Alphabet's Last Three Years: From Ad Company to AI Infrastructure Bet

The story

2023: still cautious

In Q3 2023, Alphabet's quarterly capital spending sat around $8 billion, and the only forward guidance offered was qualitative: full-year 2024 capex would be higher than all of 2023 combined. That year's 10-K listed "more than 75% of revenue from online advertising" as risk factor number one — with no language yet about AI disrupting search.

The escalation: CapEx guidance that keeps being wrong — upward

What followed was a run of guidance that kept needing to be revised up. An early-2025 estimate of "$75 billion for the year" was raised on the Q3 2025 call to "$91–93 billion." FY2026 guidance then opened at "$175–185 billion" — roughly double the prior year — and was raised again within two quarters to "$195–205 billion." On the Q2 2026 call, Alphabet disclosed a negative $5.9 billion in free cash flow for the first time, attributing it directly to capital spending — while keeping the same confident language it always used.

What changed in the filings

The FY2023 10-K said Alphabet would "phase out third-party cookies." The FY2024 10-K softened that to "a proposed user choice model." By FY2025, the topic had disappeared from the filing entirely. In its place, a new sentence appeared in the FY2025 risk factors that wasn't in FY2023 or FY2024: "We believe AI is quickly reshaping the advertising industry, including how ads are delivered online... There is no assurance that we will adapt effectively." The company also quietly revised its advertising-revenue-share language down, from "more than 75%" to "more than 70%."

A tone that never cracked

Through all of this, management's language stayed remarkably even. Right after losing the DOJ antitrust case in Q3 2024, Sundar Pichai said only that Alphabet "plan[s] to vigorously defend these cases" — no sign of alarm. Even when disclosing the first-ever negative free-cash-flow quarter in Q2 2026, the CFO called the business "very confident." Analysts pressed the same question — is this capex really paying off — in Q3 2025, Q1 2026, and Q2 2026, and got essentially the same answer each time: Cloud is already showing returns.

CapEx guidance scorecard

Pattern: the direction was right in all 6 instances, but the specific number was revised upward in 3 of 6 — always underestimated, never overestimated. That reads less like standard conservative guidance and more like the AI investment cycle outrunning even the company's own forecasts.

Source: CFO capex commentary (Anat Ashkenazi/Ruth Porat) on each earnings call; actuals from 10-K FY2025, p.33 and p.52.

Timeline

• Q3 2023Quarterly capex around $8B; cautious, qualitative guidance only.
• FY2023 10-KAdvertising named as "more than 75%" of revenue; no AI-disruption risk language yet.
• Q1 2024"Very confident" language appears; rising capex framed as conviction in the AI opportunity.
• Q3 2024Loses the DOJ antitrust case. First defensive language: "vigorously defend."
• Q1 2025"Headwind" (from depreciation) mentioned three times — first candid admission that capex is dragging on profit.
• Q3 2025Most optimistic tone of the period despite raised capex guidance ("accelerating" ×5, "strong momentum" ×3).
• Q4 2025 / Feb 2026FY2026 capex guided at double the prior year; "we are confident" regardless.
• Q1 2026Capex raised again; 2027 capex flagged as "up substantially" too.
• Q2 2026First-ever negative quarterly free cash flow (-$5.9B) disclosed; "very confident" language retained.

Our read

These three years are the story of an advertising company turning into an AI-infrastructure company. The numbers already show the transition — capex, free cash flow, the shrinking advertising-share language, the vanishing cookie language — but management's tone never once showed real strain through any of it. That gap between the numbers and the language is itself worth watching: if the payoff doesn't show up on the timeline the market expects, the next piece of bad news could land harder precisely because so little strain has been signaled so far.

What we still don't know

• Exactly why the early-2025 $75B capex estimate roughly tripled toward $200B within about a year isn't explained in the transcripts reviewed — the Q2 FY2025 call, which might hold the answer, wasn't in this set.
• Whether the Q2 FY2026 negative free cash flow is a one-quarter timing blip or the start of a longer stretch needs the Q3 FY2026 call to judge.
• The precise reason the third-party-cookie language disappeared from the 10-K — a policy reversal versus simple language cleanup — can't be confirmed from the filings alone.