Anthropic Becomes SpaceX's Largest Customer: 34.5% of Revenue and Contracts Worth $120.5 Billion

If you have ever looked at a cloud GPU bill and thought it was the heaviest line in your budget, this news shows the scale of the problem on a different level. A single AI lab now brings a space company more money than the US government does.

What happened

According to Bloomberg, Anthropic, the developer of Claude, has become SpaceX's largest customer. In the third quarter of 2026, it accounted for 34.5% of the revenue of Elon Musk's company. The US government was previously considered SpaceX's main customer, but Anthropic's payments turned out to be larger.

The commitments are a separate figure. According to the publication, Anthropic agreed to spend up to $120.5 billion on SpaceX computing capacity under contracts that run through 2029. This is a maximum amount, not a guaranteed payment: actual spending depends on the terms of the contracts and on how much capacity Anthropic actually uses.

The article gives no details on exactly what capacity Anthropic is buying or how the contracts are structured. Any conclusions about the technical side therefore remain guesswork for now.

Why this matters to practitioners

A simple analogy: AI startups used to be ordinary cloud tenants, one of many. Now one behaves like the anchor tenant of a shopping mall, around which the landlord's entire business is built. When a single lab provides a third of a supplier's revenue, the supplier starts adjusting its construction plans, procurement and priorities to fit it.

For developers and founders, several practical points follow.

First, compute remains the industry's main constraint. If the market leader is ready to sign commitments worth hundreds of billions of dollars with a horizon through 2029, capacity shortages and high inference prices are unlikely to disappear quickly. Build this into your product's unit economics: the cost per token may fall more slowly than you would like.

Second, unexpected suppliers are appearing on the market. SpaceX is known for rockets and satellites, but here it acts as a seller of computing capacity. It is a reminder that the AI infrastructure market is wider than the classic trio of hyperscalers, and labs have a choice of whom to negotiate with.

Limitations and risks

Keep in mind that all the figures so far rest on a single source: a Bloomberg publication as retold by RB.RU. We do not know how the 34.5% share was calculated, whether it includes all related entities, or how the picture will change in coming quarters. One quarterly figure does not guarantee a lasting trend.

Concentration has a flip side too. For SpaceX, dependence on one customer is a risk: if Anthropic revises its plans, revenue will drop noticeably. For Anthropic itself, commitments of tens of billions of dollars mean rigid fixed costs that must be recouped through Claude sales.

What to do about it

Product teams should check how dependent their product is on the price and availability of a single model or provider. If Claude or any other API is the core of your service, work out a fallback in advance: a provider abstraction layer, monitoring of per-request cost, and a migration plan.

Founders should treat this news as an indicator: money in AI is increasingly flowing not into applications but into the hardware and energy beneath them. That means the winners are those who know how to measure compute efficiency, not those who simply plug in the biggest model.

The bottom line is simple: the AI market is maturing and increasingly resembles heavy industry. Plan your product with the understanding that the cost of compute will set the rules of the game for several more years.