What changed
- Pentagon "never again single-threaded with any one model" — Under Secretary Emil Michael at SCSP AI+ Expo (May 7) framed eight new agreements (AWS / Google / Microsoft / NVIDIA / OpenAI / Reflection / Oracle / SpaceX) as a deliberate "counterstatement" to the Anthropic dispute. First explicit policy framing of Procurement-Driven AI Governance as a diversification strategy. Mythos called a "cyber moment."
- ECB President Lagarde studying Mythos defenses (May 8) — first G7 central-bank-level explicit reference to Mythos as a financial-stability risk. Resolves the wiki's "≥1 G7 central bank cites IMF designation within 12 months" prediction in 1 day — the fastest resolution in the wiki's history. Propagation: ECB + Bank of England (Governor's April pre-statement) + Australian ASIC May 8 + Brazil ANPD May 7 sandbox. See Track Record RESOLVED-correct.
- xAI begins renting compute to Anthropic and Cursor (May 10) — Musk's xAI monetizes idle datacenter capacity as Grok lags. Reframes xAI as compute landlord rather than frontier-model competitor.
- Cerebras IPO repriced to $150–160, ~$4.8B raise (May 10) — book oversubscribed 20×. May 13 pricing. Closes the wiki's "Cerebras IPO refile" thread from the May 4 dev-log.
- NVIDIA + SPAN + PulteGroup XFRA (May 10) — places industrial server nodes (16 Blackwell GPUs + 4 EPYC CPUs + 3 TB memory each) in residential backyards to bypass a 2,600 GW utility-interconnect backlog. First residential-datacenter deployment by a major frontier-compute vendor. Bypass-pattern for the AI Data Centers permitting constraint.
- Microsoft-G42 Kenya $1B geothermal datacenter stalled (May 10) — Kenyan government declined to guarantee annual capacity payments. Procurement-vs-deployment friction in emerging-market AI infrastructure.
- TSMC High-NA EUV through 2029 = no (Azhar/Petrovic, May 11) — per-transistor cost stopped falling in 2011; the EUV cost-down curve has now reversed. The "broken bargain of Moore's Law" framing.
- Anthropic $1T-valuation framing (FT, May 8) — round-in-progress framed as approaching $1T post-money. Extends the May 8 $50B-at-$900B announcement. Companies/anthropic Valuation Snapshot updated.
- Brockman cross-examination, Musk case "faltering" (May 10) — Brockman's personal journal as star witness; legal observers told The Information the "stolen charity" narrative is unsubstantiated. Likely RESOLVED-incorrect on the wiki's "If verdict: at least one Musk count finds OpenAI liable" prediction if verdict goes against Musk.
- TCI cuts $8B Microsoft stake 10% → 1% (May 8) — CEO Christopher Hohn cited "AI competitive uncertainty" + OpenAI >40% of Azure backlog. Microsoft -12% YTD.
- Nvidia $40B 2026 AI equity commitments (May 10) — IREN >$2B, Corning >$3B, Marvell $2B, Sept 2024 Intel $5B stake now worth ~$25B. Jensen calls circular-financing critique "ridiculous."
- Tech IT-sector unemployment 3.6% → 3.8% (April) — -13K jobs. Reinforces AI Labor Disruption contested status.
- Stratechery "Inference Shift" (May 11) — Ben Thompson argues LLM compute is bifurcating into "answer inference" (Cerebras/Groq speed, human-in-loop) and "agentic inference" (memory hierarchy outranks raw GPU; cheaper DRAM + CPUs competitive). Predicts agentic inference becomes the larger market because it isn't bounded by human attention.
- EC opens AI Act transparency-guidelines consultation (May 8) — effective August 2, 2026. Feedback due June 3. Article 50-implementing detail.
- METR time-horizon: Mythos at ~16h tasks at 50% success — Gary Marcus pushback (May 10) that 50% is a low bar and gains likely owe more to symbolic tools than pure model scaling.
What it connects to
The dominant thread is the IMF macro-prudential designation propagating faster than the wiki's prediction model anticipated. When the IMF blog landed May 7, the wiki created AI Macro-Prudential Policy with confidence: low and four 12-month predictions. Lagarde's May 8 ECB statement — one day after issuance — resolves the central prediction. Bank of England's April pre-statement + ASIC May 8 + Brazil ANPD May 7 sandbox indicate the propagation is not one central bank but a coordinated international response. This is a category change for Claude Mythos Preview from "safety policy object" to "financial-stability supervisor concern." Macro-prudential governance is now operating in parallel with AI Pre-Release Vetting and Procurement-Driven AI Governance — three governance modes simultaneously in motion.
The "Pentagon never again single-threaded" framing is the most consequential procurement-driven-governance evidence the wiki has seen. Under Secretary Michael's explicit policy framing (vs. just cohort decisions) means procurement-driven governance has moved from inferred-from-evidence to stated-as-policy. The 8-agreement counterstatement structure becomes the operative procurement model. Reflection's open-weight inclusion in the May 1 cohort is now framed as deliberate diversification, not procurement happenstance. The Procurement-Driven AI Governance thesis from Weekly Synthesis — Week 19, 2026 (May 4–10) is strengthening fast.
xAI as compute landlord + Cerebras IPO + NVIDIA XFRA + Microsoft Kenya stall + TSMC High-NA EUV = a coherent datacenter-economics signal. The Anthropic-SpaceX Colossus 1 pattern of "non-hyperscaler frontier-compute commitments" is generalizing: xAI rents to competitors, Cerebras goes public at $4.8B, NVIDIA decentralizes via residential nodes, Microsoft pulls back from emerging-market commitments. The unifying frame is capacity scarcity is real but constraints are shifting from chips to power/permitting. AI Data Centers updated with all five.
One question worth sitting with
The IMF macro-prudential prediction resolved in 1 day. The wiki's other 90-day-window prediction — the two-track Trump pre-release-vetting EO — is now 4 days from its 90-day mark (resolution date 2026-08-05/08). If the procurement-driven-governance evidence keeps strengthening (Pentagon "never again", PE-backed JVs, $1T Anthropic raise specifically for compute), does the broad pre-release-vetting EO become structurally unnecessary — and does the narrower cybersecurity-only EO (Bloomberg May 8) end up being the operative one because procurement is doing the broad EO's work? The wiki should treat the two predictions as coupled rather than independent: a strong procurement-driven outcome makes the narrow-EO prediction more likely.
Predictions touched
- RESOLVED-correct: ≥1 G7 central bank cites IMF Mythos designation within 12 months — Lagarde-ECB May 8, 1 day after issuance.
- Procurement-Driven AI Governance predictions reinforced — Pentagon "never again single-threaded" speech (May 7) materially supports the binding-procurement thesis.
- Musk v. Altman (and OpenAI / Microsoft / Brockman) predictions weakened — "case faltering" framing (May 10) tilts toward RESOLVED-incorrect for the Musk-wins prediction, pending verdict.
- AI Pre-Release Vetting two-track-EO predictions — no resolution yet; the propagation of macro-prudential framing arguably reduces the policy demand for a broad pre-release-vetting EO.
- AI Labor Disruption reinforced (weakly) — IT-sector unemployment 3.6% → 3.8%, -13K jobs (April).
- AI Bubble Debate reinforced (mixed) — Nvidia $40B equity commitments + TCI Microsoft stake cut + broken-Moore's-Law framing all point in different directions; the page's "ACTIVE — dual-readable through end-2026" status is reinforced.