May 2026 that unit took in $11.52 billion, against $1.53 billion in the same quarter a year before — Micron's core data centre revenue grew 7.5× in a yearMicron 8-K, SEC · quarter to 28 May 2026Monolithic Power Systems makes the chips that step grid power down to the voltages AI GPUs need. Its Enterprise Data segment, which sells into AI servers, grew far faster than the rest of the company in the quarter to 30 June 2026, when total revenue hit a record $980.6 million and Enterprise Data segment revenue reached $380.6 million in AI server chip revenue, up 164.3% year-over-yearMonolithic Power Systems · Q2 2026Marvell designs the custom accelerators, optics and switching silicon that wire AI clouds together, and that business has crowded out everything else it sells. In the quarter to 2 May 2026, data centre revenue of $1.83 billion out of $2.42 billion total came to 76% of Marvell's revenueMarvell Technology · Q1 FY2027 (to 2 May 2026)Marvell's record quarter to 2 May 2026 looks very different depending on which profit line you read. Non-GAAP net income was $718.0 million. Under GAAP — after $207.6 million of stock compensation, $225.2 million of amortisation on acquired intangibles and a $250.7 million contingent-consideration revaluation — net income was $34.5 million, under 5% of the adjusted figureMarvell Technology · Q1 FY2027 (to 2 May 2026)Marvell's revenue hit a record $2.74 billion in the quarter to 1 August 2026, up 37% year-on-year, led by an AI data-centre business that grew even faster — accelerating to 46% year-on-year, to $2.17 billion in Data Center revenueMarvell Technology, Inc. · quarter ended 1 August 2026Marvell's data infrastructure chip business set a fiscal second-quarter revenue record in the three months to 1 August 2026, with net revenue up 37% year-over-year to $2.739 billion, as Data Center segment revenue growth accelerated to 46% $2.739 billion in record Q2 revenueMarvell Technology · Q2 FY2027Microsoft told investors in April 2026 that its AI business had reached an annual run rate of $37 billionMicrosoft FY2026 Q3 earnings call · April 2026Microsoft's cash capital expenditure rose by more than half between the third and fourth quarters of 2025, reaching $29.88 billion in a single quarterMicrosoft 10-Q, via SEC XBRL · Q4 2025Microsoft's most recently reported quarterly cash spending on property and equipment reached $35.80 billionMicrosoft, SEC EDGAR (XBRL filings) · Q2 2026Microsoft's cash spending on property and equipment reached, in the first quarter of 2026 alone, $30.88 billionMicrosoft 10-Q, via SEC XBRL · Q1 2026A randomised online experiment gave 1,174 adults aged 25 to 45 an incentivised business problem-solving task, some with a generative-AI assistant and some without. Unaided, the higher-education group beat the lower-education group by 0.548 standard deviations. With AI the gap fell to 0.139 — about three-quarters of the education gap closedNBER Working Paper 34851 · May 2026 revisionMenaka Hampole and co-authors linked resume data to firms' AI-application adoption to isolate AI's true effect on jobs, not just correlation. In their preferred instrumental-variables specification, a one-standard-deviation rise in an occupation's AI exposure cuts its within-firm employment share by 14%NBER · Working Paper 33509, 2025Nebius rents out AI cloud capacity and is scaling from a small base. Revenue in the second quarter of 2026 reached $582.3 million against $105.1 million in the same quarter of 2025 — a 454% year-on-year rise in revenueNebius Group · Q2 2026In July 2025, Nvidia became the first US company in history to reach a $4 trillion market capitalisationAl Jazeera · July 2025Nvidia's Data Center segment brought in $193.7bn of the $215.9bn it made in its 2026 financial year — 89.7% of the company's revenueNvidia FY2026 Form 10-K · FY2026, ended 25 January 2026Nvidia's Data Center segment generated $89.0 billion in revenue in its fiscal second quarter ended July 26, 2026, up 117% from a year earlier and now 92% of total company revenue $89.0 billionNvidia · Q2 FY2027Nvidia reported revenue of $96.2 billion for its fiscal second quarter ended July 26, 2026, up 106% from a year earlier and 18% from the prior quarter $96.2 billionNvidia · Q2 FY2027Nvidia spent $18.5 billion on research and development in its 2026 financial year — three times its $6.04 billion of capital expenditureNvidia FY2026 Form 10-K · FY2026Onto Innovation's inspection systems qualify the 2.5D packaging and high-bandwidth memory built into every AI accelerator, and that demand pushed its Specialty Devices and Advanced Packaging segment to an all-time high. Total company revenue in the quarter ended 30 June 2026 reached $343.1 million in