Anthropic Q2 revenue set to double to $10.9bn as $45bn SpaceX deal redraws AI compute map

Anthropic is on track to post its first quarterly operating profit, with the artificial intelligence developer behind the Claude family of models projecting June quarter revenue of at least $10.9 billion against expected operating income of roughly $559 million. The disclosure, contained in fundraising materials and corroborated by a separate SpaceX initial public offering filing on Wednesday, also revealed that Anthropic has committed to paying SpaceX $1.25 billion every month through May 2029 for access to compute capacity drawn from the Colossus 1 data center cluster in Memphis, Tennessee. The combined arrangement could be worth more than $45 billion to SpaceX over the contract term, making it one of the largest single infrastructure commitments ever disclosed by a private AI developer. For Anthropic, the deal locks in roughly 300 megawatts of capacity and access to approximately 220,000 graphics processing units at a moment when scarcity of frontier-grade compute is the single biggest constraint on revenue growth in the sector. For SpaceX, it converts what was an internal cost centre tied to its February acquisition of xAI into a contracted, multi-year cash stream that materially reshapes the narrative ahead of its own blockbuster public listing.

How is Anthropic translating Claude demand into the AI sector’s first credible path to operating profitability?

The headline financial figure is the doubling of quarterly revenue. Anthropic told prospective investors that June quarter sales could reach $10.9 billion, up from $4.8 billion in the March quarter, with operating profit projected at $559 million. That trajectory is consistent with earlier reporting that the company’s annualised revenue run rate had already crossed $30 billion by April 2026, compared with roughly $9 billion at the end of 2025. The expansion has been driven primarily by enterprise adoption of Claude for software engineering workloads, with the number of customers spending more than $1 million annually having doubled within a two-month window earlier this year.

What makes the projected profit significant is the rarity of the achievement. Across the AI industry, the dominant financial pattern has been escalating losses tied to model training compute, inference compute, and talent costs. OpenAI, Anthropic’s closest commercial peer, has continued to operate at a substantial loss despite comparable revenue scale. The structural difference appears to lie in Anthropic’s enterprise mix. Coding and developer-tool workloads carry higher contracted margins than consumer subscriptions, and the company’s positioning around Claude Code and agentic coding has captured a disproportionate share of that segment. If the June quarter print lands within the projected range, Anthropic would become the first pure-play frontier AI developer to demonstrate operating-level profitability without the cross-subsidy of a hyperscaler parent.

Execution risk remains material. The $559 million operating profit figure is internal, not audited, and assumes continued enterprise expansion through the quarter. Anthropic is also simultaneously absorbing the front-loaded costs of the SpaceX compute agreement, which carries a discounted fee structure during the May and June capacity ramp before stepping up to the full $1.25 billion monthly rate. Margin durability over the following four quarters, once full compute costs flow through the profit and loss statement, is the more meaningful test.

Why is Anthropic paying SpaceX $1.25 billion every month and what does the Colossus 1 contract actually buy?

The compute agreement is structurally novel. SpaceX is acting as what the industry now describes as a neocloud, monetising spare capacity at the Colossus 1 facility in Memphis that was originally built to support xAI’s own model training. Following SpaceX’s acquisition of xAI in February, that capacity sits inside the consolidated entity, and Anthropic has contracted to take essentially all of it. The deal grants Anthropic approximately 300 megawatts of compute capacity and access to roughly 220,000 graphics processing units, with the contract running through May 2029.

The strategic logic from Anthropic’s side is straightforward. Frontier model training and inference at the scale Claude now operates is bottlenecked by power, data centre footprint, and GPU availability rather than by capital. By securing a multi-year tranche of high-density capacity outside the conventional hyperscaler stack of Amazon Web Services, Microsoft Azure, and Google Cloud, Anthropic reduces its dependency on any single counterparty and gains contractual certainty over a critical input. The arrangement complements rather than replaces Anthropic’s earlier multi-year compute commitment with Google and Broadcom for next-generation tensor processing unit capacity, which is scheduled to come online from 2027 and provides roughly five gigawatts over its lifetime.

