Remember when everyone said Google had already lost the AI war?
Yeah, me too. For the better part of two years the story was simple: OpenAI had the magic, Nvidia sold the picks and shovels, Microsoft cashed the checks, and Google was the big dumb search company scrambling to catch up. That narrative just got torched.
The stock market, that cold-hearted beast that doesn’t care about press releases or viral demos, has quietly made up its mind. And right now it’s screaming one thing: Google is winning the AI race. Not leading a little. Winning. Decisively.
The Moment Everything Changed
It started with a date: November 18th. That’s when Google quietly pushed Gemini 3 into the world. No fireworks, no Sam Altman-style theatrics, just a model update. But the benchmarks that followed weren’t quiet at all.
Gemini 3 didn’t just match the latest GPT models in several categories. It beat them. Sometimes by a lot. And unlike the endless “o3 coming soon” promises from the other side, Google actually shipped it to hundreds of millions of users the same week.
The market noticed.
When Proxies Stop Being Proxies
Here’s the dirty little secret of AI investing over the past two years: most of us were using proxies.
OpenAI is private, so you bought Nvidia (because every ChatGPT run uses their chips) or Microsoft (because they own a massive chunk of OpenAI). Google? Well, you just bought Google. Simple.
Except something wild happened after November 18th. For the first time since 2016, the “Google AI stack” started trading at a premium to the “OpenAI/Nvidia stack” on forward earnings multiples.
The market is saying, clear as day, that the Gemini/TPU ecosystem is now worth more than the ChatGPT/GPU ecosystem. That’s not a rounding error. That’s a regime change.
Think about that for a second. The same investors who turned Nvidia into the most valuable company on earth for a hot minute are now saying, actually, maybe the search giant with the boring name has the better hand.
The TPU Threat Is Real
Everyone loves to talk about Gemini the model, but the sleeper story here is hardware.
Google just announced Ironwood, their seventh-generation Tensor Processing Unit. And unlike every previous generation that lived exclusively inside Google’s own data centers, this one is apparently going commercial. Meaning third-party clouds and enterprises will soon be able to rent TPU clusters instead of GPU clusters.
That sound you hear? It’s thousands of CFOs doing the math.
- Google’s TPUs are purpose-built for the exact matrix multiplications that power large language models
- They claim better performance-per-dollar than Nvidia’s best offerings for inference (the part that actually costs real money at scale)
- And now they’re opening the kimono
No wonder Broadcom, Google’s long-time chip design partner, is up roughly the same percentage as Alphabet this year. The ASIC (application-specific integrated circuit) business just got a lot more interesting.
Code Red at OpenAI
Across the bay in San Francisco, they’re feeling it.
Reports surfaced this week that leadership declared an internal “code red” to focus engineering resources on making the core product better, fast. New features delayed. All hands on deck to close the quality gap.
Gemini’s user count apparently jumped from 450 million to 650 million in just a few months. That’s the kind of hockey stick growth that gets boards nervous.
In my experience covering tech for over a decade, “code red” rarely means “everything is fine.”
Nvidia’s First Real Competition in Years
Let’s be clear, Nvidia isn’t going anywhere tomorrow. Their software moat (CUDA) is still insane, and Blackwell is apparently a monster. But for the first time since the crypto winter, there’s legitimate debate about whether GPU dominance is forever.
When even rumors surface that one of Nvidia’s biggest customers might experiment with Google TPUs, the stock drops 3% in a day. That’s not normal for a company that’s been bulletproof for 24 straight months.
The irony, of course, is delicious. Nvidia spent years telling everyone that general-purpose GPUs would eat the world. Now the company that literally prints money from search ads might eat a chunk of Nvidia’s lunch with specialized silicon. Karma tastes like wafers.
Why This Feels Different
I’ve watched plenty of “AI leader” crowns change hands. Anthropic was the hotness for five minutes. Meta’s Llama models had their moment. But this shift feels structural for three reasons:
- Google finally has a model that’s unambiguously excellent and ships to a billion users overnight
- The hardware play threatens the most profitable business in tech (data center GPUs)
- The valuation flip happened without Google doing anything particularly flashy, just executing
Execution. Imagine that.
There’s something almost boring about Google’s approach that makes it terrifying. No dramatic blog posts at 2 a.m. No cryptic tweets. Just better benchmarks, faster deployment, cheaper inference, and now commercial TPU clusters. Death by a thousand paper cuts, except the paper is made of money.
What Happens Next
The honest answer? Nobody knows.
OpenAI still has brilliant people and could release something mind-blowing next quarter. Nvidia’s Blackwell ramp might crush efficiency expectations. Amazon and Microsoft have their own custom silicon efforts that shouldn’t be ignored.
But for the first time in years, the AI trade isn’t “buy everything that touches GPUs” anymore. The correlation between the big AI stocks just hit all-time lows. Winners and losers are emerging.
And right now, the market’s money is on the company that used to be called a laggard.
Sometimes the tortoise really does win. Sometimes he’s been building data centers and custom chips for fifteen years while everyone else was busy tweeting.
The AI race isn’t over. But the leaderboard just got rewritten, and the name at the top might surprise you.
In a world of overnight sensations and viral demos, maybe the quiet giant with the world’s best distribution and a decade head start on custom silicon was the smart bet all along.
The market seems to think so. For now, that’s what matters.
