July 08, 2026

Vercel’s CEO Just Declared the Single AI Vendor Model Dead

Single-vendor AI contracts are falling out of favour, according to the lead of one of the infrastructure companies best placed to witness it. Guillermo Rauch, CEO of cloud platform Vercel, told TechCrunch this week that businesses have largely stopped picking one AI lab and building everything around it. This is a habit that defined much of last year’s AI spending.

Vercel’s own numbers back the claim up. The company now handles six million deployments a day, roughly half of them triggered by coding agents. It pushes more than a trillion tokens through its AI gateway daily. That’s not remotely a small sample size to be making sweeping statements from.

Rauch commented:

“Last year there were a lot of people picking one lab partner — saying they would build everything on OpenAI or Anthropic. Now they’re saying, I understand how this all works — model, harness, data platform, sandbox, gateway — every piece is plug and play.”

Rauch’s point is that the AI stack has evolved from being one decision to becoming several. Model, harness, data platform, sandbox, and gateway. Each of these factors are now treated as their own swappable part, rather than bundled together under one vendor’s name. He pointed to Gemini’s growing usage as evidence, arguing that companies are chasing price and performance rather than sticking with whoever they signed with first. Open-weight models such as DeepSeek and GLM-5.2 are creeping in too. This is to mostly tackle the cheaper, less glamorous work that doesn’t need a frontier model to do it.

None of this is Vercel being neutral, exactly. A world where companies mix and match AI providers is a world that desperately requires infrastructure to manage the mixing. This happens to be exactly what Vercel sells. That doesn’t make Rauch incorrect, however.

Why the Timing of Rauch’s AI Comments Makes Sense

This development scans with something more prominent than one CEO’s observation. Companies spent last year throwing AI at everything to see what stuck. This year, the bills have arrived, and finance teams are probing what all that spending actually bought.

Coinbase is the clearest public example. Chief executive Brian Armstrong said on X in late June that the company had cut its AI spending nearly in half while usage kept climbing. They did so not by restricting access, but by discreetly changing which models workers got by default. Armstrong said:

“Engineers can choose any model they want, but defaults matter.”

91% of Coinbase’s staff, it turned out, had never come close to hitting their old usage limits anyway. So rather than tightening the leash, the company rerouted routine tasks to cheaper open-weight models, including GLM 5.2 and Kimi 2.7. This left the expensive ones for jobs that actually needed them. A jump in cache reuse from 5 to 60 per cent did roughly as much for the bill as the model switch itself. It’s a routine fix masquerading as a strategy redirection, and it appears to have worked.

What it Could Signal About Vendor Lock-In

For anyone buying AI for their business, the practical upshot is smaller than the headlines suggest, but still considering. Contracts written around a single AI provider are starting to look more like an unnecessary constraint. This is especially striking once a vendor’s own platform starts routing between models behind the scenes anyway.

That’s already happening inside plenty of customer-facing software, whether or not it’s advertised. It’s worth asking directly, next time a vendor renewal comes up, exactly which models sit behind the tools you’re paying for, and how much choice you actually have if one of them stops being the best or cheapest option.

The open-weight models saving companies like Coinbase money aren’t without complications of their own, though. A couple of the ones gaining ground fastest are Chinese-made and drawing regulatory attention in the US, which is a separate conversation worth having before anyone routes customer data through them casually. Essentially, though, the broader shift Rauch is describing doesn’t necessarily need resolving today.

Customer Experience