Google's flagship AI model is months behind schedule, Alphabet stock is sliding, and the only people guaranteed to lose are the retail investors who bought the AI hype.
The Gemini 3.5 Pro upgrade — widely expected at Google's May developer conference — has slipped by several months because the model's coding capabilities can't meet internal benchmarks, Bloomberg reported Thursday, citing 10 current and former employees. Alphabet shares dropped more than 4.5% to $354.17, wiping out billions in market value. The working Americans holding GOOGL in their 401(k)s felt that hit. The insiders with early warning didn't.
This is the pattern every tech bubble runs: the connected get the exits, the public gets the bag. Google's AI division can't ship a competitive product, its own engineers are fleeing to rivals, and the stock is telling the story the press releases won't.
The core problem is coding — the commercially critical capability that lets AI models actually write software. Both OpenAI and Anthropic have released models that outperform Google's current offerings, and the gap is widening. Late last month, Google updated Gemini's training data to try closing the gap. The results were disappointing, according to one person familiar with the matter.
Part of the delay is structural bloat. Multiple teams across Google — DeepMind, Cloud, Android, and Search — are all building AI coding tools simultaneously, creating overlapping efforts and competing priorities that slow execution. Co-founder Sergey Brin has been pushing the company to move faster, but his efforts have been hampered by internal factions and by engineers who believe important code should still be written by humans to meet Google's standards, former employees told Bloomberg.
The dysfunction is driving talent out the door. Senior staff have departed for Anthropic and other labs, with frustration over Google's competitive position cited as a driving factor. Engineers who try to use AI for their own work hit capacity constraints because teams inside Google are fighting over computing power — a problem external customers face too. Only some teams are even allowed to use Anthropic's Claude, with access restricted to groups doing cutting-edge research.
Google is scrambling to consolidate. Chief AI Architect Koray Kavukcuoglu is working to unite the company's internal AI coding tools, and a new DeepMind team led by research engineer Sebastian Borgeaud has been formed to tackle the problem. The company claims 75% of code at Google is now AI-generated and says it has consolidated developer tooling under an internal platform called Antigravity.
A Google spokesperson told Bloomberg the company is "shipping quickly across a wide range of models while keeping them highly cost-effective for customers" and is "productively engaged with the US government on model testing and broader frameworks." That corporate language doesn't match the reality on the ground. Customers are voting with their wallets. Freddy Vega, CEO of education platform Platzi, said Google's Flash model is more expensive and slower than its predecessor while remaining far less capable than competitors. His team shifted to Anthropic.
Benzinga framed this as a straightforward stock-move story. TNW provided the deeper picture — internal dysfunction, talent flight, customer defections, and Brin's frustrated push for speed. What neither outlet asked: who sold before the delay went public, and who's still telling retail investors to buy the dip?
The AI race isn't just about which model wins. It's about who gets wiped out when the hype meets reality. Google still has the cash and the infrastructure to grind forward. The question is whether the Americans holding the stock will still be in the game by the time Google actually ships.