What shipped
Dated changes to the AI Buildout Market Map. When figures or structure move, it shows up here.
See the full roadmapDaily data refresh: Sesterce’s $11.21B Finland bet, Biohub’s $1.8B virtual-biology push, defense AI converts to contracts, Tencent’s $5B bond and the levered buildout
Today’s AI-economy briefing, verified and folded into the data. Capital & energy: Sesterce will invest $11.21B in a Finnish AI data-center site (phase 1 200 MW from 2026) — days after Finland halted Google’s site work over EIAs, capital and permission friction collide in the same country; Black Hills will spend $1.8B (2027–29) on 564 MW of new generation for Google’s Cheyenne DC, and Google signed a 3.59-GW Constellation contract ($4.3B+); Bitdeer AI added a 67 MW Malaysia campus with an AI-cloud pipeline topping $10B. Financing: Tencent is considering a $5B offshore bond for AI and compute infra, days after SoftBank’s ~$11.1B junk-bond raise — the buildout is increasingly levered, with 2026 AI-debt issuance at $489B. Defense: Shield AI won a $150M Navy contract (+$150M company match) for the exportable X-BAT autonomous fighter; the Army expanded its NGC2 AI battle network to Indo-Pacific I Corps (up to $1.8B); ASI took a $99M five-year Army C2 IDIQ; Empower AI won a $205M DCSA contract; TurbineOne was selected to expand frontline AI across the Marine Corps. Space: Google launched a 4-TPU Project Suncatcher prototype satellite to test orbital compute. Biotech: Biohub expanded its Virtual Biology Initiative to $1.8B (Meta, Google DeepMind, Isomorphic Labs, DOE, NIH); private biotech financings keep flowing (Rouge $58M, Sensible $47M, Vitalize $31M, WhiteLab $26M, Rivercell $25M) while tech deals stall. Robotics: Mecka raised $60M for human-motion data infrastructure (> $100M ARR run rate); Helm.ai signed $70M in contracts expanding into industrial automation; Universal Quantum raised >$100M; Teradyne invested in Bright Machines; Kawasaki unveiled the Home Leo eldercare robot (fiscal-2028 launch) and NTU demonstrated privacy-preserving home monitoring for older adults. FinTech: Stuut raised $52.5M for order-to-cash AI agents (clears today’s sourcing bar). Logistics & media: Freight Right launched an MCP for freight inside AI assistants; PSA and Granite Asia launched a $50M maritime innovation fund; Dailymotion’s Ray agentic video ads enter India. Policy: California requires disclosure of AI-generated people in advertising (enacted; effective 2027; private enforcement). Perspectives: a16z’s ‘Age of Atoms’ report — capex leadership has moved from bits to atoms (Bullish, Capex & earnings); Crunchbase’s Q3 2026 review — deal pace held as funding fell, capital concentrating at the top (Mixed, Valuations).
Daily data refresh: Manus’s $500M+ post-Meta raise, Nous’s $1.5B open-source agent bet, the AI-drug-discovery IPO window, Morgan Stanley’s mid-cycle call
Today’s AI-economy briefing, verified and folded into the data. Agents & venture: Manus raised $500M+ (Boyu Capital, IDG lead) after Beijing unwound Meta’s $2B+ acquisition — the China-US AI investment decoupling in one raise; open-source Nous Research raised $90M at $1.5B (Hermes: 24M+ clones, ~2.5% of global AI token usage) and launched a cost-per-task leaderboard — the open-source monetization question is now a market test. Biotech: Nvidia-backed Iambic Therapeutics filed for a US IPO targeting up to $159.4M at ~$806M; Danaher will launch an AI-powered autonomous research lab in 2027 (8x faster molecule discovery); Healthleap raised $38M for hospital patient-risk AI (50+ hospitals). Semiconductors: Nvidia-backed Upscale AI launched ‘Token Fabric’ multi-vendor chip networking; Microsoft’s Surface Laptop Ultra pairs the Nvidia RTX Spark chip with ‘Hybrid Intelligence’ on-device AI. Defense: Dutch Intelic’s AI drone interception went live under a €30M Dutch defence contract (Kyiv office on the battlefield); TurbineOne partnered with OpenAI for frontline warfighter AI; Overland AI unveiled the IBEX autonomous ground vehicle for Marine Corps counter-UAS; Beam Global agreed to acquire ScoutDI’s inspection drones; Anduril reportedly plans a $3.7B automated shipyard (secondary-only, awaiting tier-1). the Pentagon unveiled a new four-star Autonomous Warfare Command (Project Agincourt) to accelerate drones and AI across the joint force. Energy: CoreWeave locked in 240 MW at AdaniConneX Taloja for its India entry days after the 20-year AI-data-center tax break. Finland halted site work at two Google data centers until environmental assessments are complete (Al Jazeera, Oct 7) — the permission constraint biting even where power is secured. FinTech: Ledgebrook raised $200M for AI-native specialty insurance (Allianz X, Rockefeller Capital). Legal: Teddy AI announced a $60M seed disclosing almost nothing — included with caution. Sovereign: Azercosmos launched the Platerra AI satellite-imagery platform. Perspectives: Morgan Stanley’s Simkowitz — ‘still in the middle of the AI investment cycle’ (Bullish, Capex & earnings); Goldman pegs 2026 AI investment at ~$1.019T as the $489B AI-debt question becomes the credit debate (Mixed, Capex & earnings); a16z’s Olivia Moore — consumer AI monetization is concentrated: only 7 apps rank across all three leaderboards (Mixed, Enterprise adoption).
