Gemini 3.5 Pro was promised for June at Google I/O and still has not shipped as of an API changelog entry dated September 22. Meanwhile, Google Cloud revenue grew 82% to $24.8 billion in Q2, with $8.8 billion in operating income. Pedro Dias, writing on The Inference and syndicated by Search Engine Journal, reads those two facts together as the story: a company that rents its chips to others faster than it bets them on its own frontier model.

Dias, a 20-year search industry veteran and longtime Google optimist, has changed his position. "I think Google's chances of leading in AI are slim," he wrote on LinkedIn. His argument is not that Google will fall — "Google will be fine" — but narrower: "I don't think Google has the motivation or the nerve to build the thing most likely to eat its search business, because the search business still works."

What happened to Gemini 3.5 Pro?

The confirmed timeline: at I/O in May, Google promised Gemini 3.5 Pro for June. It did not arrive. In mid-July, Bloomberg reported that the model's coding fell short of Google's own bar, and that an attempt to fix this by changing the training data disappointed. Google's statement emphasized shipping many models quickly and keeping them "highly cost-effective for customers."

The shipping part is true. Since July, Google has released Gemini 3.7 Flash, 3.8 Flash, speech, video, music and robotics models. But no 3.5 Pro. On the Q2 earnings call, Sundar Pichai was already talking about Gemini 4 and "our most ambitious pre-training run yet" — before 3.5 Pro exists.

Where the scarce compute went

The scarce resource in AI is compute, and Dias follows where Google's TPUs went. In May, The Information reported Anthropic committed to spend $200 billion over five years on Google's cloud and chips. In Q2, that showed up on the P&L: Cloud up 82%.

A Reuters exclusive from August, per Dias, reported the other side: tight computing capacity and internal disputes left areas such as coding short of resources, with Cloud's senior leaders hoping new Gemini leadership would calm a fight with DeepMind over who gets limited chips. In April, The Information reported on a memo in which Sergey Brin told DeepMind staff Google must "urgently bridge the gap" with Claude on agentic coding.

Dias's read: "A TPU rented to Anthropic is revenue this quarter. A TPU spent on another Gemini training run is a bet. A company that must win takes the bet. A company already winning on the P&L takes the contract."

How well is Gemini actually doing?

The counterarguments are real, and Dias gives them full weight:

• Google's cloud backlog stood at $514 billion at the end of Q2.
• The Gemini app passed a billion monthly users in August; Pichai says AI Mode has passed a billion too.
• Comscore's U.S. panel shows Gemini's share of AI assistant prompts rising from 17% in January to 30% in June, while ChatGPT fell from 70% to 50%.
• Search and other revenue grew 17% in Q2 to $63.3 billion, with Pichai crediting AI features for query growth.

But where money follows work rather than distribution, the picture flips. Menlo Ventures' 2025 enterprise report put Anthropic at 40% of enterprise LLM API spend, OpenAI at 27% and Google at 21%. In coding: Anthropic 54%, OpenAI 21%, Google 11%. Dias flags the caveat himself — Menlo is one of Anthropic's biggest backers — but concludes that even discounted, enterprise spend has mostly gone elsewhere.

Hasn't Google run faster under pressure?

Yes, and Dias concedes it. ChatGPT triggered a code red in December 2022. Two months later, Bard's demo error wiped roughly $100 billion off Alphabet's market value in a day. The recovery was real: Gemini 2.5 Pro topped LMArena in March 2025, and after Gemini 3, it was Sam Altman calling a code red at OpenAI.

His point is about triggers, not capability. "Nobody declares a code red when Search revenue is up 17%." He also accepts the charitable reading of the 3.5 Pro delay — holding back a model that misses your own bar is discipline. It just says nothing about whether Google will build the assistant that makes the results page optional: one that writes the code, compares the contracts, books the trip, and never shows an ad.

The Microsoft precedent

Dias draws the analogy explicitly. Microsoft had Windows, Edge and Office as defaults and moved first, putting OpenAI's tech into Bing in February 2023. Three and a half years later, StatCounter's August figures show Bing at 4.5% of worldwide search, and Copilot sending 2.8% of AI chatbot referral traffic StatCounter tracks — a rough number, Dias notes, since StatCounter counts referral clicks. Microsoft thrives renting cloud compute to model builders, including OpenAI. Nobody calls it the leader in consumer AI.

History rhymes inside Google too, in Dias's telling: Google+ had Gmail, Android, YouTube and everyone's contacts behind it in 2011, and shut its consumer version in 2019. Xerox PARC built the graphical interface and watched others ship it. Google researchers published the transformer paper in 2017; OpenAI turned it into ChatGPT.

Talent flows point the same way. Noam Shazeer, co-lead of Gemini, left for OpenAI in June. Jeff Dean departed in August after 27 years. Fortune, using Zeki Data, reports DeepMind's hires per departure fell from roughly 12 in 2023 to roughly two now, while Anthropic managed 22 in 2025.

What would change his mind?

Two signals, Dias says: a Gemini product that completes jobs end to end — no results page, no ad slot — pushed as hard as AI Mode, and Google letting that product take queries from Search while calling it good news on an earnings call. Watch for the Gemini 3.5 Pro release, any movement in Menlo's enterprise API share, and whether an ad-free agent surfaces before Pichai's Gemini 4 pre-training run delivers.