Alphabet Q2 2025: $119.8B Revenue as AI Eclipses Ads Focus

Alphabet Q2 2025: Google Is Now a Compute Company That Still Runs Ads

Alphabet posted $119.8B in Q2 revenue. Google Cloud hit $24.8B with 82% growth. On the earnings call, Sundar Pichai mentioned "TPUs" 28 times versus "advertising" 21 times. No investor asked about ads. Pichai called competing at the AI frontier "the foundation for everything we do."

This is either the same AI-first speech Pichai has given since 2016, now with better numbers — or it represents a real structural shift. The answer matters because those are two different companies to build on or against.

The case that something actually changed

Search still printed over $54B, so Google hasn't abandoned ads. But Google Cloud is now large enough that the research apparatus — DeepMind, Gemini, the TPU program — is justified on infrastructure economics rather than ad-subsidy logic. A decade ago, every dollar spent on Gemini had to trace back to a search ad. Now it doesn't. That means Google can absorb margin compression on AI search features without existential pressure, because the cloud business covers the compute cost. That decoupling is the real news under the word count.

The 28 TPU mentions on a public call is an investor relations decision, not a slip. Google is telling enterprise buyers its custom silicon is a product you can rent. That matters because AI compute pricing is currently set by NVIDIA scarcity. Google runs a full vertical stack — chip design, TSMC fabrication, and the software (JAX and XLA) to use them. If TPU v6 capacity is real and available, that's the first credible pressure on NVIDIA's pricing power in two years. The supply side was conspicuously unquantified on the call, which is worth noting.

Who wins and who loses

Builders renting Google compute benefit if TPU capacity materializes — more supply options, pricing pressure on NVIDIA. Ask for firm capacity commitments in writing, not roadmap slides.

Publishers and ad-tech vendors depending on Google Network inventory, header-bidding integrations, or Google Display Network campaigns face a structural attention shift. Network revenue is declining while Cloud doubles. Engineering and product management attention inside Google flows toward inference infrastructure, not ad serving. Expect slower Google Ad Manager feature iteration and more aggressive bundling of AI credits into Google Cloud Platform deals that squeeze standalone ad-tech vendors. The practical signal: your Google rep starts pitching Vertex AI before fixing your ad bug.

Pichai named AGI as Alphabet's first priority on the record to institutional investors. That converts AGI from a capability threshold to manage carefully into a competitive objective the board is now financially committed to. Internal teams arguing for deployment delays are now arguing against the CEO's investor promise. There was no safety language on the call in a quarter where Anthropic has been vocal about it.

The prediction

By Alphabet's Q4 2025 earnings call (around February 2026), Google will cite TPU or Cloud capacity constraints as a limiter on growth — the same supply-wall pattern AWS, Azure, and NVIDIA customers have each hit once AI compute demand became real and public. Google spent this call selling TPUs as a first-class product while saying nothing about capacity limits. That is precisely the setup before a supply-constraint admission. The only scenario where this is wrong: Google reports ample, unconstrained TPU capacity into 2026, making it the only hyperscaler not gated by fab and power limits. Confidence: medium.

Google just told you what it is, and it isn't an ad company anymore — at least not in its own head. Alphabet printed $119.8B for Q2, Cloud nearly doubled to $24.8B, and on the earnings call "TPUs" beat "advertising" 28 mentions to 21. No investor asked about ads. Pichai called competing at the AI frontier "the foundation for everything we do." For anyone building on or against Google's stack, the question is whether that reorientation changes what you can rely on from them — or whether it's the same AI-first speech Pichai's given every year since 2016, this time with better numbers.

Reversibility: Type 1 for Google internally — you don't un-ring an AGI-first strategy bell to investors. Type 2 for you the builder — you can rebalance your inventory mix or your compute vendor over a quarter or two. What's actually being decided: not "does Google still sell ads" (it sold $54B+ of search this quarter), but "where does Google's engineering and capex attention flow for the next two years, and does that leave open-web ad-tech and NVIDIA-alternative compute in a different place than you priced." Forcing function: none hard. Cloud momentum and the next TPU generation set the real clock.

The Skeptic. Pichai has been giving the AI-first speech since 2016. Word-count tea leaves make a great earnings recap and mean almost nothing operationally. Google Search still printed north of $54B this quarter — that's not a company walking away from ads, that's one embarrassed by how much it still needs them, so it talks about chips instead. Cloud growing 82% off a smaller base is table stakes for a hyperscaler; the test is whether that holds when AWS and Azure swing back, not whether it doubled once. And the open-web Network decline? Partly Google voluntarily dumping low-margin junk inventory. For the PM: a company's earnings-call vocabulary tells you what it wants Wall Street to believe, not what its engineers actually shipped.

