The AI race is moving from smarter models to whoever can secure the power
The AI bottleneck is increasingly physical: racks, power, cooling and contracts.📷 TECH&SPACE / GPT Image 2.0
- ★Anthropic's growth shows that capacity is becoming as important as model quality.
- ★AI infrastructure depends on chips, power, cooling, networking and supplier contracts.
- ★Rivals can become capacity suppliers because the compute market increasingly resembles energy.
Anthropic’s revenue has skyrocketed from $9 billion to over $30 billion annualized, but the growth has come with a brutal side effect: its infrastructure can’t keep up. The company’s 80x surge in usage—driven by developers spending an average of 20 hours per week on its Claude Code tool—has led to rate limits, outages, and a desperate search for compute power. The solution?
A deal to tap into Elon Musk’s Colossus 1 supercomputer, a move that would have been unthinkable just months ago given Anthropic’s prior efforts to distance itself from Musk’s AI ventures.
The partnership is a rare win for Musk’s xAI, which operates Colossus 1 and is preparing for its own IPO. For Anthropic, the deal provides access to 220,000 Nvidia GPUs and 300 megawatts of capacity—resources critical for maintaining service reliability as it scales. But the arrangement also highlights the broader compute crunch facing AI companies, where demand for training and inference far outstrips available hardware.
The Decoder’s report details how Anthropic’s growth has forced it into unexpected alliances, even with former adversaries.
When a product grows 80-fold, ideology quickly gives way to cables, megawatts and spare accelerator hours.
Growth is not only a model problem; it is a capacity scheduling problem.📷 TECH&SPACE / GPT Image 2.0
The source material also shows that musk’s about-face is the real story here. Just a year ago, Anthropic was positioning itself as a more ethical alternative to Musk’s xAI, with co-founder Dario Amodei publicly criticizing Musk’s approach to AI safety. Now, the two are partners, bound by the cold reality of hardware constraints.
The deal suggests that even the most well-funded AI startups can’t escape the need for external compute, especially as they prepare for public markets. For xAI, Anthropic’s business is a valuable proof point ahead of its IPO, demonstrating that Colossus 1 isn’t just a vanity project but a viable commercial asset.
Yet questions remain about the long-term implications. The terms of the partnership—exclusivity, duration, and cost—are still undisclosed, leaving analysts to speculate about whether this is a stopgap or a strategic shift. If Anthropic’s growth continues at its current pace, even Colossus 1 may not be enough. Meanwhile, the deal has sparked chatter in tech circles about Musk’s motives: Is this a genuine pivot toward collaboration, or a calculated move to control the compute infrastructure underpinning his competitors? The answer may become clearer as xAI’s IPO timeline unfolds.
