AI Business · 16 May 2026

AI Business — 16 May 2026

Analysis

House preemption talks converge with a Trump executive-order push as Mythos forces a regulatory inflection point. Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) are negotiating a federal bill that would block frontier-AI laws in California and New York for two years with a sunset, fighting over whether vetting should be compulsory or voluntary — talks moving in parallel with a Trump executive-order push to vet (or pre-clearance) advanced AI models after Anthropic's Mythos reportedly found cyber vulnerabilities human hackers missed. Industry-supported "reverse federalism" runs alongside: both OpenAI and Anthropic this week backed Illinois SB 315 mirroring CA/NY rules to set a de facto national floor, while Trump told reporters he separately raised AI guardrails and Nvidia's H200 chips with Xi in Beijing.
Anthropic's $30B at $900B valuation lands as the second AI-lab IPO-de-risking event of the week, with OpenAI signalling more rounds may follow. Dragoneer, Greenoaks, Sequoia and Altimeter each likely investing $2B+ to co-lead, nearly tripling the $350B valuation from three months ago as annualised revenue approaches $45B — a 5x jump from $9B at end-2025 and a step past OpenAI's $852B mark. Concurrently, OpenAI CFO Sarah Friar told Bloomberg the company may raise additional capital beyond its $122B round (described as the largest private fundraise ever) as the compute crunch deepens, while Anthropic separately declared certain secondary-market sales of its private shares "void," rattling retail platforms.
Asymmetry: the AI-chip rally has crossed historical bubble spreads on one BofA measure, yet Wall Street keeps buying memory and physical-AI names. Hartnett flags the SOX 62% above its 200-day average — wider than the Nasdaq's 55% spread pre-dot-com (2000) and approaching the 73% CAC pre-Mississippi-Bubble (1720) — yet Sandisk (+482% YTD) has gotten cheaper on P/E as memory demand outruns earnings forecasts, JPMorgan lifted its Taiex bull-case to 50,000 calling Taiwan "the most pure-play AI exposure," and Asia robotics names lead 2026's hottest theme as Figure's humanoids sorted packages 50 hours unattended. Capacity strain is bleeding into 8" wafer markets at ~90% utilisation, with SMIC absorbing TSMC mature-node overflow at 89% China revenue.
Safety-and-incidents converge on the audit-trail problem across consultancies, peer review and autonomous vehicles. EY-Canada retracts its loyalty-fraud cybersecurity study citing a non-existent McKinsey report and >6 fake footnotes (Deloitte-Canada and Sullivan & Cromwell were the prior precedents); journals report 60–100% YoY submission spikes as agentic tools like OpenAI's "Prism" produce publication-ready papers in 26 minutes, with STM Integrity Hub now considering submitter-authenticity attestation since fraud detection is becoming impossible. Tesla's NHTSA disclosure of 17 Austin Robotaxi incidents reveals 2 crashes were caused by humans remotely driving cars into a metal fence (8 mph) and a barricade (9 mph) — a teleoperation regime more permissive than Waymo's 2-mph cap. YouTube concurrently extends AI likeness-detection takedown to all 18+ users — the first mass-market deepfake-defence tool from a Big Tech platform.
Catalysts in the next 48–72 hours that test the day's main thesis:
- [POLICY] 18 May — Musk v. Altman jury begins deliberations; advisory verdict could land within the week, potentially threatening OpenAI's 2025 PBC conversion, Altman/Brockman roles, and the path to a sub-$1T IPO filing.
- [INFRA] 19 May — Nvidia FY27 Q1 earnings (Wednesday); consensus AI-capex outlook print into a tape that has overshot historical bubble spreads on one BofA measure.
- [INFRA] 21 May — 45,000 Samsung workers begin an 18-day strike unless a bonus deal lands; JPMorgan flags ₩21–31T ($14–21B) op-profit risk and HBM/base-die supply exposure for Nvidia and Tesla.