Daily data refresh: DeepSeek’s $12B round, Lambda’s $4B neocloud raise, Google’s second nuclear deal, Micron’s sold-out HBM
Today’s AI-economy briefing, verified and folded into the data. Capital: DeepSeek is closing at least ¥80B (~$12B) at a ~$75B valuation led by Tencent and CATL ahead of an early-2027 STAR Market IPO, with 160,000 Huawei accelerators planned for an Inner Mongolia data center (Bloomberg, Oct 6); neocloud Lambda is raising up to $4B at a $14.5B pre-money valuation — but ~$35B of its $50B backlog is a single Anthropic contract, and lenders are ‘getting choosier’ about data-center debt (WSJ, Oct 6); OpenAI is raising at least $30B at a $1.4T valuation while Anthropic eyes a November IPO at ~$2T (Bloomberg, Oct 6). Semiconductors: Micron’s 2027 HBM output is effectively sold out ($32B in advance commitments, up from $22B in June); Samsung’s Q3 profit is seen jumping ~9x on AI memory even as analysts fret about peak margins (Reuters, Oct 6); Vinci raised $250M at $1.5B for AI physics-simulation chip tools. Energy & grids: Google is nearing a second Constellation nuclear deal ($1B+, ~890 MW of US plant uprates) days after its 3.59 GW pact — ‘one contract: power + compute’ is now the default pattern — while India offered a 20-year tax break on overseas revenue from India-based AI data services (WSJ, Oct 6). Defense: Striveworks won a $9.75M Army award for agentic AI supporting Task Force RAVEN against a $70M enterprise agreement; Palladyne’s BRAIN made the Army’s xTech finalist list; Kaizen took a $49M IDIQ with a classified first task order; NATO’s new ‘Theory of Victory’ doctrine puts drone-AI mass at the core of the eastern flank. Policy: the UK accepted all 44 recommendations on regulating AI in healthcare; Australia weighed AI breach-reporting rules with OpenAI and Anthropic’s support; India weighed a national AI safety framework. Adoption: Einride’s Level-4 driverless electric trucks are hauling freight daily on Ohio roads; Lotte pilots humanoids across logistics processes; AGCO’s Tech Day pushed autonomy and AI across mixed farm fleets; AFRO TV took AI short films to national TV. Perspectives: a16z’s investment-cycle report sees capex with ‘no indication of a slowdown any time soon’ and hyperscaler free cash flow declining through ~2028 (Bullish, Capex & earnings); the mega-round pipeline shows equity, not operating cash, is the marginal dollar of AI capex (Mixed, Valuations); Lambda’s raise shows debt flowing only to contracted backlogs — $35B of $50B from one customer (Cautious, Capex & earnings).
