Token Factory

Starting with the word itself

Micky Malka opens his firm's token thesis by pointing out that the word means something different depending on who is using it, to Visa, to Google, to OpenAI, to Anthropic, and then collapses the differences on purpose: all of them are describing the same underlying thing, machine-readable information. A payment network's token and a language model's token are the same object described from two different industries.¹

The factory analogy and its three inputs

From there Malka reaches for an industrial-revolution frame. A company in the token era is a factory: it needs a supply of tokens, meaning information, money, and knowledge, it does something to that supply, and it ships something new. He breaks any delivered product, service, or promise into three required components. Identity answers who this is, and whether they are who they claim to be. Value answers what something is worth and who holds it. Intelligence answers what should happen next.¹ All three, in his framing, are raw materials of the same kind rather than a reasoning layer sitting on top of solved plumbing, which is why he can say in the same breath that money should have entered the AI buildout first and that it did not.

Where the factories sit today

Malka places frontier AI labs as the first generation of token factories, ingesting the open web, video, and the world, and emitting a different set of tokens for everyone downstream to consume. In his own factory terms, the labs are not yet the manufacturers of finished goods, they are the raw-materials suppliers. He names a layer already forming above them, companies such as ElevenLabs, Decagon, and Sierra, building their own token factories that create a service that did not exist before in a way that was not possible.¹ The end state he describes is total: "every single piece of information is to be readable by a machine."

What the frame gets right, and what it leaves open

The framing's distinctive move is treating identity and value as co-equal raw materials alongside intelligence, rather than as plumbing to be solved underneath a reasoning layer, which is not how most writing about AI-driven commerce treats them. It also relocates where margin should accrue: if frontier labs are raw-materials suppliers rather than finished-goods manufacturers, defensibility sits downstream, with whichever company does something specific to the material rather than with whoever mined it. The analogy is also doing more work than it has earned. Calling every company a factory was said about electricity, about the internet, and about cloud computing in turn, and each time the claim predicted less than it seemed to promise, since it never specified what the framing would rule out if it were wrong. Collapsing a payment credential and a unit of language-model representation into one word, token, is rhetorically effective and technically loose, and the three-part division into identity, value, and intelligence is asserted in the interview rather than derived from anything that would show why a company missing one of the three actually fails.