Musk Cautions Against Hype Over Anthropic-SpaceXAI Compute Deal

Benzinga reported Tuesday that Tesla and SpaceX chief executive Elon Musk pushed back on bullish financial projections tied to the Anthropic SpaceXAI compute deal, warning observers not to read too much into short-term contract economics.

The exchange began on social platform X after Ark Invest research director Brett Winton outlined his analysis of the agreement. Winton estimated Anthropic is paying SpaceXAI roughly $24 billion per gigawatt of compute capacity. He placed total infrastructure costs at approximately $29 billion per gigawatt.

A Deal Worth Tens of Billions Over Five Years

Winton argued the arrangement, if sustained over five years, could produce more than $50 billion in cumulative pre-tax cash flow. He added that if the contract covers all of Colossus I, which is primarily built on H100 and H200 chips, alongside portions of Colossus II featuring newer GB200 and GB300 hardware, the gross cash flow figure could climb past $60 billion.

Musk responded briefly and directly. He described the deal as short-term and cautioned against inflated long-run assumptions. His pushback drew attention given that SpaceXAI operates the infrastructure at the center of the debate.

Background: SpaceX IPO Filing Revealed the Deal’s Terms

SpaceX filed its IPO prospectus on May 20 following a confidential submission in early April. The public documents disclosed that Anthropic has committed to paying $1.25 billion per month for compute access through May 2029. The capacity spans two AI training clusters, Colossus and Colossus II. Both parties retain the right to exit the arrangement with 90 days’ notice, and the agreement includes reduced fees during an initial ramp-up window covering May and June.

Separately, Anthropic has submitted a draft Form S-1 registration statement to the U.S. Securities and Exchange Commission, moving the company closer to its own potential public listing.

Infrastructure vs. Software Margins Debate

Winton acknowledged Musk’s broader point about deal duration. He argued, however, that the current phase of AI development still leans heavily on large-scale physical infrastructure. He noted that AI software businesses will eventually command superior margins compared to infrastructure-as-a-service operations. For now, though, building and leasing compute capacity remains foundational to scaling frontier AI models.

The exchange highlights a widening debate among investors over how to value AI infrastructure commitments. Capital-intensive compute deals generate predictable revenue, but questions remain about long-term pricing power and contract renewal risk.

Read Next: SpaceX Moves Toward IPO With Confidential SEC Filing