Alphabet has signaled a transformative shift in its industrial strategy, unveiling a massive, sustained hardware build-out designed to break a persistent supply-side bottleneck.
The company issued an unprecedented 2026 capex forecast of US$175-185 billion, nearly doubling the US$91.4 billion spent in 2025. This aggressive guidance serves as a clear directive to the global tech supply chain—from foundry giants to component vendors—that the AI hardware boom is entering a phase of rapid acceleration.
Credit: Alphabet
Escaping the supply-choke: A massive push for verticalization
The internal mechanics of this investment reveal a calculated push to verticalize and secure critical infrastructure in an environment where demand currently outstrips deployment capacity.
CFO Anat Ashkenazi clarified that approximately 60% of technical infrastructure spend is allocated to servers, while 40% is directed toward data centers and networking equipment.
This allocation highlights a surge in demand for the entire "server rack" ecosystem, supported by Alphabet's strategic move to acquire Intersect, a provider of data center and energy infrastructure solutions.
This shift toward facility ownership and energy security is a response to what CEO Sundar Pichai described as a "supply-constrained" environment. With time horizons lengthening, Alphabet is being forced to plan further ahead, likely securing silicon and energy components through longer-term, high-volume agreements.
The dual-track silicon strategy: Merchant vs. custom accelerators
To maintain compute dominance, Alphabet is doubling down on a dual-sourcing strategy for processors that balances merchant silicon with aggressive internal engineering. Pichai confirmed that the company will be among the first to offer Nvidia's latest "Vera Rubin" GPU platform. Simultaneously, the company is scaling its own seventh-generation custom silicon, the "Ironwood" TPU.
For the supply chain, this ensures that while merchant silicon remains vital, Alphabet's custom ASIC fab partners and OSATs remain equally critical to the stack.
Far from being speculative, this build-out is de-risked by a Google Cloud backlog that surged 55% quarter-over-quarter to US$240 billion, representing hardware that has effectively already been "rented" by major enterprise clients and partners like Apple.
Credit: Alphabet
The efficiency paradox and the token explosion
Despite the staggering costs, Alphabet is banking on an efficiency paradox to drive further volume. The company reported a 78% reduction in serving unit costs for Gemini models over the past year through model optimizations and utilization improvements. However, this optimization has not dampened the need for new chips.
Instead, lower costs have acted as a catalyst for broader adoption, with first-party models like Gemini now processing over 10 billion tokens per minute.
For supply chain partners, this suggests that as Alphabet makes AI computing more efficient and affordable, the resulting explosion in usage volume—from "AI mode" searches that are 3x longer than traditional queries to autonomous world-building in Project Genie—will only necessitate a larger, more complex hardware footprint through 2026.
Article edited by Joseph Tsai
