Google held its Q2 earnings call on Wednesday. Revenue was $119.8 billion, up 24%. Cloud revenue jumped 82% to $24.8 billion. Nearly 90% of the Fortune 100 now use Gemini Enterprise .
But analysts weren't asking about the revenue. They were asking about the delay. The coding gap. The confidence.
CEO Sundar Pichai struck an unusually defensive tone. He was pressed by JPMorgan and Barclays analysts on whether Google had quietly ceded the frontier to OpenAI and Anthropic . His answer? "We've had clearly frontier models. There are many attributes on which we are still at the frontier; there are areas where we've acknowledged we need to improve — coding and agentic coding is an example of that" .
For a company that invented the transformer, that's a candid admission.
Gemini 3.5 Pro, originally slated for June, remains in partner testing with no firm release date . Google also confirmed it is already pretraining Gemini 4 — described as a "significantly larger" model and its "most ambitious" pre-training effort to date . Pichai said the Gemini 4 roadmap includes releases "almost at a monthly cadence" .
The concern is not just about one delay. It's a pattern. While OpenAI Codex and Anthropic Claude have been peeling off developers, Google has been slow to ship its response. A year ago, Google was closing in on Nvidia's market cap; now its CEO is in a defensive posture, answering whether the company can keep pace.
The gap extends beyond the flagship itself. Pichai's admission on coding points to a structural vulnerability. For months, Google had framed itself as "the AI-first company." But in the most contested arena of the industry — developer tools and agentic workflows — the company has been on the back foot.
Google has a counterpunch. Gemini 3.6 Flash launched this week with a 10-point gain on a key coding benchmark and up to 65% token efficiency improvement on agentic tasks . Flash Cyber is a security-focused vertical model for vulnerability detection. Flash-Lite is the cheapest, fastest option for high-volume workflows .
But those are not the flagship. They are the bridge.
Gemini 4 is the real bet. The "most ambitious" description matters less than what it implies: Google is putting its best compute resources into the next generation, not the current one . The company that defined the modern AI era with the transformer paper is now essentially saying, "Wait for the next one." That's a strategy. It's also a risk, because by the time Google ships Gemini 4, the frontier may have moved again .
The market's reaction was telling. Alphabet shares fell more than 3% after hours . The company is spending record amounts — $195 billion to $205 billion in capex this year — and investors want to see returns.
Pichai's defense was not without strong numbers. Cloud backlog hit $514 billion. Nearly 500 cloud customers have run more than a trillion tokens through Gemini over the past year . But when the CEO of the company that invented transformer architecture is spending an earnings call arguing that it's not losing, the argument itself becomes the signal.
P.S. If you're an AI developer, the question is simpler: do you wait for Google to catch up, or do you build on the tools that are already shipping? Pichai says Gemini 4 will be worth the wait. The market is hoping he's right.