Daily data refresh: Google’s twin utility deals, $489B of AI debt, Anduril’s $4.7B defense week, a $526M AI-drug-discovery wave
Today’s AI-economy briefing, verified and folded into the data. Power & utilities: on a single day (Oct 6), Google locked in 3.59 GW from Constellation ($4.3B of new investment — a 20-year 890 MW nuclear PPA plus a 2,700 MW 15-year supply agreement) and $1.8B of Black Hills-built generation (564 MW) for Cheyenne with Google covering the full cost; Constellation shares jumped 13% as utilities priced in as AI stocks. Financing: AI-related debt issuance hit $489B YTD 2026 vs Goldman’s $322B full-year forecast (Morgan Stanley sees ~$570B), hyperscaler leverage doubled in six months, and the credit market is ‘splitting in two’ — while Goldman’s Timothy Moe stays bullish (~$800B hyperscaler capex in 2026, ~$1.2T by 2027). Defense: Anduril won $4.7B of contracts in one week ($1.8B Army next-gen command-and-control with Lattice as the data layer, $2.9B Navy submarine-production), and the Pentagon signed seven firms (Google, OpenAI, AWS, Microsoft, Nvidia, SpaceX, Reflection) to deploy frontier AI onto classified networks. Compute: SpaceX closed a $1.11B/month AI hosting deal (~$13B annualized from Dec 1) toward a 2 GW-then-5–10 GW buildout. Biotech: Enveda raised $311M at $2B and BigHat $75M in a $526M two-day AI-drug-discovery wave — both companies have compounds in clinical trials; ByteDance spun off Anew Labs and raised $290M at $1.5B. Legal AI: Harvey raised $550M at $15.5B (up 41% since March), now used by 80% of the top-100 US law firms. Copyright: UMG and ElevenLabs signed a multi-year licensed AI-voice pact (opt-in artists, revenue to rights holders) — the licensing track vs. the litigation track. Agriculture: the 2026 Crop Robotics Landscape maps 400+ companies (+25% since 2024). Perspectives: Goldman’s Moe stays bullish (Bullish, Capex & earnings), Goldman/Morgan Stanley on the AI-debt wave (Cautious, Capex & earnings), and a16z’s $1.1B Machine Age Fund for AI hardware and infrastructure (Bullish, Chips & compute).
Daily data refresh: the FTC’s AI-agent probe, the first appeals ruling on AI training, Burry’s leveraged bear bet, a 3% US–China model gap
Today’s AI-economy briefing, verified and folded into the data. Policy: the FTC confirmed a formal investigation into OpenAI and Anthropic over consumer risks from autonomous agents — formal document requests within weeks — with California’s AG subpoenaing OpenAI (Oct 1) and 15 state AGs piling in; and the Third Circuit affirmed that ROSS Intelligence’s copying of Westlaw headnotes to train its legal-search AI was not fair use — the first federal appellate ruling on AI training, though the court explicitly left generative-AI questions open. Adoption & productivity: BCG’s Applied AI Index 2026 finds banks put 5.7% of revenue into AI (most of any sector), corporate AI spending doubled to 3.3% of revenue, and nearly half of companies now capture real value; NVIDIA’s healthcare survey shows $3.20 back per $1 invested in ~14 months, with 63% of health organizations actively using AI. Sovereign AI: India’s AI Mission can access only ~30,000 of 45,000 committed GPUs, so the Centre is building its own pool — 3,000 GPUs with C-DAC plus a UAE-‘gifted’ 8-exaflop supercomputer via G42 and Cerebras. Finance: 76% of treasury teams were hit by payment fraud in the past year (53% faced deepfake attempts), and Fideuram lost ~€95M to an AI-cloned voice — ~€36M still missing in crypto. Capital: Granite Asia ($11B AUM) and PSA International launched a $50M fund for AI in ports and logistics; OneByZero raised $20M to deploy governed AI agents across APAC enterprises. Perspectives: Michael Burry moved his AI-bubble timeline up and replaced shorts with leveraged June/Sept 2027 puts (Bearish, Valuations); Bloomberg Intelligence puts the US–China frontier model gap at 3% after DeepSeek V4.1 Flash (Bearish, Chips & compute); Goldman’s Peter Callahan says AI is becoming ‘a stock picker’s market’ as the trade broadens beyond semiconductors (Mixed, Valuations).
Manual data refresh (ontology test run): NYC's full-council AI hearing, Connecticut's 10^26-flops whistleblower law, Canada's AI council, Michigan's 20-year 'pay your way' power deal, a $1B Instinct round
A manual test run of the daily briefing, verified and folded into the data — and mirrored into the AI Suite ontology for the first time. Policy: all 51 NYC Council members convened an AI hearing (Oct 5) with two bills on the table — Intro. 26835 ($25,000 per-instance penalty for deploying AI without third-party validation, mandatory human kill switch) and Intro. 26887 (first-in-nation whistleblower bounty); Connecticut's Public Act 26-15 took effect Oct 1, making frontier developers above 10^26 flops let employees anonymously report 'catastrophic risk' (>50 deaths or >$1B damage) with anti-retaliation protections; Canada launched a 13-member National Council on AI (Bengio, Zatlyn, Pisano) advising on 'AI for All,' national champions, and sovereign infrastructure. Energy & grids: the Michigan PSC approved DTE's 20-year Google power deal — 80% minimum billing demand, 15-year early-termination coverage, explicitly so other customers don't subsidize the load. Capex & earnings: Instinct raised $1B at a $10B valuation (Sequoia, Benchmark, Coatue); EliseAI $350M at $4B and Armadin $255.5M at >$2.5B the same week — 9 of the week's 10 biggest US rounds were AI; GMI Cloud raised $445M in debt plus $223M in equity, another neocloud tilting toward the debt side of the financing wave.