POLICY & REGULATION

LEGPOLITICO | House talks look at blocking some state AI laws, including in California and New York | Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) are negotiating a federal preemption bill with a two-year sunset that would block California and New York frontier-AI laws, with a disputed federal vetting regime (voluntary vs. compulsory) moving in parallel with a Trump executive-order push following Anthropic's Mythos cyber-bug-finding model.
ENFBloomberg | Arm Holdings to Face US Antitrust Probe Over Chip Tech | The US Federal Trade Commission is investigating Arm's semiconductor IP licensing practices, joining ongoing global antitrust scrutiny of the SoftBank-controlled architecture supplier whose IP underpins nearly every smartphone SoC and a growing share of AI accelerators.
TRADEBloomberg | Trump Says He Discussed AI Guardrails, Nvidia's Chips With Xi | Trump told reporters on Air Force One after the Beijing summit that he raised "possibly working together for guardrails" with Xi and that Nvidia's H200 chips came up — without announcing any new restriction or expansion beyond the 75,000-chip cap and 10-firm approval list reported earlier in the week.
TRADEAxios | The 3 big conflicts in AI race against China | Commerce assessment puts DeepSeek V4 Pro about eight months behind US frontier as Treasury Secretary Scott Bessent confirms an AI-safety protocol with Beijing while Senators Coons (D-Del.) and Banks (R-Ind.) urge restraint on H200 access — and OpenAI and Anthropic both back Illinois SB 315 to seed a "de facto" national standard via state-law mirroring.

PARTNERSHIPS & ENTERPRISE

PRODWired | Greg Brockman Officially Takes Control of OpenAI's Products in Latest Shakeup | OpenAI is consolidating ChatGPT, Codex and its developer API into one core product team under Brockman (Simo remains on medical leave), naming Thibault Sottiaux to lead the merged platform and moving Nick Turley (ChatGPT's 900M-WAU lead since 2022) to enterprise — ahead of a planned IPO filing later this year.
ALLYBloomberg | OpenAI Taps Plaid to Bring Tailored Financial Advice to Masses | ChatGPT will use Plaid's bank-account aggregation to deliver personalised financial advice to consumers — OpenAI's first formal data-licensing tie-in with a US fintech-infrastructure provider as it pushes beyond generic guidance.
ENTBloomberg | UnitedHealth Tracks Workers' AI Use in Push to Transform Company | UnitedHealth is monitoring how often Optum-services workers use AI tools as part of a company-wide deployment push — among the most explicit corporate AI-adoption-tracking programmes at a Fortune 10 employer, tying tool usage to performance metrics.
PRODVentureBeat | Claude's next enterprise battle is not models: it's the agent control plane | VB Pulse February data show Microsoft Copilot Studio/Azure leading enterprise agent orchestration at 38.6%, OpenAI Assistants at 25.7%, and Anthropic Tool Use registering a first measurable foothold at 5.7% — alongside Claude's model-layer share rising from 23.9% (Jan) to 28.6% (Feb) to a directional 56.2% (Mar).

GOVERNANCE & SAFETY

INCIDFinancial Times | EY retracts study after researchers discover AI hallucinations | EY Canada withdrew its "Points of Attack" loyalty-fraud cybersecurity study after GPTZero researchers found a non-existent McKinsey citation, more than six fake footnotes and a $200bn market-size figure appearing inconsistently — the latest Big-Four AI-hallucination incident following Deloitte's Canadian provincial-government revision and Sullivan & Cromwell's apology to a New York court.
INCIDThe Verge | AI-generated research papers are overwhelming peer review | Security Dialogue submissions are up 100% YoY and Accountability in Research up 60%, with agentic tools like OpenAI's "Prism" producing publication-ready papers in 26 minutes — STM Integrity Hub now studying watermarking and submitter authenticity attestation as fraud detection becomes impossible.
GOVMIT Technology Review | Musk v. Altman week 3: Musk and Altman traded blows over each other's credibility. Now the jury will pick a side | Jury deliberations begin Monday on Musk's bid to unwind OpenAI's 2025 PBC conversion, remove Altman and Brockman, and award $134B to OpenAI's nonprofit — advisory verdict expected within the week ahead of OpenAI's planned IPO filing at a valuation approaching $1T.
INCIDWired | Tesla Reveals New Details About Robotaxi Crashes -- and the Humans Involved | Tesla's federal disclosure of 17 Austin-area Robotaxi incidents (Jul 2025–Mar 2026) reveals 2 crashes caused by Tesla's own remote drivers steering autonomous cars into a metal fence (8 mph) and a construction barricade (9 mph) — Tesla operates a more permissive teleoperation regime than Waymo's 2-mph cap on remote driving.

