Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
InfrastructureInsight · On RadarX: @rauchg→ original

Vercel CEO: Token demand grows 40% a month, while inference clouds pitch him every two days

Vercel CEO Guillermo Rauch says new inference clouds, which rent out GPUs to run ready-made models, pitch him roughly every two days. Demand for tokens (units of text volume for AI models) in AI Gateway, Vercel's gateway to various models, has been growing at about 40% a month since the start of the year. Rauch expects the number of players to rise first and then consolidate, so teams should avoid tying their product to a single small provider.

AI-processed from X: @rauchg; edited by Hamidun News
Vercel CEO: Token demand grows 40% a month, while inference clouds pitch him every two days
Source: X: @rauchg. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @rauchg ↗
Editor-in-Chief
Key metric #1Inference cloud pitches reach Rauch on average every two days
Key metric #240% a month compounded is roughly 7.5x over six months (our calculation)
Key metric #3ZDR (the provider does not store your requests) and cybersecurity are pressuring smaller players

Imagine a market where a new company offering to rent you GPUs for running neural networks appears every two days, while demand for model responses grows by nearly 40% a month. That is the picture Guillermo Rauch, Vercel's CEO, described on X. Let's go through the terms he uses.

Inference is running an already trained model so it can answer requests. A "neocloud for inference" is a company that does not train its own models. Instead, it rents large numbers of GPUs and sells access to third-party or open models.

A token is the unit used to measure the volume of text a model processes (roughly, a piece of a word), and it is what you pay for. AI Gateway is Vercel's unified gateway: an application sends its requests to it, and it routes them to the right model and provider.

According to Rauch, these neoclouds approach him with offers on average every two days. He stresses that this is not an exaggeration: he listens seriously to every founder and digs into the product. But he admits the flow looks unsustainable.

What "40% a month" means

Alongside this, Rauch cites a figure from his own business. Token demand in AI Gateway has been growing at about 40% a month, compounded since the start of the year, meaning each month is calculated from an already larger base. He says there are no signs of slowing down.

What this means in numbers, we calculated ourselves (this is our own calculation, not Rauch's words). At this rate, volume doubles in a little over two months, grows roughly 7.5 times over six months, and nearly 57 times over a year. Rauch does not disclose absolute volumes, so the only thing you can judge is the speed.

A caveat: this is data from a single platform that itself sells access to models and has an interest in a buoyant picture. Its customers are mainly web developers embedding AI into applications. The overall market may be growing differently.

An "oil rush," not a "gold rush"

To reconcile the flood of pitches with fast demand growth, Rauch offers a framing: computing is becoming a new economy, and tokens are a new unit of energy, like a kilowatt-hour. He talks about a new industrial revolution in which every sphere of life moves onto a new platform. That is why he compares the situation to an "oil rush" rather than a "gold rush": oil is needed constantly, while gold fever is remembered as a passing episode.

This is an opinion, not proof. The logic is still clear, though: once tokens are built into customer support, code writing, search and document work, demand rests on habit rather than fashion.

Why he expects consolidation

In the near term, Rauch expects even more players, followed by consolidation. He names the following reasons:

  • Economies of scale. Large providers keep their GPUs better utilized, so a token costs less.
  • Enterprise requirements and security. He singles out ZDR (zero data retention, a guarantee that the provider does not store your requests and responses) and cybersecurity as hot topics for inference providers. A bank or a clinic is unlikely to choose a small provider without such guarantees.
  • Capital. Buying and running GPUs is expensive, and not every startup will be able to keep raising money indefinitely.
  • Regulation. Rules for AI services are changing, and compliance is not cheap.

What to do in practice

First, do not build your product on a single little-known inference provider, even if it is half the price today. If the market starts to shrink, some companies will disappear or be acquired, and terms will change.

Second, keep an abstraction layer between your application and the provider. Gateways such as AI Gateway, OpenRouter, or your own proxy let you switch models or providers without rewriting code.

Third, ask providers in advance about ZDR, where data is processed, and their certifications. If you sell to businesses, your client will ask these questions anyway.

The takeaway for founders of inference startups is harsher: a low price per token will soon not be enough. They will have to demonstrate security, stability, and financial runway.

For now, this is one executive's forecast, not industry statistics. But when the person whose service carries a large share of AI traffic talks about both explosive growth and inevitable consolidation, it is worth factoring into your provider choices for the coming year.

Story

Key Questions Answered

What "40% a month" means?

Alongside this, Rauch cites a figure from his own business. Token demand in AI Gateway has been growing at about 40% a month, compounded since the start of the year, meaning each month is calculated from an already larger base. He says there are no signs of slowing down. What this means in numbers, we calculated ourselves (this is our own calculation, not Rauch's words). At this rate, volume doubles in a little over two months, grows roughly 7.5 times over six months, and nearly 57 times over a year.

What are the key details on An "oil rush," not a "gold rush"?

To reconcile the flood of pitches with fast demand growth, Rauch offers a framing: computing is becoming a new economy, and tokens are a new unit of energy, like a kilowatt-hour. He talks about a new industrial revolution in which every sphere of life moves onto a new platform. That is why he compares the situation to an "oil rush" rather than a "gold rush": oil is needed constantly, while gold fever is remembered as a passing episode. This is an opinion, not proof.

Why he expects consolidation?

In the near term, Rauch expects even more players, followed by consolidation. He names the following reasons: Economies of scale. Large providers keep their GPUs better utilized, so a token costs less. Enterprise requirements and security. He singles out ZDR (zero data retention, a guarantee that the provider does not store your requests and responses) and cybersecurity as hot topics for inference providers. A bank or a clinic is unlikely to choose a small provider without such guarantees. Capital.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…