A major drama is unfolding in the artificial intelligence market, driven by physical infrastructure limitations. Google has been forced to make a tough decision and artificially limit Meta Platforms' access to its most powerful Gemini language model family. The cause is a critical shortage of computing resources: the pace of data center construction and chip procurement cannot keep up with the explosive growth in demand.

Crisis at Mark Zuckerberg's Headquarters

For Meta Platforms, which does not yet have its own extensive cloud business, this has been a serious blow. For a long time, the company relied on purchasing computing limits from Google to support its internal business processes. However, a warning about the imminent exhaustion of capacity was issued back in March, and now the consequences are being felt.

The situation has forced Meta's leadership to take radical measures:

  • Slowing down development: Deployment schedules for corporate tools have been disrupted, and projects are stalling due to a lack of resources.
  • Internal cost-cutting: An order was issued for employees to significantly optimize the use of AI tokens to reduce the load.
  • Strategy shift: To reduce dependence on external platforms, Meta has accelerated the migration of internal processes to its own new Muse Spark model, developed by the Superintelligence Labs division.

Previously, Meta chose Gemini and models from Anthropic because they demonstrated higher efficiency compared to its own open-source solutions and the Llama lineup. Now, amid the shortage, the company is forced to restructure.

An Unexpected Ally: Google and SpaceX

The problem of server capacity shortage is systemic and even affects the developer of Gemini itself. To ensure the stable operation of its premium Gemini Enterprise platform for large businesses, Google has signed a large-scale tactical agreement with Elon Musk's company, SpaceX.

The contract details are impressive: Google has committed to paying SpaceX $920 million monthly for access to xAI data center infrastructure, including the Colossus 1 supercomputer in Memphis. The agreement is set to run until mid-2029 and guarantees Google access to approximately 110,000 high-performance Nvidia GPUs, CPUs, and memory systems.

AI Economics: Who Pays for the Race?

Despite billions of dollars being poured into "hardware," the financial model of artificial intelligence developers remains extremely unstable. Analysts note that even giants like OpenAI have not yet reached a level of net profit. Current revenues from subscription and API sales cover only a small fraction of the costs of maintaining and renting servers.

In this arms race, the main beneficiaries turn out to be the direct suppliers of infrastructure and equipment. Due to critical network overload, the market has seen a sharp rise in token prices. This is forcing many companies to review their budgets and reduce the use of AI in their products, returning to a more rational approach.