Alphabet shed nearly $200 billion in market value on July 16 after Reuters and Bloomberg reported that Google's Gemini 3.5 Pro model had slipped months past its June target because its coding performance still wasn't good enough.
The stock fell 4.4% that day, and the selloff wasn't about earnings or a fresh court ruling. It was about a ship date. At Google I/O in May, Alphabet CEO Sundar Pichai said Gemini 3.5 Pro was due in June, according to Reuters. June came and went. Google would only say that it's "currently testing 3.5 Pro, an upgraded Flash model, and other models with partners." No launch date. No fuller explanation.
Bloomberg's reporting filled the gap. According to Bloomberg, Google updated the data used to train Gemini late in June to improve coding, but the results disappointed people inside the company. Ten current and former Google employees told Bloomberg that the delay has frustrated engineers and researchers alike, and left managers worried that Google is losing ground in coding to Anthropic and OpenAI. That's the part investors heard. Coding isn't a side test anymore. It's where the AI race is being priced.
The Cost Of A Missed Date
Here's the thing about that $200 billion figure. It's larger than Alphabet's planned annual AI infrastructure spend. Alphabet told investors on its February earnings call that 2026 capital expenditures would run between $175 billion and $185 billion, with the money going into AI compute for Google DeepMind, Google Services, Cloud demand, and Other Bets. Markets erased more value in one afternoon than Google expects to spend on data centers and chips - and everything else that goes with them - for the whole year.
That sounds absurd until you look at what the market is actually punishing. Alphabet can afford the infrastructure. It has search, YouTube, Google Cloud, Android, and a balance sheet most companies would envy. The question is whether all that spending keeps producing models that developers and enterprise customers actually want to use first. If the flagship model misses its own launch window because it can't clear internal coding tests, you don't have a communications problem. You have a product problem.
Coding has become the cleanest commercial test in this AI cycle. Enterprises buy AI tools to write, review, debug, and maintain software. Developers are willing to switch when a model saves real time inside their workflow. Chatbot fluency still matters, but nobody is spending heavily just because a model gives warmer answers to general questions. Google knows this. So do Anthropic and OpenAI. That is why a delay tied specifically to coding stings more than a generic model delay would.
Google's Rivals Got A Cleaner Headline
The timing made it worse. On July 16, Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight model. Tom's Hardware, citing Moonshot's own disclosures and public benchmark results, reported that Kimi K3 jumped 17 places from its predecessor to take the top spot on Arena's Frontend Code Arena. It beat Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol on that particular frontend coding leaderboard.
Don't overread one benchmark. Kimi K3 still trails the top closed models on broader measures, and several early comparisons rely on company-published results or benchmark setups that aren't perfectly comparable. But the headline was clean: a Chinese open-weight model was winning a visible coding leaderboard on the same day Google's most important unreleased model was being described as late because of coding. Markets don't need perfect symmetry to react. They need a story they can understand quickly.
Anthropic and OpenAI haven't been waiting around either. Bloomberg's report put both companies directly inside Google's internal worry, because their recent models have raised expectations for coding and agentic tasks. Every release from a rival changes the standard Gemini 3.5 Pro has to meet when it finally appears. A model that would have looked strong in June may need to look stronger in August or September. That's the cost of delay in a market moving this fast.
Alphabet isn't suddenly fragile. That would be the wrong read. But the market reaction tells you something specific about the new investor bargain around AI. Spending huge sums is tolerated only if the models keep arriving on time and looking competitive when they land. Google still hasn't given Gemini 3.5 Pro a new release date. Until it does, the missed date remains the fact investors can trade.
Also read: OpenAI's GPT-5.6 Sol can already find browser exploits, just not finish them • SpaceX is reportedly chasing a Pentagon AI compute deal • AWS billing dashboard told a customer who spent 19 cents that they owed 2.5 billion dollars