Daily data refresh: the Super Intelligence Force, Amazon’s 15% chip-rental hike, Goldman’s AI-concentrated earnings, a $300M chip-smuggling indictment
Today’s AI-economy briefing, verified and folded into the data. Policy: Trump signed EO 14434 (Sep 29) rebranding AI as “Super Intelligence” in federal communications, then stood up the Super Intelligence Force (Oct 4) — DNI Jay Clayton chairs as formal AI czar, with a 120-day risk report due and a legal definition of “Super Intelligence” due in ~60 days that decides which commercial systems fall under national-security oversight. Enforcement: the DOJ arrested Earthmade Computer’s CEO (Oct 1) for allegedly smuggling >$300M of Nvidia A100/H100 servers to China via Malaysia and Singapore — the first prison-time case for AI-compute diversion. Capex & earnings: Amazon hiked EC2 chip-rental prices ~15% as Jassy confirmed $220B of 2026 capex with demand outstripping supply through 2027 (“the demand we already have for 2028 is striking”); JPMorgan AM counts ~$250B in bond-market plus ~$250B in bank-loan funding for hyperscaler capex — global cash underwriting the buildout, with off-balance-sheet and concentration warnings. Markets: Goldman’s Q3 preview sees S&P 500 EPS +27% with AI infra companies driving >50% of the growth — but breadth at dot-com-bubble lows and only +9% for the median company. Energy & grids: Gallup finds 71% of Americans oppose local AI data centers (worse than nuclear plants) — the permission bottleneck, quantified. Perspectives: Sequoia’s Pat Grady LP briefing (AGI arrived Nov 2025, $10–20T services prize, Bullish), McKinsey Global Institute’s 82-page workforce report (“mobility, not scarcity”, Mixed), Grayscale’s October AI-crypto outlook (+54% September, Bullish), and Goldman’s earnings-concentration note (Mixed).
Daily data refresh: Challenger's 120,136 AI-attributed layoffs, the Bank of England's AI-debt warning, Bain's $6T revenue test, Amazon's $8B chip leaseback, AWS's $1B community package
Today's AI-economy briefing, verified and folded into the data. Labor: Challenger's September tally makes AI the leading stated reason for US layoffs for the first time — 120,136 announced cuts YTD, ~21% of all US cuts — though September itself was quiet (AI ranked only fifth), and hiring plans hit their weakest September since 2011. Counterweight: NBER's new working paper prices AI as the equivalent of a permanent 32.6% productivity gain for software engineers, worth 3.6–6.5% of GDP. Capex & financing: the Bank of England's Financial Policy Committee reports global AI-related debt issuance doubled (~$450B YTD through early September vs. all of 2025), warning that 'circular arrangements' could amplify losses; Amazon is exploring an $8B Nvidia chip leaseback vehicle — off-balance-sheet financing in the wild. Bain's annual technology report sets the revenue test: AI needs $6T in annual revenue by 2031, with a $4.2T gap, requiring ~1% extra annual global GDP growth. Energy: AWS's 'Built Together' package (> $1B over five years for host communities, no NDAs, annual energy/water reporting, power costs kept off household bills) answers 100+ local moratoriums — and the $68B of US projects blocked or delayed in the June quarter. Policy: the FTC opened a broad consumer-protection investigation into Anthropic/OpenAI safety practices, and Trump is expected to name DNI Jay Clayton as AI czar. Perspectives: Bain's $6T revenue test (Cautious).
Daily data refresh: Tencent's $7B chip lease, Amazon's 20-year nuclear PPA, the halted PJM auction, TVA's data-center tariff, California's AI-layoff disclosure law
Today's AI-economy briefing, verified and folded into the data. Capex: Tencent signed a ~$7B five-year lease for ~100,000 advanced AI chips from Oracle data centers in Southeast Asia — the largest overseas lease on record, and a live demonstration of how US export controls govern physical transfer but not remote compute use. Japan's JERA, Dell, and RHAELM signed an MoU for a >$15B, 400 MW Chiba AI data center (behind-the-meter, skipping the grid queue); the FT estimates the broader Japan push at $140B. The Brookings BPEA draft now gives a unit cost for the boom: ~$8.2B per 200 MW campus (~$41B per GW), with a warning that financing risk is migrating to opaque off-balance-sheet structures. Energy: Amazon and Constellation signed a 20-year nuclear PPA covering 690 MW at Calvert Cliffs (~190 MW of new capacity, >$3B investment) — the grid's biggest customer becoming its builder. PJM's emergency 6.8 GW backstop auction was halted outright after FERC's cost-allocation objections (effective date pushed to Feb 28, 2027; TD Cowen warns of 'acute uncertainty'). And TVA's data-center tariff took effect: a dedicated rate class with a Capacity Commitment Charge, making AI loads pay for their own grid. Labor & policy: California's SB 951 requires employers to disclose AI-caused mass layoffs under the state WARN Act — the first law turning AI displacement into a reporting metric. Perspectives: Goldman's new agentic-commerce note maps a $2.6T US consumer opportunity as AI agents take over shopping, with payment networks as the winners (Mixed). Anthropic's IPO prospectus warns government attitudes could hurt its commercial ties — and that advanced AI could pose 'catastrophic or existential risks to humanity.'
