← Back to feed

Alphabet’s Q2 2026 earnings disclosed a capital expenditure figure that changes the financial story of the AI buildout: $44.9 billion spent in a single quarter, the largest capex number in the company’s history and a figure that pushed Alphabet’s free cash flow negative for the first time.

The company raised its full-year 2026 spending guidance to as much as $205 billion — approximately $15 billion above the original estimate published at the start of the year.

The Numbers That Matter

Google Cloud reported 82% revenue growth and strong operating margin expansion in the same quarter. The Cloud story was already covered when Q2 results dropped. What came into sharper focus in the days following is what that buildout is costing:

• $44.9B capex in Q2 alone
• Free cash flow: negative — first time in company history
• Inventory quadrupled to $10 billion, primarily AI hardware
• 2026 guidance: up to $205B — revised upward after Q2 results
• CFO guidance: capex will increase further in 2027

The inventory jump is significant. Alphabet has historically maintained lean hardware inventory. Quadrupling it to $10 billion indicates the company is positioning ahead of its own deployment schedule — buying chips before it has the facilities or contracts to absorb them, betting that demand will materialize and that supply constraints make early procurement necessary.

Cash Flow Context

Alphabet generated approximately $26 billion in operating cash flow in Q2. Capital expenditure of $44.9 billion means the company spent $18.9 billion more than it generated from operations. That deficit is funded from its existing cash reserves — Alphabet entered Q2 with roughly $100 billion in cash and marketable securities — but the direction of travel is notable.

For comparison, Alphabet’s total capex for full-year 2023 was $32.3 billion. The company spent more than that in a single quarter in 2026.

The CFO’s public statement that capex will increase in 2027 is unusual candor. Hyperscalers typically guide toward the next quarter; flagging increases a full year out signals that the buildout commitment is structural, not quarter-to-quarter discretionary spending.

What the Capex Is Buying

Three categories dominate Alphabet’s AI infrastructure spend:

TPU capacity. Google’s 8th-generation TPU platform, with the separate 8t (training) and 8i (inference) chips, is the primary compute substrate for both Gemini model training and the Google Cloud AI infrastructure business. Alphabet has been selling TPU capacity to third parties — including Anthropic under its five-year, $200 billion deal — and the inventory buildup reflects chips being positioned for external revenue.

Data center construction. The Texas three-campus program, the Jackson County Alabama expansion, and facilities commitments in multiple geographies are all in active construction or pre-construction procurement phase. Physical build cycles run 18-36 months; spending now shows up in Q2 capex before the facilities generate revenue.

Power and cooling infrastructure. At $44.9B in quarterly capex, a material portion is power generation, cooling systems, and grid interconnection capacity — costs that scale with data center size and that have their own multi-year lead times.

The Payback Math

Google Cloud’s $514 billion backlog — larger than Alphabet’s entire annual revenue from two years ago — provides the contractual basis for continued spending. Cloud operating margins nearly doubled to 35.6% in Q2. The internal payback model requires that Cloud revenue growth continues at elevated rates long enough to absorb the capex.

The risk is not that Alphabet lacks the cash to fund the buildout. The risk is that the transition from cash flow positive to cash flow negative, at a quarterly capex rate that the CFO says will increase, creates a new kind of financial exposure for a company whose investors have historically valued its balance sheet strength as much as its growth.

Alphabet is now spending like a construction company building infrastructure for a market that does not yet fully exist at the scale the spending implies.