quarterly revenue, up 35.3% year-over-yearOnto Innovation · Q2 2026Across January to May 2026, Epoch AI's capabilities index puts the best open-weight models behind the frontier closed models by an average of four monthsEpoch AI · Jan-May 2026Oracle's remaining performance obligations — cloud work it has signed but not yet delivered — stood, when its financial year closed on 31 May 2026, at $638 billion of contracted revenue, up 363%Oracle · Q4 FY2026, ended 31 May 2026Oracle's cash capital expenditures jumped from $21.2 billion in fiscal 2025 to $55.7 billion in fiscal 2026, an increase the company attributes directly to the expansion of its data centres $55.7 billion in Oracle FY2026 capex, up from $21.2 billionOracle · FY2026Stanford's AI Index reports that organisational AI adoption kept climbing in 2025, reaching 88% of surveyed organisationsStanford HAI AI Index · 2025 figureAI chip performance per dollar has been improving by approximately 1.37× per yearEpoch AI · 2026PJM's Dominion zone in Virginia serves the largest concentration of data centres in the world, and their power draw shows up hardest on the coldest nights. Winter peak electricity demand there reached 25,413 megawatts in the 2025–26 winter — 45% higher than the 2019–20 winterEIA · May 2026Palantir positions its whole product line as an AI platform but reports no AI revenue at all — its 2025 annual report splits revenue 54% government, 46% commercialPalantir FY2025 Form 10-K · FY2025Scale is not the only driver of AI progress. Epoch AI measures how much training compute it takes to reach a fixed level of language-model performance, and finds that requirement falling fast: pre-training compute efficiency doubles every 7.6 monthsEpoch AI · Feb 2026Quanta Services builds the transmission lines, substations and high-voltage switchgear that connect new AI data centres to the grid, and it told investors that rising demand from data centres, electrification and grid modernisation is driving orders. Its total backlog reached a record $53.4 billion, Quanta Services' record grid-construction backlog as of 30 June 2026Quanta Services · Q2 2026Samsung is racing to keep pace with HBM4 demand for Nvidia's next GPU generation, and TrendForce reports the company plans to lift its HBM wafer output from about 170,000 wafers a month to roughly 250,000 wafers a month by the end of 2026, a jump of about 47%TrendForce · Dec 2025US cloud providers signed multi-year supply agreements that stop memory makers raising their prices, so TrendForce expects the third-quarter 2026 increase to fall on everyone else: server DRAM contract prices up another 13–18% in one quarterTrendForce · 9 July 2026ServiceNow sells AI as an add-on to the workflow software companies already run, and reports separately what customers pay for it. Agentic deployments rose ninefold in nine months, and in the second quarter of 2026 ServiceNow AI reached $1 billion in annual contract valueServiceNow Q2 2026 results · Q2 2026Soaring AI-server memory prices lifted SK hynix to 79.32 trillion won of revenue and 60.54 trillion won of operating profit in the second quarter of 2026 — an operating margin of 76%SK hynix · Q2 2026SpaceX now reports an AI segment alongside Space and Connectivity. In the quarter to 30 June 2026, advertising on X brought in $367 million — less than a year earlier — while the segment's AI solutions and infrastructure line, covering Grok API access, subscriptions, data licensing and cloud, reached $2.19 billion, seven times its level a year earlierSpaceX 10-Q · Q2 2026SpaceX listed on Nasdaq in June 2026, and its first quarterly report as a public company discloses a metric no rocket company used to publish: installed GPUs multiplied by their all-in power draw, before any cooling or facility overhead. At 30 June 2026 that came to 1.4 gigawatts of installed GPU power draw, 3.5× a year earlierSpaceX 10-Q · 30 June 2026Super Micro Computer assembles and ships AI servers rather than designing the chips inside them. For the financial year ended 30 June 2026 it reported net sales of $39.1 billion, up from $22.0 billion a year earlierSuper Micro Computer 8-K · FY ended 30 June 2026AI data centres need firm, round-the-clock power, and nuclear plants are being locked up years in advance to supply it. In June 2025, independent power producer Talen Energy expanded its agreement with Amazon Web Services to deliver up to 1,920 megawatts of carbon-free nuclear power from the Susquehanna plant through 2042Talen Energy · June 2025Teradyne, whose systems test semiconductor chips before they ship, reported quarterly revenue of $1.33 billion for the quarter ended June 28, 2026, up 104% from $652 million a year earlier, with $1.12 billion