AI agents · OpenClaw · self-hosting · automation

Quick Answer

TeraWulf-Anthropic vs CoreWeave-Meta vs Oracle-Stargate: The AI Compute Deals of July 2026

Published:

Quick Answer

Three massive AI compute deals landed in early July 2026:

  • July 6, 2026 — TeraWulf signs 20-year Anthropic lease at Justified Data campus
  • July 2026 — CoreWeave and Meta sign $21B deal through 2032 (on top of existing $14.2B contract)
  • Ongoing — Oracle-Stargate buildout continues supplying dedicated OpenAI capacity

The pattern: labs are locking in multi-year compute rather than depending on spot capacity. Inference demand from every new frontier model release outpaces available supply, so long-term contracts are the risk-mitigation move for both sides.

Deal 1: TeraWulf ⟷ Anthropic (July 6, 2026)

What was announced: TeraWulf signed a 20-year lease with Anthropic for its Justified Data campus. Same-day announcement: TeraWulf sold majority interest in its Abernathy joint venture to Fluidstack.

Why it matters:

  • 20-year term is unusually long for AI infrastructure. Most AI compute contracts are 3-7 years. 20 years signals Anthropic is planning long-horizon capacity — inference for Claude products decades out
  • TeraWulf’s pivot is complete — the company was Bitcoin-mining-focused; this deal is the anchor commitment for its AI infrastructure identity
  • IPO signal for Anthropic — long-term compute commitments make the S-1 story more credible. “We have decade-plus compute secured” is exactly what late-stage AI IPO investors want to see
  • Fluidstack getting Abernathy consolidates AI-focused datacenter operators; the market is picking winners

Terms not disclosed publicly: exact GW capacity, per-year rent, ramp schedule.

Deal 2: CoreWeave ⟷ Meta ($21B, Through 2032)

What was announced: Meta and CoreWeave signed a $21 billion compute deal in early July 2026, extending capacity supply through 2032. This is on top of a September 2025 $14.2 billion contract. Total committed spend now approaches $35 billion.

Why it matters:

  • CoreWeave is Meta’s largest external compute supplier by a wide margin — bigger than any hyperscaler contract Meta discloses publicly
  • Meta’s AI supply story is now secured through 2032 — Muse Spark, Muse Image, Muse Video, Llama 5, and future models have committed compute
  • CoreWeave’s stock/valuation implications — recurring revenue commitment through 2032 changes CoreWeave’s cost-of-capital story
  • Enables Meta’s paid API strategyMuse Spark 1.1 opened for paid API at $1.25/$4.25 on July 9-10, 2026, which requires reliable inference capacity CoreWeave is providing

Terms: 2026-2032, exact capacity not fully disclosed but analysts estimate 2-4 GW of compute over the term.

Deal 3: Oracle-Stargate (Ongoing)

What’s happening: The Oracle-OpenAI Stargate buildout — the multi-hundred-billion-dollar dedicated OpenAI infrastructure project — continues adding capacity through 2026-2027. Not a single deal announcement, but the largest AI infrastructure buildout in history.

Why it matters:

  • Dedicated OpenAI capacity — Stargate sites aren’t shared with other customers, so OpenAI gets deterministic supply
  • Enables GPT-5.6 Sol via Cerebras plus GPU capacitySol’s 750 tok/s on Cerebras plus mainstream GPT-5.6 volume via Stargate GPUs = mixed inference backend
  • Different scale — Stargate is the strategic long-run bet; TeraWulf-Anthropic and CoreWeave-Meta are 2026-2032 additional capacity

Side-by-Side Comparison

TeraWulf-AnthropicCoreWeave-MetaOracle-Stargate
AnnouncedJuly 6, 2026July 2026 (extending Sep 2025)Ongoing 2024-2027+
Term20 yearsThrough 2032 (~6 years)Multi-decade
Committed spendNot disclosed$21B new + $14.2B existing = ~$35BHundreds of billions
Operator typeRepurposed BTC-miner → AIAI neocloudHyperscaler + OpenAI JV
Buyer stagePre-IPO (October 2026 expected)Public giantPrivate (OpenAI)
SignalAnthropic multi-decade confidenceMeta’s paid-API + Llama 5 supplyOpenAI moat

What This Tells Us About the AI Compute Market (July 2026)

Signal 1: The neocloud model has decisively won for AI compute. CoreWeave, TeraWulf, Nebius, Crusoe, Lambda, Fluidstack — pure-play AI compute providers are getting the biggest contracts. They ship faster than hyperscalers and give labs deterministic dedicated capacity.

Signal 2: Multi-year commitments are now the norm. 3-year contracts are last decade. 6-20 year contracts are 2026-onward. This is both risk-mitigation (labs) and cost-of-capital story (operators).

Signal 3: The gap between demand and supply is growing. Every new model release (Sol, Muse Spark 1.1, Grok 4.5, Gemini 3.5 Pro) drives inference demand faster than new capacity comes online. Multi-year contracts are essentially “book the capacity now before it’s gone” bets.

Signal 4: Pre-IPO compute stories are formal S-1 requirements. Anthropic (October 2026 expected IPO) needed the TeraWulf deal for its S-1. OpenAI (September 2026 filing expected) has Stargate. Meta already has secured supply and used it as a competitive weapon at Muse Spark 1.1 launch.

Signal 5: Watch for the next wave — Nebius, Crusoe, Lambda, and Fluidstack all likely to announce major deals in H2 2026 with Google, xAI, or Anthropic (yes, Anthropic could sign more even after TeraWulf, given the scale of demand).

Who Wins Financially

PlayerWinner?Why
CoreWeaveYes, big$35B committed revenue improves cost of capital and stock story
TeraWulfYes, big20-year revenue anchor validates the AI pivot completely
AnthropicYesLocked-in supply for October IPO narrative + long-run capacity
MetaYesMuse family + Llama 5 have compute security through 2032
OracleYesStargate revenue continues
FluidstackNeutral-positiveAbernathy acquisition adds capacity
NvidiaYesThese deals mostly buy Nvidia GPUs — every new datacenter is Nvidia rev
Nebius, Crusoe, LambdaUnclearNot in this round of announcements; risk of being outbid for future contracts

Sources