The termination clause is the most commercially interesting provision. Either Anthropic or SpaceX can walk away from the contract on 90 days’ notice. That flexibility cuts both ways. For Anthropic, it preserves optionality if model architectures shift, if internal silicon plans accelerate, or if hyperscaler pricing improves. For SpaceX, it allows the company to redirect capacity to higher-paying buyers or back to xAI’s own use if demand patterns change. In practice, a 90-day exit on a $1.25 billion monthly commitment is a meaningful constraint on either party walking, but it removes the lock-in normally associated with infrastructure of this scale.

What does the deal mean for SpaceX’s IPO narrative and the economics of its loss-making AI segment?

SpaceX disclosed the Anthropic contract inside the prospectus it filed Wednesday with the Securities and Exchange Commission, ahead of a planned Nasdaq listing under the proposed ticker SPCX. The filing reveals that the company’s AI segment, which now consolidates xAI, lost approximately $2.5 billion from operations in the March quarter on segment revenue of just $818 million. Of the company’s $7.7 billion in total AI-related spending, the bulk reflected the xAI acquisition and the build-out of the Colossus and Colossus II clusters in Memphis.

The Anthropic contract transforms the optics of that segment. At $1.25 billion monthly, the deal alone is worth $15 billion annualised. Even after deducting the operating cost of running Colossus 1 at full utilisation, the margin contribution should be sufficient to move SpaceX’s AI segment from heavy operating losses to a credible path toward break-even within the current fiscal year. The IPO filing explicitly notes that SpaceX expects to enter into similar service contracts in the future, and Elon Musk has publicly indicated that the company is in discussions with other AI developers about offering compute as a service at significant scale.

That positioning matters for the public market reception. SpaceX is reportedly targeting a valuation above $1.75 trillion at listing, which would make the offering the largest IPO in history. A loss-making AI segment burning $2.5 billion a quarter is a difficult line item to defend at that valuation. A contracted AI infrastructure segment with $15 billion in annual revenue from a creditworthy single counterparty, with optionality to add further customers, reframes the same assets as a high-growth cloud business. The narrative shift is substantial, even though the underlying physical infrastructure has not changed.

How does the Anthropic-SpaceX arrangement reshape the competitive landscape for hyperscaler compute providers?

The deal represents the most visible defection yet of frontier AI training workloads away from the established hyperscaler stack. Amazon Web Services has been Anthropic’s primary cloud partner for inference, and Google Cloud has supplied tensor processing unit capacity under the Broadcom-anchored partnership. Microsoft’s relationship with OpenAI remains the largest single hyperscaler-AI tie-up in the industry. The Colossus 1 contract opens a third procurement channel for Anthropic that bypasses all three traditional providers entirely.

For Amazon, Microsoft, and Google, the immediate revenue exposure is limited because the Anthropic-SpaceX capacity is incremental rather than substitutive. The longer-term concern is competitive. If neocloud arrangements at the gigawatt scale prove commercially workable, the implicit pricing power of the hyperscalers over frontier AI customers diminishes. Specialty providers including CoreWeave, Nebius, and now SpaceX through its AI infrastructure platform are demonstrating that purpose-built clusters can be stood up and contracted at a scale previously thought to require a hyperscaler balance sheet. The cost-of-capital arbitrage that historically favoured Amazon, Microsoft, and Google is narrowing as private credit and infrastructure debt markets become comfortable financing dedicated AI capacity.

There is also a strategic complication around xAI. SpaceX is now simultaneously the parent of one frontier AI lab and the primary external compute supplier to a direct competitor. The IPO filing acknowledges this structural tension by emphasising the 90-day termination flexibility and the firewall arrangements around Colossus 1 usage. Anthropic, for its part, is comfortable enough with the arrangement to commit $45 billion of contracted spend to it, which suggests either strong contractual protections or a calculation that the compute advantage outweighs the competitive proximity. Either reading is informative about how the industry is now thinking about infrastructure access.

What execution and regulatory risks could derail the Anthropic-SpaceX compute architecture before May 2029?

