Alphabet Reports Earnings Wednesday With Its Flagship Still Missing

Alphabet reports second-quarter results on Wednesday. It does so without the model it promised the market in May.

Sundar Pichai told the Google I/O audience on 19 May that Gemini 3.5 Pro would ship in June. It did not. It missed a second internal target on 17 July, and it has now missed a third. Bloomberg has reported that the delay comes down to coding performance, with Google unwilling to ship a flagship that underperforms on the exact workload enterprises are buying models for.

Two numbers pointing in opposite directions

The operating business looks strong. Google Cloud revenue rose 63 percent year over year to $20.0 billion in the first quarter. Cloud backlog nearly doubled to more than $460 billion. Google says its models process over 16 billion tokens per minute, up 60 percent in three months. These are not the figures of a company losing the AI market.

The stock disagrees. Alphabet shares have fallen roughly 15 percent from their 2026 peak of $408.61, closing at $356.50 on Friday. The gap between the operating numbers and the share price is the whole story going into Wednesday.

Why a delay costs more than it used to

A late model used to mean a quiet quarter and a launch two months later. The calculus has changed because the alternatives are now genuinely good. Enterprises evaluating frontier models this quarter can choose GPT-5.6, Claude Fable 5, Grok 4.5 or Kimi K3. Those are multi-year contracts with switching costs attached. Every week Gemini is absent is a week some of them get signed with somebody else.

Google appears to know it. DeepMind has registered model names including Gemini 3.6 Flash and Gemini 3.5 Flash Light, which reads like preparation for a cheaper stopgap release to keep the product line moving while the flagship is rebuilt. Shipping a Flash model to cover for a missing Pro model is a reasonable tactical move. It is also an admission.

What to listen for

The interesting part of Wednesday's call is not the cloud number, which will almost certainly be good. It is whether anyone on the call gives a date for Gemini 3.5 Pro, and whether that date comes with a hedge attached.

Three missed windows suggests the problem is structural rather than a scheduling slip. It could be the training run. It could be that Google has set itself an evaluation bar the model has not cleared, which would be the more reassuring explanation and the harder one to say out loud on an earnings call.

I would treat a refusal to name a date as the more informative outcome. A company that has fixed the problem usually says so.