Anthropic's Google Bet|Why 220K GPUs From a Rival?

Records and an Unlikely Alliance

Anthropic committed $200 billion to Google Cloud. On the same day, it signed a deal to run its AI models on Elon Musk's supercomputer.

The S&P 500 closed at a new all-time high on Wednesday, with the index up 1.19% and the Nasdaq 100 gaining 1.44% to post its own record. AMD surged 16.4% after reporting Q1 revenue of $10.25 billion, up 38% year-over-year, with data center revenue growing 57%. Super Micro Computer climbed 17% on improved margins. Eight Wall Street firms raised AMD price targets in a single session, with KeyBanc's most bullish call landing at $530. The AI infrastructure trade was dominating every screen on the floor.

At the same time, oil was collapsing. WTI crude fell 6.6% to $95.48 a barrel after President Trump indicated progress toward ending the Iran conflict and reopening the Strait of Hormuz. That oil shock, which had been the single largest macro risk on the tape for weeks, was deflating in real time. Alphabet climbed 2.5% on an analyst upgrade, and Amazon rose on the strength of its AI infrastructure narrative — CEO Andy Jassy told investors that Amazon's custom chip revenue grew 40% sequentially in Q1, and he reaffirmed that Nvidia remains a critical partner. ServiceNow crossed $1 billion in AWS Marketplace transactions. The session read like a coordinated AI earnings confirmation.

But one announcement cut across all of it in a way no earnings release could. Anthropic, the AI safety company backed by Google with a $200 billion cloud commitment, announced it had reached a deal to use the full computing power of SpaceX's Colossus 1 facility in Memphis, Tennessee. That facility houses more than 220,000 Nvidia processors — H100, H200, and GB200 accelerators. Anthropic said it would gain 300 megawatts of new compute capacity within one month. The question that hovered over the rest of the session: how does a company committed to Google Cloud end up running on Elon Musk's data center?

The Compute Constraint Nobody's Cloud Can Solve

The short answer is that Anthropic ran out of capacity. The longer answer changes the entire AI infrastructure picture.

Claude Code, Anthropic's AI coding tool, had become one of the most heavily used developer products in the market. Demand was outpacing what Anthropic could serve, regardless of which cloud contract it held. Google Cloud, despite the $200 billion commitment, could not deliver 220,000 Nvidia GPUs and 300 megawatts of power in one month. Colossus 1 could. So Anthropic went to SpaceX.

The mechanism behind this matters for anyone holding a position in the cloud hyperscalers. Google's relationship with Anthropic is financial and strategic — it owns a stake, it runs Claude models through Vertex AI, and it has a $200 billion cloud spending commitment. But that relationship does not give Anthropic exclusive access to Google's GPU allocation. Google Cloud's capacity is spread across thousands of enterprise customers. When Anthropic needed a dedicated block of 220,000 processors available immediately, the only option that could deliver that timeline was a single-purpose facility already built and running.

Colossus 1 was constructed specifically for xAI, Musk's own AI lab. But xAI does not consume 100% of its capacity around the clock. SpaceX needed a marquee customer ahead of its IPO to demonstrate that Colossus 1 is a commercial asset, not just an internal tool. Anthropic needed compute yesterday. The deal is a capacity arbitrage, not a strategic pivot.

The reversal condition sits right there. Anthropic has also expressed interest in working with SpaceX to develop multiple gigawatts of space-based data centers — one of SpaceX's core IPO narratives. That ambition would convert what looks like a temporary capacity agreement into a long-term alternative infrastructure relationship. If that second phase advances, Google's $200 billion commitment has a structural competitor embedded inside its most important AI model partner.

What Alphabet and Amazon Holders Need to Watch

Alphabet stock gained 2.5% Wednesday and is up sharply since its Q1 earnings beat. The consensus reading is that Google Cloud is an AI winner. The Anthropic deal to Colossus complicates that reading without invalidating it.

Google's Q1 showed cloud revenue growing 28% year-over-year to $12.3 billion. Anthropic's Claude models are distributed through Vertex AI, and over 100,000 customers run Claude on AWS — Amazon's competing claim on the same model. Anthropic is simultaneously the biggest argument for Google Cloud and the biggest argument for AWS. Its capacity decision this week was made independent of both.

The historical parallel is the early cloud era, when Netflix ran on AWS while competing directly against Amazon's Prime Video. Shared infrastructure and commercial rivalry coexisted for years before Amazon eventually built a competing streaming service. The Anthropic-SpaceX deal is earlier in that cycle. Anthropic is not competing with SpaceX. But it is routing capacity around both Google and Amazon, and SpaceX is now a credible compute vendor — something it was not six months ago.

The continuation scenario: Colossus 1 proves to be a temporary relief valve. Anthropic's demand normalizes, Google and AWS absorb it through expanded capacity in Q3 and Q4, and the SpaceX arrangement remains transactional. In that case, Alphabet and Amazon's infrastructure moats are unaffected. The AMD and Nvidia earnings already confirmed that data center capex is accelerating, not contracting.

The breakdown scenario: Anthropic deepens the SpaceX relationship toward gigawatt-scale space-based compute, other frontier AI labs follow the same playbook when hyperscaler capacity lags demand, and a third compute tier — purpose-built, single-tenant, outside the cloud — begins to absorb a meaningful share of AI workloads. That scenario does not require any single lab to abandon Google or AWS. It only requires that the marginal capacity decision repeatedly lands outside the hyperscaler framework.

The verification benchmark is SpaceX's IPO filing. If Colossus 1 appears as a commercial data center asset with named enterprise customers and multi-year contracts — rather than as internal infrastructure — the second scenario is advancing. That filing is expected within the next 12 months. Until then, Alphabet's 2.5% gain Wednesday reflects a market that has not yet priced the possibility that its most important AI partner just validated a competitor's compute story. Whether AMD's 38% revenue growth or Anthropic's routing decision turns out to be the more consequential signal from this session remains the question the next phase of the AI trade will answer.