Meta is in such a hurry to build AI compute that it is putting chips in tents. The company has earmarked an extraordinary $115–135 billion in AI capital spending for 2026 — nearly double last year — and is deploying unconventional, fast-to-build data centers to close the compute gap with OpenAI and Google. The scramble for processing power has reached a fever pitch.
The staggering spend
The numbers are almost hard to fathom. Meta’s planned AI capital expenditure of $115–135 billion for 2026 represents a near-doubling of the prior year’s outlay, an aggressive bet that compute is the decisive resource in the AI race. The spending funds chips, data centers and the power to run them — the physical foundation on which frontier AI is built.
Data centers in tents
Speed is everything, so Meta is improvising. The company is erecting massive tent-like structures across US campuses, filled with billions of dollars of AI chips, to bring capacity online in months rather than years. Several of these canvas data centers have already gone up, a vivid symbol of how desperate Big Tech is for compute — willing to abandon traditional construction to deploy faster.
Building its own silicon
Meta is also attacking the chip layer. In partnership with Arm, the company is developing a new class of data-center chips tailored to its AI workloads, part of a broader industry push to reduce dependence on Nvidia and control costs. Designing custom silicon lets Meta optimize performance and economics for its specific models — a long game that complements the frantic near-term buildout.
The race to catch up
The spending reflects competitive anxiety. Meta has been working to close the gap with OpenAI and Google, and recently unveiled Muse Spark, its first flagship model under Alexandr Wang’s new Superintelligence Labs, touting strong performance at lower compute cost. But matching rivals at the frontier requires staggering infrastructure — hence the willingness to spend over $100 billion and pitch literal tents.
The strain it creates
This buildout has consequences. Multi-gigawatt compute demands enormous electricity, straining power grids and drawing local opposition to data centers across the country. The capital required is colossal, raising questions about returns if the AI payoff disappoints. And the environmental footprint of this scramble — power, water, land — is becoming a flashpoint even as the spending accelerates.
The bottom line
Meta’s $115–135 billion AI spend, custom Arm chips and chip-filled tents capture the defining dynamic of 2026: an all-out race for compute. The company is betting that whoever amasses the most processing power wins the AI era — and is willing to spend and improvise on an unprecedented scale to avoid being left behind. The infrastructure arms race has never been more intense.
Photo: MDGovpics / BY via flickr