Grok’s Usage Is so Low that Elon Musk Can Sell Compute to Anthropic

Why It Matters

The agreement turns surplus GPU capacity into a revenue stream for SpaceX and gives Anthropic a high‑performance inference platform, sharpening its competitive edge in the crowded AI assistant market.

Key Takeaways

• Anthropic gains access to 300 MW compute, ~220k NVIDIA GPUs.
• xAI’s Grok models have low utilization, freeing capacity.
• Deal shifts Anthropic’s inference workload to SpaceX’s Colossus 1.
• Musk endorses Anthropic’s safety focus, boosting credibility.
• Colossus 2 now handles xAI training, improving overall efficiency.

Pulse Analysis

The AI industry is currently in a race for raw compute, with most firms scrambling to secure GPU farms for model training. In a surprising twist, xAI’s flagship Grok models have been under‑utilized, leaving a massive amount of hardware idle. SpaceX, which owns xAI, operates two colossal data centers—Colossus 1 in Memphis and the newer Colossus 2. By moving Grok’s training workloads to Colossus 2, SpaceX freed the entire capacity of Colossus 1, estimated at over 300 megawatts and roughly 220,000 NVIDIA GPUs, for external use.

Anthropic, the creator of the Claude family of chat assistants, signed a deal to lease that idle capacity for inference serving. The agreement gives Claude Pro subscribers ($20/month) and Claude Max users ($100‑$200/month) a dedicated pipeline capable of handling real‑time prompts at scale. Because inference is less GPU‑intensive than training, Anthropic can run thousands of concurrent sessions without the latency spikes that plague smaller providers. The partnership also aligns with Anthropic’s strategy to keep its models on high‑performance hardware while controlling operating costs.

The transaction signals a broader shift in how AI firms will monetize excess compute. Elon Musk’s public endorsement of Anthropic’s safety agenda adds a layer of credibility that could attract enterprise customers wary of unchecked AI. For SpaceX, leasing idle GPUs turns a fixed cost into revenue, while Anthropic gains a competitive edge in latency‑sensitive applications. Observers may see more cross‑industry compute swaps, especially as cloud providers face capacity constraints and startups look for cost‑effective, high‑throughput alternatives.