RESEARCH & MODELS

MODELWired | Mira Murati Wants Her AI to 'Keep Humans in the Loop' | Thinking Machines Lab previewed "interaction models" that natively process continuous voice/video conversation rather than transcribe-then-respond — positioned as a counter-thesis to OpenAI/Anthropic/Google autonomy-maximisation; the models are unreleased and Tinker fine-tuning API remains the lab's only shipping product since its multi-billion-dollar raise.

COMPUTING & INFRASTRUCTURE

CHIPReuters | At Samsung, the global AI boom spurred a looming strike and deep divisions | 45,000 Samsung workers are threatening an 18-day strike from 21 May over bonus disparities — Samsung proposed 607% of annual salary for 27,000 memory workers (matching SK Hynix) versus 50–100% for 23,000 logic/foundry workers who build "base die" for Nvidia and Tesla; JPMorgan estimates ₩21–31T ($14–21B) operating-profit hit and risk to global HBM supply.
CHIPSouth China Morning Post | 'Panic' over capacity crunch in mature node chips drives orders to Chinese fabs | TrendForce sees top-10 foundry 8" wafer utilisation hitting ~90% in 2026 (from 80% in 2025) as PMIC demand for AI servers absorbs capacity; SMIC's Q1 wafer utilisation hit 93.1% with China revenue at 89% as TSMC shifts 12" mature-node capacity toward high-margin AI work.
DCAxios | Seattle weighs temporary ban on large data centers | Councilmember Eddie Lin is moving a one-year moratorium after four operators proposed five facilities collectively drawing up to 369 MW (about one-third of Seattle's current average load); two of the four have withdrawn as the city drafts rules on noise, water and emissions.

GENERATIVE MEDIA & EMBODIED AI

GENThe Verge | YouTube is expanding its AI deepfake detection tool to all adult users | YouTube is opening its selfie-scan likeness-matching takedown system to all users 18+ after staged rollouts to creators, public officials and the entertainment industry — the first mass-market deepfake-detection tool from a Big Tech platform, with carve-outs for parody/satire and no coverage of voice-likeness.
BOTBloomberg | Robotics CEO Vows No Intervention in Humanoids' Viral Trial Run | Figure CEO Brett Adcock said Figure's humanoids sorted packages for around 50 hours nonstop without intervention while the company manufactures 60–70 robots weekly, targeting 90% success on package flipping for barcode scanning.

FUNDING & DEALS

RNDFinancial Times | Anthropic agrees terms of $30bn funding deal at $900bn valuation | Dragoneer, Greenoaks, Sequoia and Altimeter each likely investing $2B+ to co-lead, nearly tripling Anthropic's $350B February valuation as annualised revenue approaches $45B (5x year-end 2025); the round is expected to close as soon as this month, with Amazon and Google sitting out the new money but remaining large existing backers.
RNDBloomberg | OpenAI May Raise More Money as Compute Crunch Deepens, CFO Says | CFO Sarah Friar told Bloomberg the company may raise additional capital beyond the recent $122B round (which she described as the largest private fundraise ever) as the compute crunch deepens — landing the same day Anthropic's $30B/$900B FT scoop leaked.
RNDThe Wall Street Journal | So You Think You Own Shares in a Hot Startup? Anthropic Says Not So Fast | Anthropic declared sales of its private stock by unauthorised firms void, rattling individuals who used SPVs and secondary-market platforms to back the AI lab alongside SpaceX and OpenAI — the clarification arrived hours before the $30B/$900B primary-round terms leaked.
IPOBloomberg | STT Global Data Centres Is Said to Plan $500 Million India IPO | The Tata Communications-backed Indian data-center operator is considering a Mumbai IPO of up to $500M as digital-infrastructure investor demand accelerates in India — running parallel to Kioxia's planned US listing and on the same tape as the AirTrunk Malaysia DC plan reported Thursday.
IPOBloomberg | Kioxia Prepares US Listing After Riding AI Boom to Record Profit | The world's best-performing major stock this year — Bain Capital-controlled NAND maker — said it is preparing an ADR listing as historic memory prices drive record profit, offering a second sovereign-equity-market route for AI-memory exposure beyond the Sandisk/SK Hynix rally.

