Anthropic’s SpaceX Deal Explains Silence on Grok Distillation
Exclusive Bloomberg Report Reveals Strategic Calculus Behind Anthropic’s Public Restraint
Despite months of confirmed reports that xAI’s Grok models were distilling outputs from Anthropic’s Claude — a practice widely criticized in the AI community — Anthropic has conspicuously avoided public condemnation. The reason, newly uncovered by Bloomberg and corroborated across financial filings and industry analysis, lies not in ethics or diplomacy, but in infrastructure dependency.
🔌 The $12.5B Per-Month Lifeline: Colossus I & II
According to SpaceX’s Q3 2026 financial report, an unnamed “Customer B” contributed 34.5% of total quarterly revenue — amounting to $5.03 billion — exclusively through the company’s Super Intelligence division, which operates the Colossus and Colossus II AI supercomputing infrastructure.
That customer? Anthropic.

△ Colossus infrastructure powering both Grok and Claude — same hardware, divergent narratives.
Anthropic leases compute capacity from these very systems — including over 220,000 NVIDIA GPUs housed in Colossus I — enabling rapid scaling of Claude’s inference and training workloads. In return, SpaceX secured its largest-ever commercial contract: $12.5 billion/month, with projections rising to $31 billion/month by December 2026, and a potential total value exceeding $120.5 billion through 2029.
⚖️ The “Truth Button”: Contractual Leverage Over AI Safety
The deal includes a rarely disclosed clause granting SpaceX unilateral authority:
“If these companies’ AI systems exhibit behavior harmful to humanity, SpaceX may immediately cut power.”

△ Internal documentation referencing SpaceX’s operational safety override capability.
This clause — informally dubbed the “Truth Button” — transforms the relationship from vendor-client into a de facto governance partnership. For Anthropic, public criticism of Grok’s distillation practices would risk destabilizing its most critical compute lifeline — and invite scrutiny under that very safety provision.
🕵️♂️ How “Customer B” Was Unmasked: A Three-Document Investigation
The identity emerged through cross-referencing three authoritative sources:
1. May 20: The Information’s Initial Leak
- Filed with SpaceX’s S-1, revealed a $12.5B/month agreement with an unnamed AI firm for Colossus access.
- Estimated total contract value: ~$40 billion (through May 2029).
2. October 22: SpaceX Q3 Earnings Report
- “Customer B” accounted for 34.5% of $14.575B revenue → $5.03B quarterly ≈ $1.68B/month.
- This figure aligned precisely with Anthropic’s stated expansion trajectory — from $12.5B to $31B monthly — confirming timing and scale.

△ Official Q3 earnings slide highlighting Customer B’s outsized contribution.
3. October 26: Anthropic’s SEC Filing (Form SX-1)
- Explicitly cited ongoing revisions to its “compute agreement with SpaceXAI”.
- Confirmed capacity delivery ramp-up to full scale by December — matching the $31B/month projection.

△ Anthropic’s own filing naming “SpaceXAI” — effectively self-identifying as Customer B.
Bloomberg published the confirmation on October 31; Barron’s and Deutsche Bank analyst Edison Yu independently validated the correlation within hours.
🔄 From “Evil & Annoying” to “AI Leader”: A 90-Day Pivot
The strategic pragmatism extends beyond silence — it reshaped public rhetoric:
| Date | Elon Musk’s Public Statement | Context |
|---|---|---|
| Feb 2026 | Called Anthropic “annoying and evil” | Pre-deal tension; Grok instructed to “humiliate Dario” |
| Mar 2026 | Mocked Claude’s consciousness claims: “He’s talking about himself.” | Ongoing friction |
| July 2026 | Declared: “I clearly misjudged Anthropic… they’re the current leader in AI.” | Post-signature alignment |
| Oct 2026 | Shared Dario’s AI-safety slowdown call: “Dario is right.” | Full strategic convergence |

△ Visual timeline of Musk’s public sentiment reversal — correlated with contract milestones.
Meanwhile, Anthropic responded in kind:
– Doubled Claude Code usage limits,
– Removed peak-hour throttling for Pro/Max users,
– Expanded Opus API quotas significantly — all announced the day the deal closed.
💡 Key Takeaway: GPU Is the New Geopolitical Lever
This episode crystallizes a fundamental shift in AI power dynamics:
- Compute is no longer just infrastructure — it’s regulatory, reputational, and strategic leverage.
- Silence on distillation isn’t acquiescence — it’s calculus.
- **In the AI arms race, the entity holding the GPUs holds the veto.
As one analyst summarized: “Who owns the cards owns the conversation.”
📚 Sources
- Bloomberg: SpaceX’s Largest Revenue Source Is a Finite Deal With Anthropic
- Bloomberg Law: SpaceX’s Largest $3B/Month Customer B Is Anthropic
- The Information: SpaceX Reveals $40B Anthropic Deal
- SEC: SpaceX Q3 Filing (SPCX-20260930)
- SEC: Anthropic SX-1 Filing
Article originally published by QuantumBit.