Daily data refresh: FERC's 'data centers pay their way' ruling, the Senate's stalled ratepayer bill, Samsung's $1B Helix bet, CA's No Robo Bosses Act
Today's AI-economy briefing, verified and folded into the data. Energy: FERC partially suspended PJM's electricity procurement plan for 5 months, demanding that large users — including data centers — bear the costs of serving their demand (a regulatory 'who pays' ruling in the largest US power market). Congress stalled on the same question: the Ratepayer Protection Act fell 57–43 in the Senate, short of 60, after passing the House 417–3. Meanwhile DTE Energy's Michigan data-center pipeline topped 8 GW (Oracle 1.4 GW approved, Google 1 GW pending; up to $2B in utility capex) — with minimum monthly charges and collateral explicitly designed to shield existing ratepayers. Capex: six Samsung affiliates put $1B into Helix Digital Infrastructure, the KKR-created, Nvidia-backed vehicle now holding >$11B of committed capital for data centers plus power, transmission, and fiber. Morgan Stanley's new note counts $286B in US data-center projects delayed or cancelled since 2025 and sees hyperscaler/neocloud capex growth decelerating 93% (2026) → 61% (2027) → 14% (2028) as the cycle shifts to software. Labor: California signed the No Robo Bosses Act (SB 947) — no AI-only firings without human corroboration, operative July 2027, $500 penalties per violation. Policy: South Australia launched a Royal Commission into AI's economic impact (productivity, innovation, growth; report due July 2027). Perspectives: two new tracker rows — David George (a16z) on OpenAI's distribution moat (Bullish) and the Morgan Stanley capex-deceleration note (Cautious).
New: Investor Perspectives Tracker
What the money is saying about AI markets, in one place — a new tracker in the same row-per-item format as the Policy Tracker. Seeded with 12 perspectives from a16z, Sequoia, Social Capital, BOND, Goldman Sachs, J.P. Morgan, Morgan Stanley, and McKinsey: each with its key numbers, its stance (bullish, bearish, mixed, neutral, cautious), and a link to the original post or report. Highlights: a16z's State of Markets II (the 'bits to atoms' rotation, GPU prices climbing instead of decaying, only ~2% of the S&P 500 tracking AI metrics), Sequoia's AGI and 99.9%-cognitive-work theses, Chamath's land-power-shell bull case versus his 'tokenmaxxing' earnings warning, the $7.6T capex arc from Goldman and JPMorgan, Morgan Stanley's 38 GW power gap, and McKinsey's 11M-worker displacement read.
Daily data refresh: White House AI accord, McKinsey's 11M occupational switches, Nvidia's financing loop, Malaysia neocloud surge
Today's AI-economy briefing, verified and folded into the data. Policy: Trump and six AI-company leaders (Amodei, Pichai, Zuckerberg, Brockman, Huang, Musk) signed the White House Accord on Superintelligence — a voluntary four-tier framework (internal controls, independent external auditor, board-level committee) Trump calls 'morally binding'; it says the steps may later be codified into law. Meanwhile Nvidia and AMD are lobbying to strip China chip export provisions (AI OVERWATCH Act) from the FY2027 NDAA — Bernstein sees Nvidia's China AI-chip share falling from ~40% to ~8% by end-2026 as Huawei takes ~50%; and Italy's AI-safety criminal decree (No. 160/2026) entered into force today. Labor: McKinsey estimates ~11M US workers (~7% of the labor force) may need to switch occupations by 2035 — automation could cut 36M jobs but create 40M, with low-wage workers 8x more likely to face a switch; Workday disclosed a second 2026 layoff round (~500 jobs, 2.5%, mostly Product & Technology) in a Form 8-K. Capex: Nvidia added $150B to its buyback authorization (~$235B remaining through FY2028), and the FT reports it is in early talks with insurers on residual-value coverage for chip-backed loans to neoclouds — the debt-financed buildout keeps looking for someone to underwrite the risk. Energy: Bitdeer AI pre-committed >70% of its 21.7 MW Malaysia site (GB300 NVL72, Vera Rubin-ready) at >$1.7B expected revenue over five years — $1.7B of demand signed before the power is on.
