Microsoft CEO: Most token traffic is a commodity. Comments: open source for volume, closed source for profit

Satya Nadella made a judgment during the 20VC interview: in the future, the vast majority of token traffic will be commoditized.

His exact words:

"My suspicion is that there will always be a market for frontier. There will always be some new thing. Science may be the next thing that frontier is being set at. But I think predominantly token usage will be commodity token usage."

There will always be a market for frontier models; there will always be something new. Science may be the next frontier. But most token usage falls into the commoditized segment.

A token is the smallest unit of text that a model processes; token traffic roughly equals actual model usage.

This conversation took place on Harry Stebbings' podcast 20VC. Harry has talked in a tweet about where he started: at 18, in a bedroom in London, building from nothing, with not a single acquaintance in the tech world. The guest he brought on this time took Microsoft's market cap from $300 billion to $3.9 trillion.

The interview is not only about open versus closed. There are several points in the full episode worth pulling out separately: why you should assume all AI models have already been compromised, why chain-of-thought is unreliable, how private evals can become a company's most valuable IP, and whether $100 billion-scale training investment is sustainable.

The question Harry posed in this tweet was: will most tokens flow through open models or closed models?

AI researcher Nathan Lambert's reply was just one sentence:

"most tokens through open models, most of the profits through closed models"

Most tokens go through open-source models; most profits go to closed-source models.

This sentence deserves unpacking. Open-source models carry everyday traffic, while closed-source models take the profits—similar to the early internet pattern in which open-source software became widespread and commercial companies made the money. The two paths are not necessarily mutually exclusive; they look more like layers.

A commenter also pointed out a detail: Nadella shifted the question from 'open vs. closed' to 'frontier vs. commodity.' That shift is itself a stance. Microsoft is, on one hand, an important compute provider behind OpenAI, and on the other hand, it puts open-source models into the Azure catalog. Whichever path wins, the infrastructure business cannot be escaped.

Some also raised the data sovereignty angle: for enterprise customers, the real question is not open or closed, but where the data goes. If data cannot leave the jurisdiction, open source naturally wins. For ordinary users, the future may be running open-source models locally on every device—no servers, no worries.

There was one more comment worth pondering:

"Whatever my competitors have is commodity, whatever we have is moat."

What competitors have is a commodity; what we have is a moat.

Whether a model is a commodity or a moat depends on whose hands it is in.

Listen: YouTube · Apple Podcasts · Spotify