of that revenue in its Semiconductor Test segment Teradyne's revenue up 104% year-over-year to $1.33 billionTeradyne Inc. · Q2 2026Tesla's own quarterly filing attributes its capital spending mainly to global AI and operational infrastructure. Across the first half of 2026 that spending more than doubled against the same six months of 2025, from $3.89 billion to $8.28 billion of capital expenditureTesla Form 10-Q · Q2 2026Tesla reports its AI training compute the way it reports factory capacity — in installed megawatts. At the end of the second quarter of 2026 both Texas clusters, Cortex 1 and Cortex 2, were in production, and the compute installed there had more than doubled in six months, to over 205 MW of AI training capacityTesla Q2 2026 Update · Q2 2026Tesla's driver-assistance software is one of the few AI products consumers pay for directly, and the company discloses the subscriber count every quarter. At the end of the second quarter of 2026 it reported 1.48 million active FSD subscriptions, up 56% in a yearTesla Q2 2026 Update · Q2 2026Compute used to train frontier language models has grown roughly 5× per year since 2020Epoch AI · 2026Orders for Nvidia rack-scale platforms and in-house cloud ASICs came in stronger than expected, so in August 2026 TrendForce revised its forecast for this year's global AI server shipments up from 28% growth to nearly 31% growth year-on-yearTrendForce · August 2026TrendForce forecasts that global foundry industry revenue will grow 24.8% year-over-year to approximately $218.8 billion in 2026, with TSMC posting the largest increase of around 32% Global foundry revenue forecast to grow 24.8% to $218.8 billionTrendForce · published March 2026AI servers have repriced memory faster than any forecast could keep up with. In a single quarter TrendForce raised its estimate for 2026 global memory market revenue from $551.6 billion to $889.3 billion — 61% above its own forecast from three months earlierTrendForce · May 2026 forecastTSMC's high performance computing platform, the category that carries AI accelerators, accounted for 58% of its 2025 revenueTSMC 4Q25 management report · 2025In the first quarter of 2026, TSMC's share of revenue across the world's ten largest foundries reached 72%TrendForce · Q1 2026Stanford's AI Index found the United States trailing most of the world's advanced economies on population-level AI adoption. Citing Microsoft AI Economy Institute telemetry for the second half of 2025, the index ranked the US 24th out of the top 30 economies worldwide, with an AI usage rate of 28.3% of Americans, still in the bottom half of the global rankingStanford AI Index (Microsoft data) · H2 2025Across US businesses surveyed between December 2025 and May 2026, AI use ran between 17% and 20% — and even among firms with 250 or more employees, only 37% reported using AIUS Census Bureau, BTOS · December 2025 – May 2026Counting the notable AI models released in 2025, Stanford's AI Index puts the United States on 59, against China's 35Stanford HAI, AI Index 2026 · 2025US data centres used 176 terawatt-hours of electricity in 2023 — on Berkeley Lab's measurement, 4.4% of the country's totalLawrence Berkeley National Laboratory · 2023Berkeley Lab projects that by 2028 US data centres will draw between 6.7% and 12% of all US electricityLawrence Berkeley National Laboratory · Projection to 2028Generative AI reached 53% adoption in three years, faster than the personal computer or the internet — but unevenly: Singapore sits at 61% and the UAE at 64%, while Stanford's AI Index finds the US is 24th for generative AI adoption, at 28.3%Stanford AI Index · 2026 reportEIA forecasts US electricity use rising 1% in 2026 and 3% in 2027, driven largely by large computing facilities including data centres — which would be the first four-year run of demand growth since 2007EIA Short-Term Energy Outlook · Jan 2026Vertiv, which builds data-centre power and cooling, ended 2025 with an order backlog of $15.0 billion — up 109% in a yearVertiv Q4 2025 results release · Q4 2025Vistra signed 20-year power purchase agreements with Meta Platforms in January 2026 to supply a total of 2,609 MW of carbon-free power and capacity from its Perry, Davis-Besse and Beaver Valley nuclear plants in PJM, with full delivery of operating capacity by 2027 and uprate capacity by 2034 2,609 MW of nuclear power under 20-year Vistra-Meta PPAsVistra Corp. · January 2026Stanford's AI Index, citing Brynjolfsson et al., finds AI's labour-market effect landing on hiring pipelines rather than on broad job losses. By September 2025, employment for US software developers aged 22 to 25 had fallen close to 20% from its 2022 peakStanford AI Index · 2026 report
LIVE FACTS

