Quick answer. In 2026 the compute behind AI is being committed years in advance. Alphabet spent $35.7 billion on infrastructure in a single quarter, Meta booked $237.67 billion in long-term capacity commitments, and Broadcom locked multi-year custom-silicon deals with Meta and Google. The spending is shifting decisively toward custom chips.
Capital is being committed years ahead of the workloads
The defining feature of the 2026 buildout is the scale and duration of the financial commitments behind it. Alphabet reported first-quarter 2026 capital expenditure of $35.7 billion, the overwhelming majority directed at technical infrastructure for AI. Meta disclosed $237.67 billion of non-cancelable contractual commitments as of March 31, 2026, most tied to third-party cloud capacity and technical infrastructure. Oracle, on the revenue side of the same trend, reported fiscal 2026 total revenue of $67.4 billion with Cloud Infrastructure up 77 percent to $18.1 billion. Together these disclosures describe capital being locked to future AI demand rather than current usage.
The spend is shifting from merchant GPUs to custom silicon
Where that capital goes is changing. Broadcom and Meta extended their partnership through 2029 to support multiple gigawatts of Meta's custom silicon and to co-develop what they describe as the industry's first 2nm AI compute accelerator. In a separate SEC 8-K, Broadcom disclosed a long-term agreement to develop and supply Google's future TPUs, plus a supply-assurance deal for networking and AI-rack components running through 2031. AMD, positioned further up the supply chain, announced more than $10 billion of Taiwan ecosystem investments to scale the advanced packaging that these chips require. The common thread is a multi-year bet on proprietary accelerators and the packaging to build them.
Forward guidance points sharply higher
The forward numbers are larger still. Pichai said Google began delivering TPU systems to customer datacentres in the second quarter and raised its 2026 capital-expenditure guidance to $195-205 billion, citing accelerated capacity delivery to meet demand. CoreWeave said it had surpassed 1 GW of active power and expanded total contracted power to more than 3.5 GW. Read as guidance rather than settled results, both figures signal that operators expect demand to keep outrunning today's capacity.
FAQ
How much is Alphabet spending on AI infrastructure in 2026?
Alphabet reported $35.7 billion of capital expenditure in the first quarter of 2026, mostly on technical infrastructure, and Pichai raised full-year guidance to $195-205 billion.
Why are these companies designing custom AI chips instead of buying GPUs?
Broadcom signed multi-year deals to build Meta's 2nm accelerator through 2029 and to supply Google's future TPUs through 2031.
What did Meta commit to for future compute capacity?
Meta disclosed $237.67 billion of non-cancelable commitments as of March 31, 2026, largely for third-party cloud capacity and infrastructure.
Sources
Oracle — investor.oracle.com (2026-06-10)
Broadcom — investors.broadcom.com (2026-04-14)
Meta — www.sec.gov (2026-03-31)
CoreWeave — investors.coreweave.com (2026-05-07)
Broadcom — www.sec.gov (2026-04-06)
Google / Alphabet — abc.xyz (2026-04-29)
Sundar Pichai — abc.xyz (2026-07-22)
AMD — ir.amd.com (2026-05-21)
