Latest / Minds, Bodies, and Terawatts / Google's Gemini 3.5 Pro Is Months Late; Alphabet Loses $200 Billion in a Day
Transcript
- 0:00[Host] This is Minds, Bodies, and Terawatts, where being the smartest isn't the same as winning. Bloomberg reported Thursday, July sixteenth, that Google is months behind on Gemini three-point-five Pro, and Alphabet shares fell about four percent, wiping out two hundred billion dollars in a single day.
- 0:17[Guest] Google unveiled that model back in May as its most powerful ever. Now people close to the work say it's stalled, mostly on coding, and the launch keeps slipping.
- 0:29[Host] Why coding specifically? Is that really the thing worth two hundred billion dollars?
- 0:34[Guest] Coding is where the frontier labs prove themselves right now, and it's where OpenAI and Anthropic have pulled ahead. On that same May stage, Sundar Pichai said the Pro version would ship in June. June came and went, and it still hasn't.
- 0:50[Host] The irony is thick, though. Google invented the transformer, the "T" in ChatGPT, back in 2017. How does the company that wrote the recipe fall behind on the dish?
- 1:02[Guest] Inventing the technique and shipping the best product are different sports. Google has the talent and the data centers; what it lacks is a model that clearly beats the rivals on the test that matters, a working app.
- 1:16[Host] And investors punished a delay like it was a death. Why so brutal?
- 1:21[Guest] Because they're not pricing the delay. They're pricing the fear that Google can't hold a lead in the one business it was supposed to own. When your flagship slips and rivals don't, "late" quietly becomes "losing."
- 1:35[Host] But here's what doesn't look like just a scary headline. A few weeks earlier, Bloomberg reported that two of Google's senior Gemini researchers, the people who helped build the model, were leaving for Anthropic, part of a wider talent exodus. Bad week, or a slow bleed?
- 1:53[Guest] Talent walking to a rival stings, no argument. But step back, because a bleed only matters if raw capability is the moat. There's a piece on unscarcity dot ai called The Commoditization of Intelligence, and one line fits perfectly: the best model wins the demo, the cheapest good-enough model wins the invoice.
- 2:14[Host] Meaning the frontier crown might be worth less than the market thinks.
- 2:18[Guest] Right. Most AI work in the world isn't frontier work, it's routine. Once several labs are good enough, raw capability stops being a moat, and a few months of delay matters far less than today's panic assumes.
- 2:33[Host] But if the smartest model isn't the moat, what actually protects Google here?
- 2:38[Guest] Distribution. There's a companion piece called "Network Effects" arguing the durable advantage isn't intelligence, it's being the default. Gemini already lives inside Search, Android, and Gmail, on a few billion phones.
- 2:50[Host] So Google can lose the benchmark and still win the invoice, because it's already in everyone's pocket.
- 2:57[Guest] That's the bet. A slightly worse model that's already everywhere beats a slightly better one you have to go find and install. OpenAI got the trophy; Google still holds the distribution.
- 3:09[Host] Which makes you wonder if that two-hundred-billion-dollar drop was investors watching the wrong scoreboard.
- 3:14[Guest] Maybe so. The scoreboard everyone watches is capability. The one that actually pays is who owns the pipe the intelligence flows through.
- 3:23[Host] So a week like this is really about intelligence itself, scarce and expensive for all of history, sliding toward something closer to a utility. Patrick Déglon calls that the Foundation in Unscarcity, the point where genius gets cheap enough to be infrastructure.
- 3:39[Guest] And the fight moves from who's smartest to who owns the tap. For more on the commoditization of intelligence and where the next AI bottleneck shows up, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.