Rent the Frontier
The most capable model ever made public shipped on a Tuesday, and within a day a payments company had used it to rewrite fifty million lines of code that would have taken a team two months. Claude Fable 5 — the first Mythos-class model anyone can rent rather than be vetted for — led every independent benchmark that matters and became the first model past 80 percent on the hardest production-code test, eleven points clear of the field. The same week, the money behind all of this stopped being a question: Oracle booked a $638 billion contracted backlog, SpaceX raised the largest IPO in history without a tremor, and global data-center spending crossed a trillion dollars for the first time. And in the middle of it, the lab that built the model put $200 million behind researching the unemployment it expects to cause — Dario Amodei calling AI-driven displacement "an intrinsic property of the technology" and proposing a tax on companies like his own. P-012 (the bubble bursts by 2026) holds at refuted with two weeks to its resolution date: the cracks from Broadcom's miss are real, but nothing this week looked like a burst.
Key Developments
The most capable model ever made public shipped June 9 — and a day later Stripe had used it to do two months of migration work in an afternoon
Anthropic released Claude Fable 5 on June 9, the first time a Mythos-class model has been available to rent rather than restricted to vetted defenders (GA on API, Bedrock, Vertex and Foundry at $10/$50 per million tokens). It took the #1 spot on the independent Artificial Analysis Intelligence Index, scored 53 percent on Humanity's Last Exam — a new high — and, the number that matters for software, hit 80.3 percent on SWE-bench Pro, the version built from real production GitHub issues, eleven points ahead of the next model and the first ever to clear 70 percent there. Stripe, in early access, reported a fifty-million-line Ruby migration finished in a day, work it had budgeted two months of team time for. The companion Mythos 5 stays locked behind Project Glasswing for approved cyberdefenders, and Fable still hands about 9 percent of tasks back to Opus 4.8. This is the clearest single-release jump in coding capability since Opus 4.8 in May.
One caveat, and it is the named one — Benchmark Theater: the 95 percent SWE-bench Verified figure Anthropic led with is its own reported number on a benchmark that is saturating, so treat it as directionally correct, not independent. The signals that aren't Anthropic's are the ones that hold up — the 80.3 percent on SWE-bench Pro is from an independent tracker (morphllm) and the #1 Intelligence Index ranking is Artificial Analysis's, not the lab's. And capability is not deployment: the same 9 percent Opus-4.8 fallback that makes Fable safe is 9 percent of tasks where the frontier model quietly isn't the one answering.
Oracle booked a $638 billion backlog — the largest contracted demand in enterprise-tech history, and more than half of it is one customer
Oracle's Q4 FY2026 (June 10) put remaining performance obligations at $638 billion, up 363 percent year over year and $85 billion higher than the prior quarter — more than triple the previous record for any software company. Cloud infrastructure revenue grew 93 percent, the Multicloud AI Database 404 percent, and four individual customers each signed for more than $8 billion in the quarter. Roughly $75 billion of the backlog is prepaid or customer-supplied GPU capital, contracted demand no rival can replicate on the same terms. It is the supply-side answer to the bubble question: you do not pre-pay $638 billion for capacity you don't intend to use.
The backlog cuts both ways. More than half of that $638 billion is OpenAI — a company losing about $1.22 for every dollar of revenue — so Oracle's most-durable-demand-in-enterprise-tech is concentrated in the single least-proven business model in it. And the print sent the stock down 11 percent, because behind the backlog is FY27 capex guided near $70 billion net and a planned $40 billion capital raise that dilutes equity against negative free cash flow. The demand is contractual; the open question is whether Oracle can finance the buildout to deliver it before the cash math turns.
The largest IPO in history cleared without a tremor — SpaceX raised $75 billion the same week OpenAI filed to go public and data-center spending crossed a trillion dollars
SpaceX priced at $135 a share on June 11, raised $75 billion at a $1.75 trillion valuation — the biggest IPO ever, more than double Saudi Aramco's — and closed its first day up 19 percent. OpenAI filed a confidential S-1 on June 8 at an $852 billion valuation targeting a September listing, roughly 34 times its $24-25 billion run-rate, and CoreWeave was added to the Nasdaq-100 fifteen months after its own IPO. Dell'Oro put 2026 global data-center capex above $1 trillion for the first time. P-012 — the boom bursts by 2026 — resolves at month-end with none of its five burst criteria met; the market just absorbed the largest offering in history without flinching.
A clean debut is what a top looks like, not proof there isn't one. The same week, a Fortune analysis put S&P 500 long-term earnings-growth forecasts at 20.2 percent — above the 18.6 percent peak in 2000 — with Ray Dalio's bubble indicators near 1929 and 2000 levels and Jamie Dimon calling the mood 1999. OpenAI asking public markets for $852 billion while losing $1.22 per dollar of revenue is the precise trade a correction springs. P-012 refutes because nothing burst by its June 30 date — but the debt-financed channel, GPUs depreciating in a year against 7-to-15-year loans, is a 2027-2028 story this resolution doesn't reach.
What the Evidence Moved
Claude Fable 5 (June 9) ships as the first publicly accessible Mythos-class model — 95 percent SWE-bench Verified (Anthropic-reported) and 80.3 percent on the independently tracked SWE-bench Pro, #1 on Artificial Analysis; Stripe reports a 50M-line migration in a day. At the extreme boundary (0.84); the remaining gap is reliable multi-step orchestration, with a ~9 percent fallback to Opus 4.8.
