Alphabet hit $109.9B in Q1 and is starting to sell TPUs to outside data centers

Alphabet posted $109.9 billion in Q1 2026 revenue. Google Cloud grew 63% to $20.03 billion, the backlog nearly doubled to over $460 billion, and Sundar Pichai used the earnings call to announce that Google will start selling TPUs into outside customers’ data centers.

That last line is the one that resets the chip-supply story for 2026. Until this quarter, the only path to a TPU was renting time on Google Cloud. Pichai’s commentary opens a second channel: customers buy or lease silicon and stand it up under their own roof. The pool he named, “AI labs, capital markets firms, and high-performance computing,” reads as the same buyers Nvidia has been quoting at the GTC stage every spring. Google now sits in that conversation as a vendor, not just a hyperscaler.

The numbers

Alphabet’s Q1 release shows the AI-driven shape of the business in clean detail. Total revenue of $109.9 billion is up 22% year over year. Google Cloud revenue of $20.03 billion is up 63%. Net income of $62.58 billion almost doubled the year-ago $34.54 billion, helped by a one-time gain on equity holdings. Operating income hit $39.69 billion. The standout leading indicator is the Cloud backlog, which nearly doubled to over $460 billion of contracted-but-unrecognized revenue.

• Revenue: $109.9 billion, up 22% from $90.2 billion a year ago.
• Net income: $62.58 billion, almost double last year’s $34.54 billion. The jump is partly a one-time gain on equity investments, but operating income also climbed to $39.69 billion.
• Google Search: revenue grew 19%, with AI-experiences pushing query volume to all-time highs. Pichai stated Search latency is down “more than 35% over the past five years.”
• YouTube ads: $9.88 billion versus $8.93 billion year over year.
• Google Cloud: $20.03 billion, the 63% growth headline. The backlog of contracted revenue, the leading indicator analysts care about, sits at $460 billion.
• Subscriptions, Platforms & Devices: $12.38 billion (up from $10.38 billion). Google now reports 350 million paid subscriptions across services, a 25-million quarterly increase, with Gemini app and YouTube Music + Premium driving the gain.
• Other Bets: $411 million in revenue, $2.1 billion operating loss. Waymo crossed 500,000 fully autonomous rides per week.

TPUs go retail

The TPU disclosure is short on contracts and long on signal. Pichai said Google will offer Tensor Processing Units to “a select group of customers in their own data centers,” without naming the buyers, the chip generation, or the price. The framing matters more than the specifics today: Google is opening a second commercial path for TPUs, one that lets customers run them outside Google Cloud’s perimeter.

Two things change with that. First, customers who refused to consolidate workloads on Google Cloud for compliance or vendor-balance reasons now have a way to buy the silicon without rewriting their networking stack. Second, the TPU roadmap becomes a product line that has to compete on its own merits, not as a captive to a cloud SKU. Pricing transparency, support contracts, and a public driver story all move from “internal tool” to “vendor expectation.”

The Gemini-side numbers underline why Google can afford to widen the TPU door. “Our first-party models, like Gemini, are now processing more than 16 billion tokens per minute via direct API use by customers,” Pichai said, attributing a 60% quarter-over-quarter increase in token throughput. Gemini Enterprise paid monthly active users grew 40% quarter over quarter. With internal demand still climbing, the only way to free TPU supply for outside sales is to keep building, which is exactly what the capex line is paying for.

What about I/O

Pichai also previewed Google I/O 2026. He said the company is “excited to share more about Search at I/O” and that the next show will focus on “pushing the next frontiers of foundation models, including intelligence, agents and agentic coding.” Coding-agent hints at I/O fit a year in which Anthropic’s Claude Code, OpenAI’s Codex, and Cursor’s agent modes have all raced to ship comparable surfaces; Google’s pitch presumably runs through Gemini and the IDE integrations that already ship with Android Studio and Cloud Workstations.

The I/O dates have not been published in the Q1 release. Google’s developer site historically uses mid-May, and a Sunday-newsletter-style confirmation usually arrives within two weeks of the earnings call.

What this means for you

If you’re an enterprise architect deciding where AI inference lives, the TPU-out-of-cloud lane is the development to watch over the next two quarters. Buying chips into your own data center has a different cost shape than paying for inference per token: lower marginal cost at scale, but a bigger up-front bill and operational ownership of cooling, scheduling, and driver updates. Google’s pitch will have to include the runtime support that’s been Nvidia’s moat. That story is not yet public.

If you’re shipping product on Gemini, the throughput gain matters more than the chip news in the short term. 16 billion tokens per minute on first-party APIs is rate-limit headroom. Plan capacity for usage that doesn’t hit a cap. And mark I/O on the calendar; the Search and agent demos are where Google tends to drop the actual SDKs that turn earnings-call slogans into shippable code.