Why Enterprise AI Spend Grew 13x | Harpreet Arora artwork

Why Enterprise AI Spend Grew 13x | Harpreet Arora

MTS

October 19, 2026

Vercel's agentic infrastructure lead Harpreet Arora reveals enterprise AI adoption trends from Vercel AI Gateway data, including open-weight model usage reaching 25% of enterprise spend, Jevons paradox driving a 13x total spend growth despite falling token costs, and why frontier lab spending...

0:00

Harpreet Arora

Since December, price per token has reduced by almost half. So it's like, it is getting cheaper, for sure. But the overall usage has gone up 26 times.

And so therefore, the total spend has increased 13x.

0:15

Sophia

Hello everyone, and welcome back to MTS. I am here live at Vercel Ship, where I am joined by Harpreet Arora from Vercel, who leads a Gentic infra at Vercel. So welcome to MTS.

0:27

Harpreet Arora

Thank you for having me, long time listener, first time caller.

0:30

Sophia

Long time listener, first time caller. Those are my favorite. Well, I really wanted to thank you for joining us today. I think one of the interesting things about your background, which our audience is going to be really curious about, is a lot of your work with Vercel's AI Gateway, you get to understand about enterprises, how they're using models, what they're calling them for. And I want to get into some of these really interesting insights.

0:51

Harpreet Arora

For sure. Yeah, yeah. I often say that we're kind of in a privileged position, not only because of the enterprise customers and the developers that we have on our platform, but we also have strong partnerships with the labs, as well as the open weight providers. So we're sitting at the intersection of this development that's happening. And so we have a unique perspective, which is really helpful.

1:13

Sophia

I want to get into this unique perspective. Yeah, you get to talk to the labs, you get to understand open weight model usage, you get to talk to the customers and enterprises. So what do you think is maybe one of the top insights that the mainstream public doesn't totally know about what's happening?

1:28

Harpreet Arora

Yeah, for sure. I mean, if you're in AI with deep in AI, you'll find it hard to keep up with everything that's happening, right? Like just this week, we had Gemini coming out with Argon, we had Meta drop an open weight model this morning, we had Meta also drop Wonder. And so the frontier keeps on changing so quickly that it's hard to keep up. And on the other hand, we're kind of seeing open weight models come about, right? So between all of that, you're seeing customers realizing, well, there's a lot of choice, but at some point, the choice gets overwhelming. And how do you make infrastructure decisions in a world that's fluid? And so that's where we come in with Vercel Gateway, right? And so that's why we built AI Gateway. Because as a customer and as a developer, you want to have choice. You want to have the flexibility to try the latest thing.

You also want your infrastructure to be reliable. And so when something like that, a problem like that emerges, you need infrastructure from a provider like Vercel, which can provide you that flexibility as well as reliability.

One of the things that we just published our October production index blog. We published that monthly. It covers all the insights that we see each month from our customers as well as from the partners on the lab side. And one of the most interesting things you will see is open-weight models for the first time have started taking dollar share, which is not something that they were doing before.

3:05

Sophia

Open-weight models for the first time are starting to see dollar share. What do you mean by that?

3:09

Harpreet Arora

Yeah. So if you look at the trend from June of this year, open-weight models started picking up a lot of fraction. Yeah. And so you started seeing token volume go to open-weight models. And so we saw that steady acceleration. In August, the share flipped. And so it was the first time that more than half the tokens on AI Gateway were going to open-weight models.

But open-weight models, because they were cheaper, the spend was still concentrated on frontier close-weight models, right? But in October, we saw more than a quarter of the dollars actually going to open-weight models as well. And that's the first indication that enterprises are actually using these in production, which is where we sit, right?

3:51

Sophia

Yeah. And I would actually argue it's because of these gateways, it's because of these routers that enterprises are using more open-weight models. Actually, trying to change your internal system to use open-weight models just on your own, difficult. But then if you're going to a router and you get to share what you care about, you share around cost, it then can route it for you differently.

4:11

Harpreet Arora

Exactly.

4:12

Sophia

I am curious though, because when it comes to open-weight models, the majority of the strongest open-weight models are Chinese models. And so, what are you noticing about enterprises using maybe more, an increase in growth of how they use open-weight Chinese models?

Read the other 9 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000797096068