Week 23 · June 1 – June 7, 2026

high signal

Open the Books

The two most valuable private companies in AI filed to go public within ten days of each other — Anthropic on June 1 at a $965 billion valuation, OpenAI around May 22 targeting a September listing — and SpaceX targets a June 12 Nasdaq debut, the largest private-to-public capital transition the industry has attempted. The filings will force the first audited unit economics for a frontier AI lab into daylight: inference cost, gross margin, customer concentration. They land in the same week the boom got its first real crack — Broadcom's June 3 guidance miss wiped 14 percent off the stock and dragged the whole semiconductor complex down, while Ray Dalio called the bubble — and the labor data crossed a line, with AI becoming the single largest self-reported reason for US layoffs for the first time (40 percent of May's 97,006 cuts) even as headline payrolls beat at +172,000. Microsoft chose the same week to ship its first in-house frontier models trained without a line of OpenAI; Google's Co-Scientist landed six wet-lab-validated discoveries in Nature; and the only hard US AI-safety law on the calendar — Colorado's — was repealed by its own governor before it could take effect. P-012 (the bubble bursts by 2026) resolves refuted at month-end with zero of five burst criteria met — the cracks are real, but they are not a burst.

6 predictions updated5 milestones25 companies refreshed

Key Developments

Anthropic filed a confidential S-1 on June 1 at a $965 billion valuation — and with OpenAI (around May 22) and SpaceX (June 12) the three biggest names in the boom are going public at once, forcing the first audited frontier-lab economics into daylight

Anthropic filed a confidential draft S-1 with the SEC on June 1, 2026 — confirmed by CNBC, the Washington Post and CBS — the first frontier AI lab to take concrete steps toward a public listing. It follows the $65 billion Series H at a $965 billion valuation (closed May 28) and a disclosed annualized run-rate of $47 billion, up roughly 5x from $9 billion at year-end 2025. OpenAI filed confidentially around May 22, targeting a September listing at $852 billion to $1 trillion despite losing about $1.22 for every $1 of revenue (~$14 billion annual burn on ~$24 billion ARR), and SpaceX targets a June 12 Nasdaq debut. Together they are the largest simultaneous private-to-public AI capital transition ever attempted. The S-1, once public, will produce the first audited unit economics for a frontier lab — inference cost, gross margin, customer concentration — numbers that have been opaque through every private mega-round to date.

Challenges P-012 ↘ AI investment boom is a bubble likely to burst by ...Confirms P-030 ↗ AI-native companies can achieve billion-dollar rev...Challenges ↘

Counterpoint

The disclosure cuts both ways. Anthropic's $47 billion is annualized gross billing, not GAAP-recognized revenue — OpenAI's CRO already disputed the April $30 billion figure by roughly $8 billion on gross-vs-net grounds, so recognized revenue is likely $33-40 billion. OpenAI's own numbers are the sharper warning: a company losing $1.22 per $1 of revenue asking public markets for a $1 trillion valuation is pricing in a profitability inflection it has not demonstrated. IPO timing is still speculation plus pre-filings, not a confirmed date, and a $965 billion private company with no clean precedent faces SEC scrutiny that will force the gross-vs-net question into resolution one way or the other. The filings are the strongest possible refutation of P-012's bubble-burst thesis — you do not file to open your books if they are empty — but they also set the exact trap a real correction would spring.

Source →

Broadcom's June 3 guidance disappointed for the first time in eight quarters — the stock fell 14 percent and dragged the whole semiconductor complex down, giving the AI trade its first real crack just as P-012 prepares to resolve refuted

Broadcom reported Q2 FY2026 on June 3 with revenue $22.2 billion (+48 percent year over year) and AI semiconductor revenue $10.8 billion (+143 percent) — both beat — but management declined to raise the $56 billion full-year AI target the sell-side had assumed would move up after a front-loaded first half. The stock fell about 14 percent, pulling AMD, Intel and Micron down with it, the first major AI-chip disappointment in eight quarters. Ray Dalio used the same week to call the AI bubble bursting on liquidity grounds, and the SaaS rout deepened (the IGV software ETF is down ~20 percent year to date, Figma down ~86 percent from its high). Against that, the boom printed records: Q1 2026 venture funding hit $330.9 billion (81 percent AI), NVIDIA's Q1 grew 85 percent, and hyperscaler 2026 capex was raised, not cut, to $660-725 billion. Of the five operational burst criteria, zero are met — P-012 resolves refuted at 2026-06-30.

Challenges P-012 ↘ AI investment boom is a bubble likely to burst by ...Confirms P-017 ↗ Annual data center CapEx will grow from $500B (202...Confirms P-038 ↗ Approximately 1% of total US economic growth in 20...Confirms ↗

Counterpoint

The honest bear case is not equity, it is debt. Broadcom's miss is a valuation-multiple story — 200 percent year-over-year AI revenue growth is not demand destruction — but the data-center build-out is increasingly financed with debt ($200 billion-plus issued in 2025, $250-300 billion projected for 2026) against GPU collateral that depreciates in roughly a year while the debt assumes 7-15-year lives. The Federal Reserve has flagged AI infrastructure as a top systemic-stability concern, and CoreWeave carries a $4.2 billion 2026 refinancing wall. None of this is a June 2026 burst, but it is the precise channel through which a version of P-012 could come true in 2027-2028 — after this prediction's resolution date. The cracks are real; the timing is early.

