TODAY'S HEADLINE: Google told Meta it could not have more AI computing power. Then Google rented 110,000 chips from a rocket company.

Here is a sentence that would have sounded absurd a year ago: one of the richest companies on earth was told to use less AI, by another of the richest companies on earth, because there was not enough computing power to go around. The Financial Times reports that Google capped how much of its Gemini AI that Meta could use, and that Google itself is so short on capacity it is paying SpaceX nearly a billion dollars a month for backup chips. It is the clearest sign yet that the real bottleneck in AI is not ideas, it is raw compute. Here is what happened and why it matters.

SECTION 01 // What actually happened

Around March, Google Cloud told Meta it simply could not supply as much Gemini computing capacity as Meta was asking for. The restriction, still in place, delayed and disrupted some of Meta's internal AI projects. Meta had been leaning on Gemini because, for certain jobs, it outperformed Meta's own Llama models.

In response, Meta told its engineers to conserve "tokens," the units that measure AI usage, and to be more efficient. A few other Google customers were affected too, but Meta was hit hardest because its demand was so large. One of the biggest companies in the world was, in effect, put on an AI diet.

FT via CNBC: https://www.cnbc.com/2026/06/28/google-limits-metas-use-of-its-gemini-ai-models-ft-reports.html

SECTION 02 // The twist

The striking part is that Google, which runs one of the largest pools of AI infrastructure on the planet, is itself running short. To plug the gap, Google agreed to pay SpaceX about 920 million dollars a month for access to roughly 110,000 Nvidia chips, capacity it openly described as a temporary "bridge" to meet demand for its Gemini Enterprise product.

And Google is not alone in leaning on the rocket company. Anthropic, the maker of the Claude chatbot, struck a similar deal to rent capacity from SpaceX. When the companies that build AI are renting emergency computing power from a space firm, you get a sense of how tight things have become.

The SpaceX deal: https://thenextweb.com/news/google-caps-meta-gemini-compute-shortage

SECTION 03 // The bigger number

The scale of unmet demand is staggering. Google Cloud passed 20 billion dollars in quarterly revenue, up 63 percent from a year earlier, yet CEO Sundar Pichai admitted revenue would have been higher if the company could have met demand. Its backlog of signed-but-not-yet-delivered cloud contracts nearly doubled in a single quarter to more than 460 billion dollars.

That backlog is the real tell: customers are signing up for computing power faster than Google can physically build it. Meanwhile Meta is pouring up to 135 billion dollars into AI infrastructure this year, and has begun shifting work to its own new in-house model, Muse Spark, to depend less on rivals.

Backlog detail: https://www.investing.com/news/company-news/google-limits-metas-gemini-ai-access-as-compute-demand-outpaces-supply-4764079

SECTION 04 // Why it matters now

For months, headlines have warned of an AI "bubble." But a bubble usually means too much supply chasing too little demand, like warehouses of goods nobody buys. What is happening here is the opposite: capacity is spoken for before it is even built, and the largest buyers on earth are being turned away. You cannot have a glut of something the market is rationing.

That does not mean AI stocks cannot fall, or that the spending will pay off. But it reframes the story. The constraint right now is physical: chips, data centers, and the electricity to run them. Whoever controls that scarce capacity holds real power, and that is shaping up to be the defining business battle of the AI era.

The bubble debate: https://www.forbes.com/sites/jonmarkman/2026/06/29/google-limits-metas-gemini-usage-over-compute-shortages/

FAQ // Quick answers

Why did Google limit Meta's use of Gemini?

Around March, Google Cloud told Meta it could not supply as much Gemini computing capacity as Meta wanted, because demand exceeded available supply. The cap delayed some of Meta's internal AI projects and pushed Meta to use AI more efficiently.

Why is Google renting chips from SpaceX?

Despite running one of the world's largest AI infrastructure pools, Google is compute-constrained. It agreed to pay SpaceX about 920 million dollars a month for roughly 110,000 Nvidia GPUs as temporary "bridge" capacity to meet surging demand for its Gemini Enterprise product.

What is the $460 billion backlog?

It is the value of cloud contracts Google has signed but not yet delivered, which nearly doubled in a single quarter. It signals that customers are committing to computing power faster than Google can physically build and supply it.

Does this mean the AI boom is a bubble?

Many analysts argue it is the opposite. A bubble typically involves excess supply with weak demand, whereas here capacity is fully spoken for before it is built and even the largest buyers are being turned away. That said, heavy AI spending could still underperform expectations.

How is Meta responding?

Meta told employees to conserve AI usage and has accelerated development of its own internal model, Muse Spark, to reduce reliance on rivals like Google. It is also investing heavily, with up to 135 billion dollars budgeted for AI infrastructure in 2026.

We will keep tracking this and bring you the next chapter as it lands. Stay sharp out there.

This newsletter is for general information only and is not investment advice. Always do your own research before making financial decisions.

TECH DAILY // www.techdailynews.org