Companies have spent two years chasing the smartest AI on the market. Now they're chasing a bill they can actually pay.
That shift is the opening Google has been waiting for.
Pichai said recently that companies are torching through their yearly token budgets just five months in. (Tokens are the units AI models charge by - every question, every reply, every step an AI agent takes eats them up.)
Google's own numbers show why. Monthly token usage across its AI products has jumped to 3.2 quadrillion - seven times higher than a year ago.
And the bills are starting to bite. Uber's COO recently said the company's AI costs are getting harder to justify, while venture capitalist Chamath Palihapitiya said his firm 8090 moved away from the coding tool Cursor after token spending got out of hand.
The new wave of AI agents - programs that run long tasks on their own - burn tokens faster than chatbots ever did. The more useful AI gets, the more it costs to run.
Every weekday morning, Market Briefs breaks down the moves shaping the market in five minutes - plus you get a free investing masterclass when you sign up.
Google builds its own AI chips, called TPUs, and runs them in its own data centers. It also buys the parts that go into those chips straight from the makers.
Analysts at William Blair estimate that setup gives Google a 50% to 75% discount on its own AI compute compared to rivals. Every dollar a competitor spends on AI workloads, Google spends 25 to 50 cents.
OpenAI doesn't have that. Every ChatGPT request runs through Microsoft or Oracle's cloud, those clouds pay Nvidia for the chips underneath, and each layer takes a cut.
That's why Pichai can pitch Gemini 3.5 Flash as the "good enough" option for most jobs - and back it up with the math. He said Google Cloud's biggest customers could save over a billion dollars a year by shifting 80% of their AI work to a mix of Flash and top-tier models.
Flash is built to handle the bulk of everyday AI work - summarizing documents, answering customer questions, sorting through data - for a fraction of the cost of top-tier models. The trade-off is that it's not as sharp at the hardest tasks.
For most jobs, companies don't need the smartest model on the market. They need one that won't break the bank.
This is the same play Google ran with Search more than 25 years ago. Cheaper hardware, faster results, "good enough" wins.
The bet now: the AI race isn't really about who has the smartest model. It's about who can run AI for less.
OpenAI president Greg Brockman said the quiet part out loud recently - the model alone is no longer the product.
Google has been building for this fight since before most of its rivals existed.
Join 350,000+ investors reading Market Briefs every morning and get a 45-minute investing course thrown in as a bonus.
