When a compute contract is described two different ways in public, the timeline itself becomes part of the story.

On one side is Elon Musk’s statement on X: SpaceX’s arrangement with Anthropic is a 180-day lease, with 90 days’ notice for mutual cancellation after that. On the other is SpaceX’s S-1 filing, which describes a cloud services agreement with Anthropic that runs through May 2029, with the company’s disclosure on page F-62 placing the relationship in a multi-year frame.

Those are not just different phrasings. They imply different operating assumptions about exclusivity, pricing durability, and how much notice each side needs before capacity can be reallocated. For Anthropic, which is tied to Colossus compute access, that difference affects how much of its deployment plan can be treated as committed supply versus conditional access.

Two timelines, one deal

The contradiction matters because compute is not an abstract line item for an AI lab. It is the constraint that shapes training schedules, inference availability, rollout sequencing, and product promises.

Musk’s public description centers on a short-term lease with a possible off-ramp. The S-1’s language, by contrast, reads like a longer contractual container: a three-year cloud-services agreement extending through May 2029, with a 90-day termination provision. That structure does not eliminate flexibility, but it does suggest a longer planning horizon than a simple half-year rental.

The difference is especially relevant in the context of Colossus access. Earlier coverage described xAI’s compute deal with Anthropic as a major arrangement for exclusive use of Colossus capacity. If the working assumption inside the market is a durable multi-year commitment, the pricing and deployment expectations look one way. If the binding reality is a short lease with a relatively fast off-ramp, they look another.

What the terms imply for exclusivity and access

A 180-day lease with mutual 90-day cancellation is a very different instrument from a three-year services agreement, even if both contain a 90-day termination mechanism.

In practical terms, the shorter framing implies:

  • less certainty that capacity stays allocated for the full product cycle,
  • more exposure to repricing if utilization tightens,
  • and a stronger need to keep alternate compute plans warm.

The longer framing implies something closer to a capacity commitment that can support roadmaps, staffing, and model-training schedules with more confidence. Even if the agreement is still terminable, the presence of a May 2029 endpoint changes how much of the supply can be treated as predictable.

That matters for exclusivity too. A service agreement through 2029 can be read as a stronger signal that Anthropic’s access is not purely opportunistic or month-to-month. A lease framed as 180 days, however, leaves more room for reallocation if either side decides the capacity should be used differently.

Why the discrepancy lands now

The timing is important because Anthropic’s Colossus rollout depends on clean assumptions about compute availability. If the deal is effectively short-term, Anthropic has to hedge against interruptions, capacity reduction, or a changed billing posture. If the deal is effectively multi-year, then deployment can proceed with more confidence that the underlying infrastructure will still be there when each phase of the rollout comes due.

The same ambiguity also cuts into SpaceX’s own planning. A public short-term framing signals optionality and preserves the ability to recover capacity quickly if internal needs spike. A filing that describes a multi-year agreement signals something more durable to investors and regulators, even if the contract still contains off-ramps.

That tension is what gives the current dispute real operational weight: it is not only about who said what, but about how the same compute pool is being described to the market versus in regulatory paperwork.

Market signaling and what to watch next

For investors and competitors, the immediate question is not whether the arrangement exists. It clearly does. The question is which timeline should be used when modeling Anthropic’s access to AI-grade compute.

If the market prices the deal as a multi-year commitment, Anthropic’s runway looks steadier and the value of Colossus access rises. If the market concludes that the practical term is closer to Musk’s 180-day framing, then the exclusivity premium is thinner and the risk of reallocation is higher.

That is why this kind of discrepancy tends to invite follow-up disclosure. Any amended filing, later statement, or clarification around the contract term could materially change how readers interpret the deal’s durability.

What to verify next

The next useful checks are straightforward:

  • the final contract language, especially any term, notice, and renewal provisions,
  • whether SpaceX amends or elaborates on the cloud-services agreement in later filings,
  • whether Anthropic or SpaceX make any additional public statement on the May 2029 timeline,
  • and whether the 90-day termination language appears as a true mutual off-ramp or only as a narrow contractual backstop.

For editors and technically minded readers, the key is to separate the existence of the deal from the duration of the commitment.

Operational milestones that will reveal the real commitment level

The clearest indicators will not come from rhetoric. They will come from execution.

Watch for changes in:

  • capacity provisioning for Colossus access,
  • billing cadence and whether pricing tracks a long-horizon services relationship,
  • how often termination language is referenced in later disclosures,
  • and whether Anthropic’s public roadmap continues to assume stable access on a multi-quarter basis.

If those operational markers line up with the S-1’s May 2029 framing, the deal behaves like a longer commitment regardless of Musk’s short-term description. If they begin to resemble a lease with an active off-ramp, then the 180-day account becomes more than a rhetorical hedge.

For now, the contradiction itself is the signal: one public narrative emphasizes flexibility, while the filing emphasizes duration. In AI compute, that gap can change how companies schedule models, secure capacity, and price risk.