Daily data refresh: the buildout's debt bill, the 66 GW grid number, White House AI meeting, AMD's $8.2B world-model bet
Today's AI-economy briefing, verified and folded into the data. Capex: the buildout is increasingly borrowed — nearly $500B in AI-related debt issued globally by Aug 5, 2026, with debt set to finance 35% of hyperscaler capex in 2027; Goldman now sizes cumulative 2026–31 AI infrastructure capex at ~$7.6T ($765B in 2026 → $1.6T in 2031). Chips: AMD is buying Fei-Fei Li's World Labs for $8.2B in stock (expected close end of 2026, pending approval) — the chip race moves from language models to physical-AI 'world models'; Li becomes AMD EVP & Chief Scientist. Energy: US data-center power demand runs 31 GW (2025) → 41 GW (2026) → 66 GW (2027), hitting 8.5% of peak summer US demand — and only 50–60% of the next two years' planned capacity is expected on time. Labor: the New York Fed finds just 4% of service firms laid anyone off because of AI in six months vs 13% hiring specifically to deploy it; over a third of AI-using firms retrained workers rather than replacing them. Policy: Trump and Speaker Johnson met AI CEOs at the White House today; Google/OpenAI/Anthropic are near launching the SAFA self-regulatory body (pre-release testing, incident reporting, auditor accreditation, targeted late-2026 launch); Florida asked a court to stop OpenAI building new models without outside guardrails.
Daily data refresh: Goldman $1.42T revenue hurdle, 2030 grid gap, Italy's AI crime, fresh layoff totals
Today's AI-economy briefing, verified and folded into the data. Capex: Goldman's new revenue-hurdle math — the six big US hyperscalers need ~$1.42T in cumulative 2028–2030 revenue (~$11.6B/GW/year at 15% ROIC) to justify 2026–27 AI compute capex; ~$300B/year just to break even; $1.69T in cloud backlog is the offset case. Energy: TrendForce counts 161 GW of data-center power demand in 2026 (AI servers 33.4%, past 40% by 2027) and a ~268 GW grid supply-demand gap by 2030 — power is the hard ceiling, not chips. Labor: fresh tracker totals — 225,122 tech layoffs across 519 events (SkillSyncer, Sep 28), Layoffs.fyi at 128,536 already past all of 2025; AI has been the leading stated reason for cuts three months running. Policy: Italy made failing to adopt safety measures on high-risk AI systems a criminal offense (Art. 437-bis) — with the company itself criminally liable. Chips: Beijing reportedly signaled it may let ByteDance and Alibaba buy Nvidia's new RTX Pro 5500; ByteDance is weighing ~1M units.
Daily data refresh: Moody's $700B+ capex call, AI-cited layoffs, Australia summons Altman
Today's AI-economy briefing, verified and folded into the data. Capex: Moody's forecasts US hyperscaler capex above $700B in 2026 — 6x 2022 — and $870B in 2027, but warns lack of electricity will constrain AI capacity through 2027 and that a data center takes 12–24 months from spend to revenue; bond spreads are widening as the giants borrow. Labor: Crunchbase's tracker counts 94,046 US tech layoffs Jan-Aug 2026 (+16.8% YoY), with AI cited in 33% of layoff events vs 1% in 2024 — though Layoffs.fyi notes little evidence AI is doing the laid-off workers' work. Policy: Australia's September consultation proposes mandatory tiered data-center standards (submissions close Oct 9, legislation early 2027); OpenAI agents breached US federal websites and a UN API, and Australia summoned Altman and Amodei to a Senate inquiry; ~75 US projects (~$130B) are blocked or delayed as 70% of Americans oppose a data center nearby. Energy: US data-center construction ran at a $59.3B/month pace by May 2026 (+129% vs two years prior, Census), and Bloom Energy claims its 800V DC fuel cells cut non-compute capex 27% ($3.6B) on a 1 GW campus — Oracle's $165B Jupiter campus hit schedule risk, not cancellation.
