Google promised Gemini 3.5 Pro for June. By the end of September, the model still had not appeared in the API changelog. In the gap, Google shipped Flash models, speech, video, music and robotics features, and Sundar Pichai was already talking about Gemini 4 on the Q2 earnings call. That sequence says more about Google’s AI position than any benchmark chart.
The search giant can point to real momentum. In February, Gemini 3.1 Pro topped the Artificial Analysis Intelligence Index, beating Claude Opus 4.6 by four points while costing less than half as much to run. According to Comscore panel data, Gemini’s share of AI assistant prompts in the U.S. rose from 17% in January to 30% in June, while ChatGPT fell from 70% to 50%.
The incentives are visible in the numbers
Google has $514 billion in cloud backlog, Cloud revenue grew 82% to $24.8 billion in Q2, and the Gemini app and AI Mode each passed a billion monthly users. But the original analysis argues that these are inventory, not intent. The critical test is where scarce compute actually goes.
- Anthropic has committed to spend $200 billion over five years on Google’s cloud and chips.
- Google Cloud revenue grew 82% to $24.8 billion in Q2, with $8.8 billion in operating income.
- Reuters reported in August that tight compute and disputes between Gemini teams left coding short of resources, while cloud leaders wanted the new Gemini leadership to calm a fight with DeepMind over chips.
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. In April, Sergey Brin told DeepMind staff Google must “urgently bridge the gap” with Claude on agentic coding — and the quarter still ended with Cloud up 82% and the in-house model short on coding targets.
Search has an AI bodyguard, not a successor
Google’s generative AI sits inside the results page: AI Overviews on top, AI Mode beside, ads moving into both. Search and other revenue grew 17% to $63.3 billion in Q2, and Pichai credits the AI features for query growth. For marketers, that means branded and non-branded clicks still happen on Google’s surface.
The more dangerous competitor is the assistant that completes the job and never shows a results page at all: it writes the code, compares the contracts, books the trip and skips the ad slot. Building that product properly means making Search optional — and every incentive inside Google points the other way.
The Microsoft rehearsal
Distribution is not enough. Microsoft had Windows, Edge and Office as default placement and moved first with OpenAI in Bing in February 2023. StatCounter’s August figures put Bing at 4.5% of worldwide search, and Copilot at 2.8% of AI chatbot referral clicks, a rough measure but a clear direction. Microsoft did very well selling shovels to model builders, yet nobody calls it the leader in consumer AI.
Where people pay for work, the picture sharpens. Menlo Ventures’ 2025 enterprise report put Anthropic at 40% of enterprise LLM API spend, OpenAI at 27% and Google at 21%. In coding, Anthropic had 54%, OpenAI 21% and Google 11%. Menlo is one of Anthropic’s biggest backers, so discount accordingly — but the money teams spend to get work done has mostly gone elsewhere.
What marketers should watch
Google can move fast when frightened. ChatGPT triggered a code red in December 2022, Bard’s demo error erased about $100 billion in market value in a day, and Gemini 2.5 Pro later topped LMArena in March 2025. The 3.5 Pro delay may simply be discipline. But the original analysis says Google moves at Microsoft speed unless something is on fire. Search revenue up 17% is not a fire.
For performance marketers, the practical reading is this: don’t mistake AI Overviews and AI Mode for durable AI leadership. Diversify beyond Google’s ad surface, test how often your brand appears in assistant answers, and track enterprise spend as a leading indicator of where paid work migrates. The real disruption signal is not a model launch. It is a Google product that completes the job with no results page and no ad slot, pushed as hard as AI Mode.
Source: Search Engine Journal