Calendar

POLICY18 May — Musk v. Altman jury begins deliberations; advisory verdict may land within the week.
INFRA19 May — Nvidia FY27 Q1 earnings (Wednesday); AI-capex outlook print after BofA flags historic bubble spreads on the SOX.
INFRA21 May — Samsung 45,000-worker strike start (18 days planned); JPMorgan flags up to ₩31T operating-profit risk and HBM/base-die supply impact.
DEALSMay (TBD) — Anthropic $30B/$900B round expected to close "as soon as this month" per FT.
POLICYmid-June — Apple ordered to produce SVP Craig Federighi's documents to Musk under court order; Apple expected to unveil revamped Siri in June.

Data Coverage: ~260 articles scanned across 9 query domains over the 24-hour Friday window, ~24 retrieved at full text. Cross-day dedup: yesterday's OpenAI-Apple legal-options story rolled forward into NYT's "Considers Legal Action" headline and Ars Technica's "burned" backstory (absorbed into Analysis 4 / cross-context, no standalone bullet — judge has now ordered Federighi document production for mid-June); yesterday's Cerebras $185/$5.55B/$40B IPO open + $350/100B/$3.2B Feldman-fortune coverage rolled forward into WSJ's editorial "Nvidia fatigue" framing (absorbed into Analysis 3 as market-thesis read); yesterday's Codex-on-mobile [PROD] bullet rolled forward into Semafor commentary (deferred — no new fact); yesterday's PJM 76% power-price coverage scanned again on TechCrunch byline (deferred — same fact). Within-window dedup: FT preferred over WSJ for Anthropic $30B/$900B given primary-source $2B-per-co-lead detail; WSJ preferred for the share-void story given retail-investor specificity; Bloomberg preferred over Verge for OpenAI-Plaid on Plaid mechanics; Wired preferred over TechCrunch for Tesla Robotaxi crashes on teleoperator specificity; MIT Tech Review preferred over CNBC/WSJ for Musk-Altman week 3 on trial-mechanics depth. US 60% / Europe 9% / China 14% / APAC ex-China 12% / Other 5%. POLICY at 4 (House preemption, Arm probe, Trump-Xi AI guardrails, US-China 3 conflicts). PARTNERSHIPS at 4 (Brockman/OpenAI reorg, OpenAI-Plaid, UnitedHealth AI tracking, Claude agent control plane). GOVERNANCE at 4 (EY hallucinations, peer-review slop, Musk-Altman week 3, Tesla Robotaxi crashes). RESEARCH at 1 (Murati/Thinking Machines interaction models). COMPUTING at 3 (Samsung strike, SMIC capacity crunch, Seattle DC moratorium). GENERATIVE MEDIA at 2 (YouTube likeness detection, Figure 50hr trial). FUNDING at 5 (Anthropic $30B/$900B, OpenAI compute-crunch follow-on, WSJ Anthropic share-void, STT India IPO, Kioxia US listing). Pershing Square Microsoft stake, Alphabet ¥576.5B yen bond, JPMorgan Taiex bull case, AI chip-bubble historical comparison, Cannes filmmakers + AI acceptance, Claude Code product-lead interview, US heavy job-losses AI feature, and Runway profile scanned and absorbed into Analysis as market-context rather than standalone bullets.