Daily data refresh: policy wave, AI net-jobs math, Anthropic wet lab
Today's AI-economy briefing, verified and folded into the data. The Policy Tracker grows 16 → 19: Maryland's executive order creating a single ≥25 MW review process, a task force, a monthly public dashboard and an NDA ban — plus a push to repeal the 2020 tax exemption that cost the state ~$22M in forgone revenue in 2026; Illinois' planned Power Act (bring-your-own-energy, closed-loop water) headed for the November veto session; and the bipartisan federal bill barring Chinese optical transceivers (Eoptolink, Innolight named) from sensitive systems, with a 5-year phase-out. Labor math: The Economist estimates AI has created ~1M US jobs vs ~200K AI-attributed layoffs since mid-2023 (~5:1), though customer-service and admin roles fell 10-15% and Challenger counts 116,175 AI-cited cuts through August. Healthcare: Anthropic's life-sciences push — a Bay Area wet lab for preclinical research, a Novo Nordisk R&D partnership, a CRISPR-like enzyme system Claude found autonomously, and a free clinical tool for ~100 lower-income countries.
Ontology connection: Impact Suite becomes tenant #1 (ontology branch)
The AI-Economy knowledge graph is now defined as a versioned ontology schema (8 object types, 6 link types, 4 governed actions), with an idempotent backfill script and a merge-sync GitHub Action that keeps the AI Suite ontology current after every data merge — without touching the daily PR workflow. The policy tracker also gains a ‘Follow the links’ panel: traverse each policy to its affected sectors, cited claim, jurisdiction, citation chain, and observation history. Work is on the ontology branch for review; nothing is automated into the morning briefing yet.
Sidebar: policy and economics trackers join the Map nav
The two new trackers are now one tap away in the sidebar — the AI Policy Tracker (frontier-AI rules, who-pays policy, copyright, national strategies) and the AI Buildout Economics Tracker (cost per MW, component math, orbital data-center economics) sit in the Map navigation alongside AI Economy Metrics.
New: AI Buildout Economics Tracker
Launched a dedicated economics tracker with 20 figures across five categories — build cost per MW ($8.9–23.3M excl. GPUs, +21% since 2024), component economics (GPUs are ~65–70% of a $3.4–5.5B all-in 100 MW facility), power & operating costs ($58.68/MWh PPAs, the CEPR who-pays number), land/water/resources (NoVA's $3.7M/acre record, 2.6M gallons per MW per year), and orbital data-center economics: the $2,720/kg → $500/kg → $200/kg → $100/kg breakeven chain, Suncatcher's thermal constraint, and Barclays' ~3× build-cost estimate. Derived math included: $1B buys ~18–30 MW of fully loaded AI capacity.
New: AI Policy & Regulation Tracker
Launched a dedicated policy tracker with 16 policies across four categories — frontier-AI safety (EU GPAI enforcement, CA SB 53, NY RAISE, China's agent rules), data-center 'who pays' policy (CA disclosure package, NY moratorium, PA consumer protections, NJ ratepayer law), AI & copyright (Anthropic $1.5B settlement, DOJ fair-use stance, CA AB 2013), and national AI strategy & governance (EO 14365, CO/TX state laws, China companion-AI ban, Australia's 2027 AI-law push). Everything cited; several flagship items take effect January 1, 2027.
Midday data refresh: record bond sale, DOE grid money, orbital AI
Afternoon sweep of the day's AI-economy news, verified and folded into the data: SoftBank's $11.1B bond sale — the largest high-yield corporate bond deal on record — to fund its $64.6B OpenAI commitment, as hyperscaler bond sales top $200B in 2026 (LSEG); the DOE's ~$2B grid-modernization push for AI-driven demand across 26 states; Atlas Energy's 283 MW (+328 MW framework) behind-the-meter gas power deal with a frontier AI lab; Google's Project Suncatcher orbital-AI test satellite launching October 1; and Basecamp Research's $140M Series C for AI-designed medicines built on proprietary biological data.
Daily data refresh: the $10.3T buildout, Goldman capex math, crowding-out
Today's AI-economy briefing, verified and folded into the data: Brookings projects $10.3T in data-center + AI infrastructure spending from 2025-2032 (~3.6% of GDP/yr, the largest single-industry buildout in US history as a GDP share); Goldman puts 2026 US AI investment at ~1.8-1.9% of GDP, sees hyperscaler capex hitting ~$800B in 2026 and $1.2T in 2027, and estimates ~$300B in annual AI revenue is needed to break even (>$500B for a 20% return); FactSet sees $4.2T in combined big-five capex through 2029, increasingly debt-funded. The power & grid page gains two new key metrics: construction crowding-out (data-center construction +$9B vs. all other private construction -$46B through July) and the Mississippi aluminum smelter that lost its power — and ~1,000 jobs — to a data center. Plus Gartner's finding that ~30% of employees laid off because of AI will need rehiring by 2029, at higher cost.
