← Back to feed

Alphabet’s full-year capital expenditure guidance for 2026 now stands at $205 billion — $15 billion above the original estimate of up to $190 billion. The revision came during the Q2 2026 earnings call, where CFO Anat Ashkenazi attributed the increase directly to rising compute demand and the need to expand capacity ahead of constrained availability.

Ashkenazi also flagged that capex is expected to keep climbing into 2027, making this an upward guidance revision with no declared ceiling rather than a one-time adjustment.

Why the Increase

The driver is straightforward: enterprise AI demand is outrunning the infrastructure Alphabet built to service it. Google Cloud posted $24.8 billion in Q2 revenue, up 82% year-over-year, with a $514 billion backlog. That backlog is not speculative demand — it represents signed commitments, including Anthropic’s five-year $200 billion Google Cloud agreement. Spending into confirmed revenue has a different risk profile than speculative buildout.

The Q2 earnings call also marked the first quarter in which Google reported external revenue from its Tensor Processing Unit systems. TPU racks deployed to customer data centers — a model Nvidia has dominated with GPU cloud deployments — generated external revenue for the first time. That hardware-to-customer channel is new for Google, and scaling it requires supply that only capital expenditure can produce.

Industry Context

Synergy Research Group, reporting alongside the earnings, put the broader data center buildout in context: overall US data center capacity is expected to double over the next three years. Google is not outlier-spending relative to that trajectory. AWS and Azure have both flagged capacity constraints as a Q2 2026 theme, and Microsoft is running a $190 billion capex program of its own.

The relevant comparison is whether Google is growing its infrastructure share relative to its hyperscaler peers. The Cloud backlog of $514 billion — growing faster than revenue, which is itself growing faster than any comparable period — suggests it is.

What $205 Billion Actually Buys

For context: $205 billion in a single fiscal year is larger than the annual revenue of every company in the S&P 500 outside the top 15. At a rough 40% allocation to AI-specific compute infrastructure (data centers, networking, custom silicon), that implies approximately $80 billion directed at AI capacity in 2026 alone.

The 8th-generation TPU platform, unveiled at Google Cloud Next ‘26, sits at the center of this buildout — specifically the split between the TPU 8t training variant and the TPU 8i inference variant. The inference-optimized chip targets agentic workloads that are running continuously rather than in batch, which has fundamentally different infrastructure requirements from training runs.

The capex trajectory also reflects Alphabet’s view that the revenue per compute dollar is improving. Cloud margins expanded to 35.6% in Q2, nearly double the year-ago rate. At that margin level, incremental infrastructure spending has a shorter internal payback period than it did twelve months ago.

Capacity as Strategy

For the first time in its cloud history, Google is capacity-constrained. The $205 billion guidance increase, and the forward signal that 2027 spending will exceed 2026, represents an acceleration into that constraint rather than a managed pace-match.

The risk of building ahead of demand is real. But given the structure of the Cloud backlog — long-term contracts with known counterparties — Alphabet’s management is treating capacity as signed revenue that hasn’t been delivered yet, not as a speculative bet on demand materialising.