THE INTELLIGENCE CLUB

Everything you need to navigate the AI decade

Inspired by the best of investing research and AI intelligence platforms — rebuilt around one obsession: facts you can act on.

Daily AI Facts

The signal, minus the noise. Every fact sourced and dated — receipts on the card, every time.

146 facts on file

Go to today's facts

Model & Chip Leaderboards

Who's really winning? Live rankings of the models, the chips, and the companies behind them.

Top: Claude Opus 5 (Adaptive Reasoning, Max Effort)

Go to the leaderboard

Investing Intelligence

The AI share of revenue — published only where the company itself discloses one. No estimates.

Go to AI exposure

AI Stock Screener

Every AI-exposed listing in one table, each price stamped with the minute it was captured.

72 listings tracked

Go to the screener

Weekly Video Briefings

The AllFactsAI channel — the week in AI, decoded in minutes, every week.

Go to the channel

The Members' Club

A community of investors comparing notes: early access, member calls and AMAs.

See the three tiers

Not sure where to start? Start free with The Daily Receipts →

Menu

THE AI VALUE CHAIN

Follow the money through the boom

Every layer of the AI build-out, from the power grid to the software that has to pay for it all. Pick a layer to see who owns it and what decides whether they win.

Accelerators

The engine

GPUs and custom silicon convert capital into compute. This layer captures the largest share of AI spending — and sets the price of intelligence.

What to watch
Data-center revenue, backlog, custom-silicon share shift, gross margin.

Price this layer →
NVDANvidiaAMDAMDAVGOBroadcomMRVLMarvell
Menu

THE AI STOCK SCREENER

MARKET CLOSED · 17:52 UTCPRICES 15-MIN DELAYED · EODHD₩ RESTATED AT 1,377.16/$ · 30 Aug

Every layer of the chain — priced

72 listed companies, each placed in the layer of the AI value chain where its AI revenue is actually recognised. Sort by any column; a figure we do not have prints as a dash, never as a zero.

S&P 500

 

7,711.76

-0.25%

NASDAQ

 

26,402.42

-0.52%

DOW

 

53,559.99

-0.02%

RUSSELL 2000

IWM ETF

$295.75

-1.35%

VIX

 

14.43

-0.55%

Compare stocks

#30-Day
1
Nvidia
Accelerators$219.24-3.83%+1.32%+11.54%+3.16%$5.28T$118.34M27.7$164.98 – $235.47
2
Apple
Hyperscalers$319.61+1.60%+3.35%-4.04%+2.54%$4.66T$20.84M36.7$201.41 – $339.79
3
Alphabet
Hyperscalers$346.07+1.59%+0.51%+3.88%-8.82%$4.23T*$11.45M17.4$188.59 – $402.38
4
Microsoft
Hyperscalers$517.07+2.38%+6.27%+14.05%+14.27%$3.84T$15.54M28.8$352.17 – $537.65
5
Amazon
Hyperscalers$265.06+3.43%+3.02%+13.13%-1.56%$2.86T$31.06M21.3$198.79 – $284.02
6
TSMC
Foundry & Memory$421.28-1.41%-0.34%+3.52%-0.04%$2.18T$5.07M—$225.26 – $477.57
7
Broadcom
Accelerators$367.27-1.15%+0.09%-4.91%-17.32%$1.75T$8.48M61.1$286.52 – $480.77
8
Meta Platforms
Hyperscalers$575.95+0.85%+5.11%+7.23%-8.53%$1.48T*$9.54M21.7$525.23 – $787.42
9
Tesla
Applications$346.71-2.28%-3.89%+12.92%-19.97%$1.37T$20.52M321.0$298.32 – $489.88
10
Micron Technology
Foundry & Memory$924.76-1.14%-3.51%+6.65%-3.91%$1.04T$15.74M20.9$104.71 – $1,213.37
11
AMD
Accelerators$471.55-1.07%-1.62%-4.08%-9.79%$769.79B$8.58M121.2$151.14 – $580.91
12
SpaceX
Model Labs$140.22-0.46%+3.31%+26.11%—$684.13B*$39.36M—$108.27 – $211.39
13
ASML Holding
Semi Equipment$1,698.19-2.12%-3.83%+2.71%+5.31%$654.51B$736,163—$685.12 – $1,986.87
14
Intel
Foundry & Memory$89.27-3.06%-0.67%-1.82%-21.98%$450.28B$58.43M—$19.31 – $140.94
15
Palantir Technologies
Applications$185.99+0.03%+3.53%+52.37%+19.00%$446.91B*$15.49M160.3$107.27 – $207.18
16
Oracle
Hyperscalers$150.61-0.88%+2.99%+18.26%-32.95%$433.83B$10.86M25.8$114.99 – $324.63
17
Lam Research
Semi Equipment$304.50-4.42%-3.85%+1.40%-5.05%$381.03B$2.77M52.9$94.32 – $433.33
18
Applied Materials
Semi Equipment$463.25-3.96%-6.23%-7.89%+2.69%$367.63B$3.22M40.0$155.39 – $722.23
19
Dell Technologies
Data Centers$460.12-2.57%+3.20%+12.70%+8.57%$298.62B*$2.18M36.6$110.54 – $494.51
20
Arm Holdings
Accelerators$242.05-5.16%-1.75%-1.03%-32.34%$257.55B$2.10M247.0$104.55 – $439.46
21
Arista Networks
Networking & Optics$195.91-2.58%+3.57%+14.24%+22.52%$247.09B$1.66M61.8$116.13 – $210.50
22
GE Vernova
Power & Cooling$918.82-3.67%-4.69%-7.13%-5.78%$244.71B$1.02M26.4$546.95 – $1,174.86
23
KLA
Semi Equipment$176.11-4.17%-4.59%-2.55%-8.55%$230.09B$3.68M48.0$83.94 – $301.37
24
Alibaba
Hyperscalers$118.25+1.67%-0.37%+2.22%-3.40%$219.71B$6.62M—$94.81 – $187.62
25
Sandisk
Foundry & Memory$1,484.99+0.00%-6.96%+16.02%-12.39%$217.43B$6.21M20.3$40.69 – $2,335.00