Tools suite trimmed to the strongest four
The usefulness eval (4.7/5 across 8 tools) found real problems in four tools — skills explorer, classroom planner, syllabus check, and action plan. Rather than ship them half-working, they're removed for now; the suite keeps the four strongest: study plan, leverage audit, career moat builder, and policy brief builder. The removed tools can return once their issues are fixed.
Tools evaluation: privacy fixes + usefulness eval suite
Ran every /tools persona tool interactively with fictional persona data (4.7/5 usefulness across 8 tools) and fixed what the eval found: tools no longer save anything to local storage before you enter data, and “Delete my data” now stays deleted. Also fixed sentence casing in the policy brief builder, made action-plan logic shareable and testable, and added a permanent eval suite (evals/tools) with persona fixtures and ratings so regressions get caught.
Persona tools suite: 4 interactive /tools, private by design
Four guided tools for students, professionals, and policymakers — AI-proof study plans, AI leverage audit, career moat builder, and policy brief builder. All answers stay in the browser's local storage; download or delete your data anytime as JSON. No accounts, no tracking.
Perspectives index + Collaborators tab
The six thin perspective stubs are retired in favor of a single honest index: Power & grid stands as published — the data-centre buildout deep dive — while materials & mining, VC & capital, the application layer, jobs & labor, and students & skills show as "In research" with their open questions. The duplicate power & grid page redirects to the buildout hub. Also new: the Collaborators tab, publishing human perspectives from people building the AI economy — practitioner opinions under their own bylines, clearly separated from our data research. First two pieces: Bhavin on how public markets price AI, and Nishi on AI’s healthcare ROI starting with the back office.
Feedback emails, chat export, share buttons
The About-page feedback form now actually delivers: submissions are emailed to the team via Resend (with honeypot, rate limiting, and a Supabase backup row). The Ask chat panel gains a Download .md button that exports the conversation with timestamps. Tracker pages get Share buttons (X, Facebook, LinkedIn, copy link), and selecting any passage on article pages offers Share quote.
Glossary + methodology go live
Two new reference pages: a 20-term glossary in plain English (linked in the footer) and a full methodology page — source hierarchy, confidence tags, vintage policy, corrections — linked from the About page's 'Open methodology' heading, which previously opened nothing.
Start-here guide: three 10-minute paths
New visitors get a proper on-ramp: /start offers guided paths for students, educators, and strategists — 4–5 ordered steps each, every step saying what you'll learn. Linked from the homepage hero.
Sectors index: the 22-sector taxonomy goes public
The full 22-sector taxonomy is now public under Sectors in the top nav — each sector marked 'Brief published' or 'In research'. First three briefs ship today: Healthcare (Abridge's 150+ health systems, ~$150B/yr projected savings), Manufacturing ($3.8T projected AI contribution by 2035), and Global FinTech ($447B+ projected banking savings by 2030). Every figure sourced; every brief carries an honest 'where the data runs out'.
Citeable pages: citations + CSV downloads
The power & grid map and AI Economy Metrics now carry a 'How to cite this page' box with a copy button, and both metrics tables (AI Economy Metrics, Numbers to know) get a Download CSV button for the visible rows — exports, citations, and methodology for researchers and educators.
Trust fixes: feedback form, stub badges, nav
About page feedback form now asks about the Impact Suite instead of another project. The 13 in-research perspective pages carry a visible 'Coming soon — in research' badge (with a 'soon' tag in the chip nav), so stubs no longer read as finished content. AI Economy Metrics joins the top nav, footer, and mobile menu.
Daily data refresh: AI capex, power & who pays
AI Economy Metrics: Goldman Sachs puts AI investment at almost half of 2026 S&P 500 EPS growth; Dell'Oro reports data-center capex +92% YoY in Q2 2026. Power & grid map: California's 7-bill 'who pays' package (signed Sep 22) — data centers must fund their own grid and water upgrades.
AI ecosystem bird's-eye map
Added a React Flow overview under Map — the full buildout market on one canvas. Click any node to open that perspective; the power & grid map remains the deep dive for numbers and who wins / who pays.
Market sleeve: hub + layer strands
Connected perspectives across the buildout market — chips, materials, models, capital, policy, safety, jobs, and more — now sit alongside the power & grid map, with roadmap and changelog under Updates.
Public roadmap & product focus
Published the AI Buildout Market Map roadmap and changelog. Soft-hid proposals and tracker datasets from public chrome so the front door is the buildout map. Perspective exploration folder opened for layer and impact strands.
v1 power-edge impact map
Shipped the first living map: players, key metrics, charts with connecting annotations, cascade and graph of who wins / who pays, confidence tags, disclosed data gaps, and Ask / Context / Path exploration.
Something wrong or missing? Send feedback.