The first risk is operational. Colossus 1 is a relatively young facility, and ramping it to deliver 300 megawatts of consistent, high-utilisation compute for a single demanding customer is not a trivial undertaking. Any sustained outage, power constraint, or hardware reliability issue would directly impact Claude availability and Anthropic’s enterprise service-level commitments. The fact that Anthropic is taking essentially all of the available capacity amplifies the concentration risk on the supplier side.

The second risk is regulatory. The Federal Trade Commission and the Department of Justice have shown sustained interest in concentration within the AI compute supply chain. A long-term, exclusive-capacity arrangement between two of the largest AI infrastructure players in the country, structured around a recently acquired competitor lab, is the kind of transaction antitrust authorities may want to examine, particularly if SpaceX adds further AI customers and approaches a position of effective market power in the neocloud segment. The 90-day termination clause provides some defence against an exclusivity argument, but the practical reality of a $15 billion annual contract is closer to exclusivity than the legal structure suggests.

The third risk is geopolitical and energy-related. Memphis sits inside a Tennessee Valley Authority service area that is already absorbing rapid load growth from data centre expansion. Sustained operation of Colossus 1 and Colossus II at full capacity will require continued cooperation from local utilities and regulators, and any constraints on power delivery would cascade directly into Anthropic’s available compute. Both companies have publicly expressed interest in jointly developing multi-gigawatt orbital data centre capacity, which is a long-dated and speculative element of the relationship rather than a near-term mitigant.

Key takeaways on what the Anthropic-SpaceX compute deal means for AI economics, hyperscalers, and the road to IPO

• Anthropic’s projected June quarter revenue of $10.9 billion against an expected operating profit of $559 million would make it the first frontier AI developer to demonstrate pure-play operating profitability without hyperscaler cross-subsidy, validating the enterprise coding workload as the most monetisable segment of generative AI to date.
• The $1.25 billion monthly compute commitment to SpaceX, running through May 2029 and potentially exceeding $45 billion in aggregate, locks in roughly 300 megawatts of capacity and 220,000 graphics processing units, securing one of the largest contracted AI infrastructure footprints in the industry.
• For SpaceX, the contract converts a loss-making AI segment that bled $2.5 billion on $818 million of revenue last quarter into a $15 billion annualised neocloud business, materially reshaping the prospectus narrative ahead of a Nasdaq listing under the proposed SPCX ticker at a targeted valuation above $1.75 trillion.
• The 90-day mutual termination clause is unusual at this contract size and signals that both parties retained meaningful optionality on what is otherwise a multi-year capital commitment, reflecting continued uncertainty about model architecture, GPU generation cycles, and pricing trajectories.
• The arrangement is structurally a defection from the traditional hyperscaler procurement stack of Amazon Web Services, Microsoft Azure, and Google Cloud, and it validates the neocloud business model at frontier scale, with knock-on implications for the pricing power of incumbent cloud providers over their AI customer cohorts.
• SpaceX’s role as both the parent of xAI and the primary external compute supplier to a direct competitor in Anthropic creates a structural tension that the IPO filing addresses through termination flexibility but does not fully resolve, and it will likely attract regulatory scrutiny as the relationship matures.
• The deal complements rather than replaces Anthropic’s earlier multi-year compute partnership with Google and Broadcom for next-generation tensor processing unit capacity from 2027, indicating a deliberate multi-supplier strategy designed to avoid single-counterparty dependency at the infrastructure layer.
• Power availability in the Tennessee Valley Authority service area, Memphis-specific utility cooperation, and the operational reliability of the Colossus 1 facility constitute the most immediate execution risks to the contracted capacity ramp through May and June 2026.
• The disclosure of the Anthropic-SpaceX contract inside the SpaceX prospectus rather than as a standalone announcement increases legal certainty around the commercial terms but also exposes Anthropic’s compute cost structure to public scrutiny in a way that may inform competing bids and counterparty negotiations for OpenAI, Meta Platforms, and other frontier developers.
• With Anthropic itself reportedly targeting an initial public offering window later in 2026 at a valuation that market participants have placed near $1 trillion, the demonstration of operating profitability and contracted multi-year compute access materially strengthens the company’s prospectus narrative and positions it as the most financially mature of the major frontier AI labs heading into the public market window.