Source →

For the first time ever, AI was the single largest reason US employers gave for layoffs — 40 percent of May's cuts — even as headline payrolls beat at +172,000 and the New York Fed disputed the attribution

The Challenger, Gray and Christmas report for May 2026 recorded 97,006 announced US job cuts (+16 percent month over month, the highest May total since 2020), of which 38,579 — 40 percent — were attributed to AI. It is the first month AI has been the single largest self-reported layoff driver since the category began in 2023, up from 7 percent in January; year-to-date AI-attributed cuts (87,714) already exceed the full 2025 total of 54,836. The June 6 BLS report complicated the picture: payrolls beat at +172,000 with unemployment steady at 4.3 percent, but financial-activities employment fell 22,000 and is down 107,000 since its May 2025 peak. This is The Hollowing in the hard data — Goldman puts AI substitution at -16,000 net jobs a month, Yale finds software-developer employment for ages 22-25 down ~20 percent — but a New York Fed study (May 14) found AI-exposed posting declines began before ChatGPT and show no clean junior-vs-senior divergence. The trend is real; the attribution is the most contested number in the labor data of 2026.

Confirms P-002 ↗ AI will disrupt 50% of entry-level white-collar jo...Confirms P-020 ↗ AI-exposed industries will see job losses of ~20,0...Confirms ↗

Counterpoint

Challenger tracks employer self-attribution, not verified causation — and AI is a convenient narrative for post-pandemic over-hiring corrections, a point Sam Altman himself has made. Two Federal Reserve studies are the counterweight: the NY Fed's event study around ChatGPT finds no statistically significant divergence between high- and low-AI-exposure occupations after launch, and the Fed Board's BTOS-linked analysis (through November 2025) finds AI adoption explains 0.04-0.13 percent of posting changes. The catch is that both Fed studies cut off before the Q1-Q2 2026 acceleration that Challenger captures — so the real question, which September's BLS occupation data will answer, is whether a genuine attribution break happened this spring or whether the self-reported number is running ahead of the employment reality. The +172,000 beat says the macro is not cratering either way.

Source →

What the Evidence Moved

P-001AI models will handle most aspects of software engineering t...

Microsoft shipped MAI-Thinking-1 + MAI-Code-1-Flash at Build (June 2) — first in-house frontier models without OpenAI distillation, Flash live in all Copilot tiers; SWE-bench Verified is saturating (Mythos 93.9%) and METR now flags autonomous task-horizon measurements above 16 hours as unreliable. Counter-signal: a 33,000-PR study finds bug-fix is the task type agents fail most.

82%→83%▲ +1pp

P-004AI systems will be able to discover genuinely novel scientif...

Google Co-Scientist published in Nature with six independent wet-lab validations (Stanford repurposed Vorinostat for liver fibrosis, 91% organoid effect; ERA beat the CDC COVID forecasting ensemble). Strongest evidence yet for AI generating externally validated novel scientific hypotheses.

75%→78%▲ +3pp

P-012AI investment boom is a bubble likely to burst by 2026

Final pre-resolution check, ~3 weeks to target_date 2026-06. First genuine bear signals appeared (Broadcom guidance miss -14%, Dalio bubble warning, SaaS rout) but all sub-threshold — zero of five burst criteria met. Q1 VC $330.9B, NVIDIA +85%, Anthropic + OpenAI both filing S-1s. Resolves refuted at 2026-06-30.

5%→5%▼ 0pp

P-017Annual data center CapEx will grow from $500B (2025) to $1.4...

Hyperscaler 2026 capex confirmed at $660-725B (+77% YoY) with Dell ($16.1B AI servers), Marvell and NVIDIA ($75.2B Data Center, +92%) printing supplier-side revenue at matching scale; Dell'Oro and Morgan Stanley now forecast above ARK's $1.4T-by-2030 number. Broadcom miss + GPU-debt mismatch are the downside channel.

55%→57%▲ +2pp

P-022Over 40% of agentic AI projects will fail by 2027 due to uni...

MSR 2026 peer-reviewed study (11,771 PRs: agents introduce 79% of CI failures, fix 61%) plus a 33,000-PR companion study (bug-fix worst merge rate); Deloitte State of AI 2026 finds only 8.6% of companies have agents in production, 79% report adoption challenges, 54% of C-suite say AI is 'tearing the company apart.' Deployment without value capture remains dominant.

Data refresh

Q1 FY2027 — revenue $81.6B (+85% YoY), Data Center $75.2B (+92%), Q2 guided $91B excluding China after BIS June 1 guidance blocked Blackwell to Chinese-parented firms worldwide (Huawei now ~60% China share). Dimensions at ceiling; score holds 8.50.

Open the Books

Key Developments

Anthropic filed a confidential S-1 on June 1 at a $965 billion valuation — and with OpenAI (around May 22) and SpaceX (June 12) the three biggest names in the boom are going public at once, forcing the first audited frontier-lab economics into daylight

Broadcom's June 3 guidance disappointed for the first time in eight quarters — the stock fell 14 percent and dragged the whole semiconductor complex down, giving the AI trade its first real crack just as P-012 prepares to resolve refuted

For the first time ever, AI was the single largest reason US employers gave for layoffs — 40 percent of May's cuts — even as headline payrolls beat at +172,000 and the New York Fed disputed the attribution

What the Evidence Moved

Company Impact

Palo Alto Networks

Amazon

Applied Materials

Abbott Laboratories

American International Group

American Express

American Tower

Microsoft

Broadcom

NVIDIA

Funding Snapshot

Regulatory Calendar

Capability Milestones

Sources