Market capDERIVED and P/EDERIVED are computed from SEC filings and the current price

Menu

THE MODEL LEADERBOARD

SNAPSHOT · 30 AUGUST 2026

Smartest per dollar — and who owns it

Intelligence against blended price, every lab tagged with the listed company that actually gives you exposure to it. The curve is the value frontier: models nothing cheaper beats.

Artificial Analysis Intelligence Index vs blended $ / 1M tokens

Up and to the left is better

5658606264$0.2$0.5$1$2$5$10$20Blended price per 1M tokens (log scale) — cheaper is leftIntelligence indexClaude Opus 5Grok 4.6GLM-5.3GLM-5.3-Flash
Closed / API onlyOpen weightsWeights not classifiedValue frontier

Selected

Claude Opus 5 (Adaptive Reasoning, Max Effort)

Anthropic · closed

Intelligence63.1$ / 1M tokens$10.00Tokens / sec53.776

How you get exposure

Private. Backed by Google and Amazon; runs on TPU and Trainium.

GOOGLAMZN
Artificial Analysis Intelligence Index and blended price per 1M tokens, 30 August 2026
ModelLabIntel.$ / 1MListed proxy
Claude Opus 5 (Adaptive Reasoning, Max Effort)frontierAnthropic63.1$10.00GOOGL · AMZN
Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)Anthropic62.1$20.00GOOGL · AMZN
GPT-5.6 Sol (max)OpenAI60.9$8.00MSFT
Grok 4.6 (high)frontierSpaceXAI60.9$3.00SPCX
Kimi K3 (max)Kimi59.7$6.00BABA
GLM-5.3 (max)frontierZ AI59.5$2.15—
Qwen3.8 MaxAlibaba58.1$3.00—
Qwen3.8 2.4T A95BAlibaba57.7$3.00—
GLM-5.3-FlashfrontierZ AI57.5$0.24—
Claude Opus 4.8 (Adaptive Reasoning, Max Effort)Anthropic57.3$10.00GOOGL · AMZN
Muse Spark 1.2 (xhigh)Meta56.8$2.00META
GPT-5.6 Terra (max)OpenAI56.6$4.50MSFT
GPT-5.5 (xhigh)OpenAI56.3$11.25MSFT
Gemini 3.7 Flash (high)Google56$1.50GOOGL

Most of the frontier is still private, and the listed exposure sits in the compute, the clouds and the backers underneath it. The exceptions are the labs owned outright by a listed company — Google, Meta, and xAI, inside SpaceX since their 2026 merger.

