I keep hitting the buy button on Alphabet (NASDAQ:GOOGL | GOOGL Price Prediction), and the reason is simple. It is the only hyperscaler on the planet that runs AI inference without paying somebody else’s margin to do it.
Every rival renting NVIDIA (NASDAQ:NVDA) GPUs is handing over a 75.0% gross margin toll on every token processed. Google runs its own silicon, its own models, and its own distribution. The receipts back it up.
The Cost Advantage Shows Up in the Filings
On the Q1 2026 call, Sundar Pichai said the new TPU 8i delivers “80% better performance per dollar than the prior generation” on inference, and that after upgrading Search to Gemini 3, Google “reduced the cost of core AI responses by more than 30%”. Trillium (v6) delivers roughly 4.7x better performance-per-dollar and a 67% reduction in power consumption per token compared to equivalent NVIDIA clusters, which is why Midjourney slashed its monthly inference bill by 65% overnight after moving from Nvidia GPUs to Google TPU v6e pods. Anthropic, Character.AI, and Perplexity are running massive portions of their scaling infrastructure on TPUs for the same reason.
The economics are landing on the income statement. Google Cloud revenue grew 63% to $20.03 billion in Q1, backlog nearly doubled quarter-on-quarter to over $460 billion, and Cloud operating margin expanded from 17.8% in the first quarter of last year to 32.9%. Consolidated operating margin reached 36.1% with operating income up 30% YoY. First-party Gemini models now process more than 16 billion tokens per minute… up from 10 billion last quarter.
Q1 EPS of $5.11 cleared the $2.63 consensus, the fourth consecutive quarter of EPS beat, and management raised the dividend 5% to $0.22 per share. Full-stack economics compound.
Why Not Just Buy NVIDIA?
NVIDIA is a superb business. It is also the supplier whose margin every hyperscaler is now engineering around. NVIDIA trades at a P/E of 43 versus 27 for Google, a P/FCF of 53, and a dividend yield of 0.019% versus 0.24% here. Google’s forward P/E is 25. I pay less per dollar of earnings, collect a growing dividend, and own the customer relationship rather than the toll booth people are routing around. Jensen Huang called Blackwell “king of inference today”. Midjourney’s bill and Google’s 30% response-cost cut tell me the toll is being renegotiated in real time.
The Risk I Refuse to Wave Away
CapEx more than doubled to $35.67 billion in Q1, free cash flow fell 46.63% YoY, and 2026 CapEx guidance was raised to $180 billion to $190 billion. If that capital does not earn its return, the thesis bruises. What keeps me adding: CFO Anat Ashkenazi said just over half of the total backlog will convert to revenue in the next 24 months, Cloud operating margin nearly doubled while CapEx doubled, and Pichai stated flatly that “We are compute constrained in the near term. As an example, our Cloud revenue would have been higher if we were able to meet the demand.” The open question is execution, and I like the odds.
What Keeps the Buy Button Active
Owning the silicon, the model, and the distribution in a business where inference cost decides who keeps the customer is a moat I have yet to see anyone else assemble. I will keep adding at $370.92 and above until that stops being true.
Contact [email protected] for any questions or corrections.