The desperate hunt for AI computing power is upending Silicon Valley

The intense pursuit for AI computing power is reshaping Silicon Valley, with the industry scrambling to secure the necessary resources for advanced AI projects. Google's acquisition of pioneering AI researcher Noam Shazeer for $2.7 billion in 2024 demonstrated the high stakes, but even this couldn't guarantee Shazeer the computing resources he desired. He later left Google for rival OpenAI due to cutbacks on his team's allocation of resources.

Compute, the term for computing power, is crucial for the AI boom and is currently scarce. This has led to competitive battles among rivals and the formation of new alliances, such as between Elon Musk and Dario Amodei. Venture-capital firms are leveraging access to AI chips to attract startups, while the race for computing power has intensified globally.

Companies like San Francisco Compute, which vets and rents out data center computing power, are thriving as AI companies scramble to access the necessary resources. Their chief technology officer, Eric Park, has traveled the world to source computing power, including unconventional locations like rural America and Manila during monsoon season.

Despite the hype surrounding announcements of new computing power, the actual availability often falls short. This discrepancy has led to startups signing deals with less reliable providers, compromising on security and infrastructure. The unpredictability of securing computing power has prompted some companies to overestimate their needs, leading to inflated contracts and higher costs.

Among the most notable developments is Anthropic's agreement with Elon Musk's SpaceX to lease over 300 megawatts of computing power, worth up to $45 billion. This significant deal comes at a time when Anthropic's success with Claude Code has created a computing crunch, leading to frequent outages and increased tensions within the company.

Co-founder Tom Brown even visited Musk's AI company xAI to secure computing resources for Anthropic. The intense competition for computing resources has also drawn top AI executives into the negotiations, with Anthropic's CEO, Dario Amodei, speaking with Meta's chief AI officer, Alexandr Wang, to secure additional capacity.