Scores and prices: Artificial Analysis, read 30 August 2026 — artificialanalysis.ai. One source only; other leaderboards score differently and are not comparable. Proxy notes describe exposure, not ownership, and are not investment advice.

Menu

THE MONEY LOOP

The money goes out. Then it comes back

Nvidia invests in the companies that buy its chips, and those chips come back as Nvidia's revenue. Pick an arrow for the amount, the date and the source behind it.

Nvidia

NVDA · Supplier

OpenAI

Customer

Anthropic

Customer

Microsoft

MSFT · Investor

Bond market

Lender

Pike County, Ohio campus

Landlord

Nvidia → OpenAI

$30bn (reported)

27 February 2026equityreported

Forbes, quoting OpenAI's own announcement ↗

Part of a $110bn round at a $730bn pre-money valuation.

COMPANY ANNOUNCEMENTS AND NAMED PRESS REPORTS · 1 AUGUST 2026

Menu

THE AI CAPEX TRACKER

FROM SEC FILINGS · 30 AUGUST 2026

The number that moves everything

Four balance sheets fund most of the boom. Their combined capital spending is every other company's revenue — which is why one guidance revision can reprice the entire chain.

Combined quarterly capex · big four

$165.05B+212% vs Q2 24

$0B$90B$180BQ2 24$52.85BQ3 24Q4 24Q1 25Q2 25Q3 25Q4 25Q1 26Q2 26$165.05B
MSFT$35.80Blatest
GOOGL$44.92Blatest
AMZN$54.21Blatest
META$30.12Blatest
Data table
QuarterMSFTGOOGLAMZNMETATotal
Q2 24$13.87B$13.19B$17.62B$8.17B$52.85B
Q3 24$14.92B$13.06B$22.62B$8.26B$58.86B
Q4 24$15.80B$14.28B$27.83B$14.43B$72.34B
Q1 25$16.75B$17.20B$25.02B$12.94B$71.91B
Q2 25$17.08B$22.45B$32.18B$16.54B$88.25B
Q3 25$19.39B$23.95B$35.09B$18.83B$97.26B
Q4 25$29.88B$27.85B$39.52B$21.38B$118.63B
Q1 26$30.88B$35.67B$44.20B$19.00B$129.75B
Q2 26$35.80B$44.92B$54.21B$30.12B$165.05B

SEC filings (10-Q, 10-K) via the SEC XBRL API, read 30 August 2026 — every quarter traced to the filing that reported it (EDGAR). These are cash purchases of property and equipment and exclude finance leases, which Microsoft in particular uses heavily — their headline capex figure is higher than the one here. They are also total company capex, not AI capex: AI infrastructure drives the growth, but fulfilment centres and ordinary facilities are in these numbers too. Alphabet has since reported Q2 2026 capex of $44.9bn. The chart stops at Q1 2026, the last quarter all four have filed.

Full quarterly table, SEC filings and CSV →

Menu

AI EXPOSURE

DISCLOSED FIGURES ONLY · 26 JULY 2026

How much AI is really in there?

Everyone claims to be an AI company. We publish the AI share of revenue only where the company itself reports one — and say so plainly where it does not. No estimates, no invented precision.

89.7%OF REVENUE

Nvidia

Accelerators

AI share of revenueCompany disclosed

Data Center segment revenue of $193.74bn on $215.94bn total. FY2026, ended 25 January 2026.

Data Center is not a pure AI line — it includes networking and non-AI accelerated computing — but it is the closest figure Nvidia reports.

Nvidia FY2026 Form 10-K ↗
From the filingsDerived from filings
Revenue
$215.94B
R&D / revenue
8.6%
Capex / revenue
2.8%

Fiscal year ended 2026-01-25. Capex intensity measures spending on AI, not selling into it — it runs high for hyperscalers and low for the companies supplying them.

Where we exercise judgementOur read
Chain criticalityVery high
Moat durabilityHigh

The purest expression of the trade, and the company with the least anywhere to hide if demand turns.

Only four of these companies tell you what share of their revenue is AI. That silence is itself the finding: the further you get from the silicon, the less anyone is willing to put a number on it.

AI revenue shares are the companies' own disclosures, linked on each card. Revenue, R&D and capex come from SEC XBRL filing data. Criticality and moat are our editorial judgement and are labelled as such — they are not folded into any number. Nothing here is a rating or investment advice; see the methodology.

The whole card, company by company

Every figure above, plus the cash flows, margins and valuation behind it. Nvidia's card is open to everyone; the rest are Insider.