Fable 5's 80.3 percent on SWE-bench Pro (real production GitHub issues) is the first model above 70 percent there, and the Stripe 50M-line migration-in-a-day demo is the clearest 'engineers orchestrate, AI writes the code' data point yet. Uber separately disclosed 70 percent of committed code is now AI-generated.
Dell'Oro put 2026 global data-center capex above $1 trillion for the first time (June 10), and Oracle's Q4 RPO hit $638 billion (+363 percent YoY), the largest contracted backlog in enterprise tech. Supply and demand both printing at the trajectory's high end.
AI-attributed cuts kept accumulating (Amdocs ~2,900 on an AI pivot, Uber cut 23 percent of its People division, Skillsyncer tallies ~152k AI-attributed 2026 cuts) and Amodei put $200M behind displacement research. Counter-weight: PIIE's Kolko calls the research 'first inning' and BLS revised Q1 productivity down (+0.8 to +0.3 percent) — net small move up.
Uber's 23 percent People-division cut alongside 70 percent AI-generated code is the clearest single-company productivity-to-headcount pipeline yet; Anthropic's own labor paper finds a 14 percent entry-rate decline for ages 22-25 in exposed roles — though it barely reaches statistical significance, which caps the move.
The EU AI Act Digital Omnibus defers Annex III high-risk obligations from August 2026 to December 2027 — removing the main near-term regulatory forcing function that would push enterprises off facial/biometric verification by year-end. With the deadline pushed 16 months, 30-percent-abandonment-by-2026 loses its primary driver.
Final pre-resolution check, two weeks to target_date 2026-06-30. SpaceX absorbed the largest IPO in history (+19 percent), OpenAI filed its S-1, data-center capex crossed $1 trillion — none of the five operational burst criteria met. Resolves refuted at 2026-06-30.
Company Impact
Oracle
Score changeQ4 FY2026 (June 10) — RPO $638B (+363 percent YoY, +$85B QoQ), over half tied to OpenAI, ~$75B prepaid GPU capital; IaaS +93 percent. md 8->9 — the backlog is now a contractual moat no peer can match. Score 7.6->7.8. Bear: ~$40B capital raise and negative FCF.
Oracle Q4 FY2026 results — Investor Relations · Oracle stock -11% on capital raise — CNBC
ServiceNow
Score changeExpanded IBM partnership (June 11) — Workflow Data Fabric + watsonx.data + legacy-Java modernization targets the #1 enterprise-AI blocker; FY26 AI revenue target raised to $1.5B (from $1B). md 7->8. Score 7.2->7.4. Maturity Index is vendor-commissioned (flagged).
IBM and ServiceNow expand collaboration — IBM Newsroom · ServiceNow Enterprise AI Maturity Index 2026
Home Depot
Score changeGemini-powered AI voice agents in production across a 50-store pilot (April 22, Google Cloud co-announcement) — 4x faster call resolution vs phone menus, national rollout planned; a confirmed production deployment beyond the March-2025 Magic Apron suite. md 8->9. Score 7.2->7.4. The 4x figure is co-reported (flagged).
Home Depot + Google Cloud Gemini voice agents — Google Cloud
Anthropic
Data refreshClaude Fable 5 + Mythos 5 launched June 9 — Fable 5 GA at $10/$50 per M tokens, #1 Artificial Analysis, 80.3 percent SWE-bench Pro; $350M Economic Futures commitment (June 10); TCS Global Premier partnership (50,000 associates). Dimensions at/near ceiling (are 10, aam 10); score holds 9.1.
Claude Fable 5 and Mythos 5 — Anthropic · TCS-Anthropic Global Premier partnership
Uber Technologies
Data refreshExhausted its full-year 2026 AI-coding budget in four months and capped engineers at $1,500/month (June 3); 70 percent of committed code now AI-generated; cut 23 percent of its People division. The COO says he cannot draw a clean line from AI usage to consumer outcomes. Dimensions saturated (aam 9); score holds 6.7 (devil's-advocate hold — a heavy-adoption story, not a clean disruption-risk move).
Uber caps AI spending after blowing through budget — TechCrunch · Uber cuts 23% of People division — CNBC
Salesforce
Data refreshSummer '26 brings multi-agent orchestration to GA on June 15 (forward-looking as of this run) with Gemini 3.5 Flash native in Agentforce; Q1 FY27 Agentforce ARR $1.2B (+205 percent YoY). Materiality re-check opens at Q2 FY27; score holds 7.0.
Salesforce Summer 2026 release · Salesforce Q1 FY27 earnings
Sources
- Claude Fable 5 and Mythos 5 — Anthropic
- Claude Fable 5 is a version of Mythos the public can access — TechCrunch
- SWE-bench Pro leaderboard — morphllm
- Oracle Q4 FY2026 results — Investor Relations
- Oracle stock -11% on capital raise — CNBC
- SpaceX IPO live updates — CNBC
- OpenAI files confidential S-1 — Fortune
- Dell'Oro: 2026 data-center capex above $1 trillion
- Anthropic proposes taxing itself for AI job loss — Fortune
- Uber caps AI spending after blowing through budget — TechCrunch
- KPMG + Microsoft scale enterprise AI agents globally
- TCS + Anthropic Global Premier partnership — TCS
- IBM and ServiceNow expand collaboration — IBM Newsroom
- NEURA Robotics $1.4B Series C — The Next Web
- China $295B AI buildout plan — Bloomberg
- EU AI Act Omnibus defers high-risk deadlines — Gibson Dunn
- Home Depot + Google Cloud Gemini voice agents
Next issue drops Monday
Subscribe to get the briefing before the market opens.