Meta has been put on a Gemini diet

TL;DR: Google has reportedly capped Meta’s Gemini use because it can’t provide the compute needed to fuel the demand, according to the Financial Times. It’s a sign of the great compute crunch many companies now face, raising the stakes on the massive data center buildout currently underway—one stymied by logistical issues and, increasingly, political pushback.

What happened: Google delivered the bad news to Meta in March, prompting the latter to tell its staff to use tokens more efficiently, per the FT—a big change of pace from the “tokenmaxxing” culture widely reported earlier this year, when employees collectively burned a baffling 60 trillion tokens in a 30-day period.

The Gemini restriction has reportedly delayed some AI projects Meta was working on, and is the most recent example of how compute bottlenecks are hampering even the biggest, best-resourced firms’ AI use. Due to the squeeze, AI labs themselves have changed how they charge for their products, with both Anthropic and Microsoft moving to pay-as-you-go pricing for some of their models.

The great compute grab: Suddenly, the staggering data center spending estimates we’ve seen from hyperscalers don’t sound quite as crazy. Google already commands a lot more compute than its competitors, per research firm Epoch AI, but it’s pouring in as much as $190 billion on capex this year, much of it likely on building data centers. It also recently signed a deal to pay SpaceX $920 million a month for access to another 110,000 Nvidia GPUs and other compute-related components. Google, Meta, Microsoft, and Amazon appear to be essentially writing blank checks to ensure they can meet AI demand—but the question is how they’ll make do until that new compute comes online.

The big if: Unfortunately for the compute-constrained, over 60% of the data centers slated for completion in 2027 hadn’t started construction as of May, according to a JPMorgan analysis cited by the Wall Street Journal. Supply chain snags, power shortages, and permitting issues are the main culprits—but political opposition is also a growing cause of projects being postponed or canceled.

Bottom line: Google limiting Meta’s Gemini access is a compelling case for why companies might want to limit their reliance on AI. No matter how deep your pockets, or how big your company is, unlimited AI access isn’t guaranteed when there’s not enough compute to go around. —WK