Nvidia

AI COMPANY FACT CARD

Nvidia

NVDA · Accelerators · FY2026, ended 25 January 2026

Sample card
71.1%
Gross margin
60.4%
Operating margin
44.8%
Cash flow margin
89.7%
AI share of revenue

Revenue & free cash flow

Billions of dollars per fiscal year, with revenue growth under each. Levels, not growth rates: one year off a small base makes a growth chart unreadable.

27.0
3.8
60.9
27.0
130.5
60.9
215.9
96.7
FY2023
baseline
FY2024
+126%
FY2025
+114%
FY2026
+65%
RevenueFree cash flow

From the filings

From exposure

Nvidia FY2026 Form 10-K, via SEC XBRL

Revenue
$215.94B
Free cash flow
$96.68B
Cash & equivalents
$10.61B
Shares outstanding
24.3bn
Open the filing ↗

Revenue by market

The company's own market split, latest fiscal year against the one before. Bars are scaled to the largest line, not to total revenue.

Data Center$193.7B+68%
Gaming$16.0B+41%
Professional Visualization$3.2B+70%
Automotive$2.4B+39%
OEM and Other$0.6B+59%

Revenue by region

United States69.3%
Taiwan19.6%
China (incl. Hong Kong)9.1%
Other2.0%

Revenue by customer billing location. Nvidia discloses that 76% of the revenue billed to Taiwan is attributable to end customers based in the United States and Europe — billing location is not the end customer's location.

Nvidia FY2026 Form 10-K, revenue by region ↗

Quality

Our read

Our judgement, kept out of every measured figure.

Strengths — more is better

Chain criticalityVery high
Moat durabilityHigh
Disclosure qualityHigh

Risks — more is worse

Customer concentrationVery high

Two direct customers each accounted for 22% and 14% of Compute & Networking revenue in FY2026 — up from 12% and 11% a year earlier. Nvidia names neither. That segment is most of the company, so this is close to company-wide concentration, though it is not reported as such.

The read

Our read

The purest expression of the trade, and the company with the least anywhere to hide if demand turns.

  • Reports a segment figure at all — most of the chain does not.
  • Free cash flow grew with revenue rather than lagging it.
  • Capex intensity of 2.8% while selling to buyers spending ten times that.
  • Data Center is not a pure AI line; it is an upper bound.
  • Almost the whole business now rides on one end market.
  • Cash and equivalents are small next to the market's expectations.

What the price assumes

Our read

The growth today's quote already implies, on our other assumptions.

No live quote available, so this is left blank rather than computed from a stale price.

Terminal multiple
25×
Discount rate
12%

Our valuation

Our read

Low, base and high cases. A range, never a target.

$130 – $208

Base case $165

$165
$130
$208

Free cash flow grew 59% in FY2026. All three cases assume that decelerates sharply, because five more years at anything close to the recent rate is not a forecast anyone can defend. Net cash is cash and equivalents less total debt; marketable securities are excluded rather than estimated, which makes the band conservative.

Sources on every figureallfacts.ai · Not financial advice
Open the Nvidia card ↗
TSMC
TSM

Foundry & Memory

Insider

58%

High performance computing platform, 58% of 2025 revenue

7 blocks locked
Unlock with Insider
allfacts.ai
Broadcom
AVGO

Accelerators

Insider

31.3%

AI revenue of $20bn on $63.89bn total

7 blocks locked
Unlock with Insider
allfacts.ai
Microsoft
MSFT

Hyperscalers

Insider

13.1%

AI business annual run rate of $37bn, against $281.72bn of FY2025 revenue

7 blocks locked
Unlock with Insider
allfacts.ai
Meta
META

Hyperscalers

Insider

Not disclosed

Capex, FY2025: $69.69bn — 34.7% of revenue

7 blocks locked
Unlock with Insider
allfacts.ai
Vertiv
VRT

Power & Cooling

Insider

Not disclosed

Backlog, end of 2025: $15.0bn, up 109% year on year

7 blocks locked
Unlock with Insider
allfacts.ai
Palantir
PLTR

Applications

Insider

Not disclosed

Revenue split, FY2025: 54% government, 46% commercial — no AI line reported

7 blocks locked
Unlock with Insider
allfacts.ai
Menu

FACT PROVENANCE

Every fact has a receipt

The internet is filling up with confident nonsense about AI. Every number we publish carries its source and its date — visible, on the card, every time. If it can't be sourced, it isn't a fact and it doesn't ship.