The Compute Pragmatist. Twenty-eight TPU mentions on a public call is an IR decision, not a slip. Google is telling hyperscaler buyers its custom silicon is a product you can rent, not an internal cost hack. That matters because the whole AI economy is priced off NVIDIA scarcity right now, and Google runs a full vertical stack — chip design, TSMC fab, and the software (JAX, XLA) to actually use them. If TPU v6 yields hold and capacity is real, that's the first credible pressure on NVIDIA's pricing power in two years. The open question is supply: are 28 mentions describing chips you can buy today, or chips Google hopes to have before Amazon's Trainium eats the same demand? For the PM: Google is trying to become a place you rent AI chips, not just a place you run ads.

The Researcher. The vocabulary inversion is a real signal even if the Skeptic's right that it's stage-managed. Here's the mechanism that makes it matter: at $24.8B Cloud with 82% growth, Google's research apparatus — DeepMind, Gemini, the TPU program — is now justified on infrastructure economics instead of ad-subsidy logic. That decoupling buys Google something specific. It can eat margin compression on AI search overviews without existential panic, because the compute business pays for the compute. A decade ago every Gemini dollar had to trace back to a search ad. Now it doesn't. That's the quiet structural change under the word count. For the PM: Google's AI can now lose money on search and still make sense, because the cloud business foots the bill.

The Safety Lens. Pichai named AGI as Alphabet's first priority — to institutional investors, on the record. That reframes AGI from a capability threshold you manage carefully into a competitive objective the board is now fiduciarily tied to. Once "reach the frontier faster" is the stated mandate, every internal safety team arguing for a deployment delay is arguing against the CEO's investor promise. Notice what wasn't on the call: any safety language at all, in a quarter where Anthropic won't shut up about it. The TPU capex says compute scaling isn't slowing. Race frame, not responsibility frame. For the PM: when the boss tells shareholders that building superhuman AI is job one, "let's slow down and test" becomes a harder sentence to say inside the building.

The Builder. Forget the AGI poetry — what changes on your Tuesday? If you depend on Google Network inventory, header-bidding integrations, or GDN-heavy campaigns, the signal is structural. Network revenue is declining while Cloud doubles, which means PM and engineering attention inside Google flows toward inference, not ad-serving. Expect slower GAM feature iteration, longer waits on publisher-side fixes, and more aggressive bundling of AI credits into GCP deals that squeeze standalone ad-tech vendors. The 90-day tell: your Google rep starts pitching Vertex AI before they fix your ad bug. For the PM: the team fixing your Google ads tooling is getting smaller, and the team selling you AI chips is getting bigger.

Where they split. Three real disagreements. The Skeptic says the word count is theater and Search's $54B proves nothing changed; the Researcher says the reason the theater is possible — Cloud now paying for its own compute — is the actual news. They can't both be right about whether this quarter mattered. Second: the Compute Pragmatist sees a credible NVIDIA challenger in TPUs; that entire thesis rests on supply Google conspicuously did not quantify. Third: the Safety Lens reads AGI-as-priority as a governance red flag, while the Builder reads the same capex as good news for anyone renting Google's compute. Same sentence from Pichai, opposite implications depending on whether you're buying chips or worrying about what runs on them.

What it hinges on. Two facts, both checkable. One: is TPU capacity real and rentable at scale, or aspirational positioning ahead of supply constraints? That determines whether the "NVIDIA alternative" story is a 2026 reality or a 2028 hope. Two: does the open-web Network decline keep accelerating? If Google is genuinely deprioritizing third-party publisher inventory, that reprices every open-web ad-tech vendor downstream of it. The council leans toward the Researcher and Compute Pragmatist over the Skeptic: the word count is stage-managed, but the thing being staged — Google monetizing compute directly — is a genuine break from the ad-subsidy era. If you build on Google, verify your Network dependency and price in slower ad-tooling iteration. If you buy compute, ask Google's cloud team for firm TPU v6 capacity commitments in writing, not roadmap slides.

Prediction: Google Cloud's next TPU-capacity signal will be a shortage story, not an abundance one — by Alphabet's Q4 2025 earnings call (early Feb 2026), Google will cite capacity or supply constraints as a limiter on Cloud growth, the same way it did NOT this quarter.

Confidence: Medium — every hyperscaler hits the supply wall once demand is proven.

Why: Google spent this call selling TPUs as a first-class product and posted 82% Cloud growth, which means demand is now real and public. The pattern across AWS, Azure, and NVIDIA's own customers is consistent: once AI compute demand is proven, the binding constraint flips from "will anyone buy this" to "can we build it fast enough," and management starts blaming supply for not growing faster. Google leaning this hard into TPU messaging while saying nothing about capacity is exactly the setup before a supply-constraint admission. The opposite outcome — Google reporting abundant, unconstrained TPU capacity into 2026 — would make it the only hyperscaler not gated by fab and power limits, which the vertical-integration story alone can't buy.

Revisit by 2026-02-15: We're right if Alphabet's Q4 2025 call names TPU/Cloud capacity or supply as a constraint on growth. We're wrong if management describes TPU capacity as ample and demand as the only limiter.