At Anthropic’s developer conference last week, I ran into Guillermo Rauch, CEO of AI startup Vercel. We started chatting, and the first company he mentioned wasn’t Anthropic; it was Google.
Overall, demand for AI is off the charts, he said. But Google’s models, in particular, are in demand amongst Vercel customers. Rauch even mentioned that he had to call a top Google executive recently to ask for more Gemini tokens, the core unit of AI usage.
While all the talk lately has been about Anthropic and OpenAI, Google’s Gemini has been quietly gaining ground, he said.
You can see this on Vercel’s AI Gateway, which lets companies connect their apps to different AI models through a single system. It’s mainly used by AI startups, software companies, and enterprise product teams that run AI features like chatbots, coding assistants, search tools, and copilots.
Check out this chart. In March, Anthropic models led, based on the number of tokens (traffic) handled by Vercel’s AI Gateway. In early April, Google’s Gemini 3 Flash model jumped into the lead and has stayed there. And this is even before Google’s big annual conference, I/O, which kicks off next week. The company is likely to unveil a host of more capable AI models, tools, and features. (I’ll be there, so tune in next week).
Vercel
Gemini Flash is less powerful than the full Gemini 3 model, but it’s faster and cheaper to use. That makes it popular, especially among corporate customers of Vercel.
“Enterprise teams tend to pick Gemini Flash and Claude Haiku, the smallest, fastest, cheapest models each lab ships,” Rauch told me. “Flash in particular is seeing strong B2C adoption because it doesn’t hallucinate much, uses tools effectively, and it’s fast and affordable.”
In AI, the answers are never simple, though. There are other ways to measure success, such as how much users spend on models
“I’m often asked which lab is ‘winning’, but what we see in production looks nothing like the benchmark leaderboards,” Rauch said. “AI Gateway reflects a variety of models winning different use cases.”
By token use, Google clearly won in April. But, based on dollars spent, Anthropic led with 61% share. “Some models win on inexpensive, high-volume traffic. Others win on expensive, quality-critical work. They’re solving different problems,” Rauch explained.
OpenAI’s spend share tripled from March to April (going from 4% to 12%) due to the launch of its new GPT-5.4 and 5.5 AI model series. Google climbed from 8% to 21%, based on dollars spent, as Gemini Flash usage scaled.
“A snapshot from any one month doesn’t predict the next,” Rauch warned.
That’s especially true with Google I/O just around the corner.
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at [email protected].
