Nandini Roy Choudhary, writer
By TechSun News Desk | techsunnews.com | June 30, 2026 | Tech / AI | 4 min read 🚨
Meta wanted more AI power. Google said no.
That’s the short version of a story that broke this weekend — and it says a lot about where the AI industry actually stands right now, underneath all the hype.
According to the Financial Times, Google told Meta back in March that it simply could not supply the full amount of Gemini AI computing capacity Meta had requested. The shortfall has been quietly disrupting Meta’s internal AI projects for months. This weekend, it became public.
What Actually Happened 
Meta was leaning hard on Google’s Gemini models — not its own. Internally, Meta used Gemini for content moderation, scam detection, advertiser chatbots, customer service, and even coding tasks. Gemini reportedly performed better than Meta’s own open-source Llama models for these jobs.
Then Google ran out of room.
Google informed Meta it couldn’t fulfill the full capacity Meta wanted to purchase. The restrictions disrupted and delayed multiple internal Meta projects, forcing the company to tell employees to use AI tokens more efficiently. Several other Google customers were affected too — just not as badly as Meta.
Both companies declined to comment when reporters asked.
Why Google Couldn’t Deliver
This isn’t about Google playing favorites. It’s simpler than that — Google doesn’t have enough compute to go around.
Google Cloud pulled in over $20 billion in quarterly revenue, up 63% year-on-year. But the company is sitting on a backlog of nearly $460 billion in unmet demand. CEO Sundar Pichai admitted as much during earnings: Cloud revenue would have been even higher if Google could actually meet demand. (Source: Business Standard)
To plug the gap, Google signed a deal worth roughly $920 million a month to lease additional GPU capacity from Elon Musk’s SpaceX — access to 110,000 Nvidia GPUs, described internally as “bridge capacity.” Anthropic has reportedly struck a similar arrangement with SpaceX.
Translation: even the companies building the AI infrastructure are now renting compute from a rocket company to keep up.
How Meta Is Responding
Meta isn’t sitting still. The company has been cutting jobs and redirecting spending toward AI for months — 8,000 roles eliminated in May, with 7,000 employees reassigned to AI-focused teams under its Superintelligence Labs division.
Now Meta is accelerating development of its own model, Muse Spark, and shifting workloads away from Gemini wherever it can. The company has also pledged up to $600 billion in US infrastructure investment through 2028 — partly to reduce exactly this kind of dependence on a competitor.
There’s irony here. Meta doesn’t operate its own cloud business, unlike Google, Microsoft, or Amazon. That makes Meta structurally dependent on renting compute from rivals — and this episode is a clear example of what that dependence can cost.
This Is Bigger Than Google and Meta
Compute scarcity isn’t a one-company problem right now. It’s industry-wide.
We covered this exact dynamic in our piece on the global memory chip shortage hitting Apple and Microsoft — chipmakers have redirected nearly all their production toward AI infrastructure, squeezing supply everywhere else. The Google-Meta situation is the same root cause, just one layer up the stack: it’s not chips running out, it’s the data centers and GPUs built from those chips running out.
Even Apple is exposed to this. The company partners with Google to use Gemini models for its next-generation Siri — meaning Apple is also competing for a slice of Google’s strained compute capacity.
| 🟡 EDITOR’S OBSERVATION
What’s notable here isn’t that Google ran short on capacity — every cloud provider is. It’s that a company as large as Meta, spending $135 billion a year on AI infrastructure, still couldn’t buy its way out of the shortage. If Meta can’t out-spend this problem, smaller AI startups renting the same compute are in a far tougher spot. |
💬 WE WANT TO HEAR FROM YOU
| Do you think AI compute shortages will slow down AI progress in 2026?
A) Yes — infrastructure can’t keep up with demand B) No — companies will solve it with money and new deals C) It’ll slow smaller players, but not Big Tech Drop your take in the comments — we read every one. |
❓ FREQUENTLY ASKED QUESTIONS
| Q: Why did Google restrict Meta’s access to Gemini AI?
Google simply doesn’t have enough computing capacity to meet everyone’s demand right now. Around March 2026, Google told Meta it couldn’t fulfill the full amount of Gemini compute Meta had requested. It’s not a pricing dispute or a falling-out between the companies — it’s a straightforward infrastructure bottleneck affecting multiple Google customers, with Meta hit hardest due to its unusually high usage. |
| Q: What is Meta doing instead of relying on Gemini?
Meta is fast-tracking its own internal model, Muse Spark, built under its Superintelligence Labs division, and shifting workloads away from Gemini wherever possible. The company has also pledged up to $600 billion in US infrastructure investment through 2028 to reduce its dependence on outside AI providers like Google. |
| Q: Does this mean the AI industry is running out of compute everywhere?
Not entirely, but the strain is real and widespread. Google itself is leasing GPU capacity from SpaceX to cover its own shortfall, and Anthropic has reportedly done the same. Combine that with the ongoing memory chip shortage affecting consumer devices, and the picture is clear: AI demand is currently outpacing the physical infrastructure needed to support it, across multiple layers of the supply chain. |
Disclaimer: This article is based on reporting from the Financial Times, CNBC, Business Standard, Cryptopolitan, AI Magazine, and Data Center Dynamics. All figures reflect information available as of June 30, 2026.