Today's facts · 31 August 2026

  • Primarystraight from the source
  • Reportedvia a named outlet
  • Estimatea named analyst's projection
Supply chainPrimary

Every advanced AI chip needs packaging and testing before it ships, and ASE Technology, the world's largest outsourced semiconductor packaging and test house, is racing to add capacity for it. Equipment spending in its packaging and test segment jumped in the quarter to 30 June 2026, from $942 million a year earlier to $1.646 billion in ASE's advanced-packaging equipment spending, up 74.7% year-over-year

SourceASE Technology Holding·Q2 2026
Open this fact →
CapitalPrimary

Baidu's AI Cloud Infra revenue grew 50% year-over-year in the quarter to 30 June 2026, powered by GPU Cloud demand up 283%, and the company poured capital into keeping pace. Group-wide capital expenditure nearly tripled year-over-year to RMB11.39 billion ($1.68 billion), Baidu's Q2 2026 capital expenditure, up 199.7% year-over-year

SourceBaidu·Q2 2026
Open this fact →
HardwarePrimary

Monolithic Power Systems makes the chips that step grid power down to the voltages AI GPUs need. Its Enterprise Data segment, which sells into AI servers, grew far faster than the rest of the company in the quarter to 30 June 2026, when total revenue hit a record $980.6 million and Enterprise Data segment revenue reached $380.6 million in AI server chip revenue, up 164.3% year-over-year

SourceMonolithic Power Systems·Q2 2026
Open this fact →
EnergyPrimary

Quanta Services builds the transmission lines, substations and high-voltage switchgear that connect new AI data centres to the grid, and it told investors that rising demand from data centres, electrification and grid modernisation is driving orders. Its total backlog reached a record $53.4 billion, Quanta Services' record grid-construction backlog as of 30 June 2026

SourceQuanta Services·Q2 2026
Open this fact →

Browse every fact — the full archive, filterable by topic and source tier.

Menu

THE CHANNEL

Watch the boom get decoded

AllFactsAI on YouTube is where it started: the AI gold rush, the chips, the money and the winners — explained for investors in minutes, not hours. Start with the trailer.

Latest from the channel

Watch on YouTube

New videos every week

Menu

MEMBERSHIP

Three ways in. One club.

Spectator is live today: The Daily Receipts every weekday and The Weekly Brief every Saturday, both free. Insider and Pro are in development — founding members hear pricing before anyone else.

FREE

Spectator

Available now

  • First look at the Saturday film, and the AI stock of the day
  • The Daily Receipts (weekdays) and The Weekly Brief (Saturdays) — free, sign-up required
  • The full AI stock screener
  • Value chain, money loop and exposure map
Join the club + get the free brief

MEMBER

Insider

In development

  • Everything in Spectator
  • AI company fact cards, every company
  • The complete fact library, searchable
  • Every past issue of both briefs, searchable
  • Fact alerts when a tracked number moves
  • Watchlists, leaderboard and capex history
Free brief + founding list

ALL ACCESS

Pro

In development

  • Everything in Insider
  • Every card figure that changes at each earnings release
  • Pro Charts with indicators
  • Portfolio Connect and AI-exposure X-ray
  • Exclusive Pro videos
  • Data export and API
  • Custom tracked series
Free brief + founding list

THE DAILY RECEIPTS & THE WEEKLY BRIEF

Free, six days a week

The Daily Receipts, every weekday: four sourced AI facts, the market open, and one part of the site worth your time. The Weekly Brief every Saturday. Free, and every figure traces to its filing.

Edition of 31 August 2026

Every advanced AI chip needs packaging and testing before it ships, and ASE Technology, the world's largest outsourced semiconductor packaging and test house, is racing to add capacity for it. Equipment spending in its packaging and test segment jumped in the quarter to 30 June 2026, from $942 million a year earlier to $1.646 billion in ASE's advanced-packaging equipment spending, up 74.7% year-over-year

Every advanced AI chip needs packaging and testing before it ships, and ASE Technology, the world's largest outsourced semiconductor packaging and test house, is racing to add capacity for it. Equipment spending in its packaging and test segment jumped in the quarter to 30 June 2026, from $942 million a year earlier to $1.646 billion in ASE's advanced-packaging equipment spending, up 74.7% year-over-year

Read this morning's edition →

Free forever. Six emails a week — The Daily Receipts on weekdays, The Weekly Brief on Saturday — drop either in one click.