A billion dollars for the paperwork
micro1, a San Francisco data lab, said on 9 October that it will spend $1bn over the next 12 months acquiring and licensing enterprise operational data, with the capital provided by Citi and Hercules Capital. The release does not say how that financing is structured, and no valuation or equity figure is attached to it.
The money runs through what the company calls its Company Data Partnerships programme. The stated purpose is not to add more text to a pre-training corpus. micro1 says it takes the data in de-identified form and builds reinforcement-learning environments from it — simulated workplaces in which AI agents can practise the sequences of steps that make up real jobs.
What it is buying
The company’s own programme page lists what it wants: standard operating procedures, process documentation, internal knowledge bases, workflows, templates and other business-process artefacts, plus CRM data, project histories and quality-assurance processes. Two further categories are less about documents than about judgement — how teams evaluate information and reach operational decisions, and human feedback on AI outputs in real business settings.
The examples it gives run across industries: source-code documentation and bug tracking in engineering, reconciliations and approval chains in finance, ticket workflows in customer support, playbooks and lifecycle documentation in sales, contract review processes in legal, fulfilment and inventory procedures in logistics. The common thread is the recorded residue of work rather than the output of it.
Eligibility is narrower than the headline figure suggests. micro1 says it is looking for operationally mature companies with more than 30 employees and established documentation, and that it currently prioritises US companies, followed by other Western markets.
The price list
Compensation is set out in three tiers on the programme page, “based on the quality, uniqueness, and value of the data provided”: more than $100,000 for a qualified enterprise data partnership; more than $500,000 for large-scale operational datasets or ongoing participation across multiple teams and business functions; and more than $1m for highly unique proprietary operational data with significant value for frontier AI development.
There is also a referral scheme paying up to $50,000. A four-step process — assessment, evaluation, a discovery call, then an agreement — precedes any transfer, and micro1 says the scope, security standards, anonymisation and redaction requirements and internal approvals are all defined before a company participates. Personally identifiable information, it says, is removed where relevant and access is restricted to a limited number of authorised staff.
What the public terms do not settle
The published material is clear about what micro1 wants and what it will pay. It is less clear on three questions a general counsel would ask first.
The first is whose data it actually is. CRM records, support tickets and project histories describe customers, suppliers and employees as much as they describe the company selling them. De-identification reduces that exposure; it does not by itself resolve the contractual and privacy rights attached to the underlying material.
The second is who ends up using it. The public terms do not name the labs or model developers that will train on the environments built from a company’s data, and do not state whether a partnership is exclusive.
The third is what happens afterwards — whether a company can withdraw, and what obligation attaches if the data leaves the arrangement.
None of that makes the programme unusual: data licensing deals are routinely private, and the specifics live in the contract rather than on the marketing page. But the scale here is new. A billion dollars aimed at a year of corporate paperwork is a bet that the bottleneck for agents is no longer text on the open web but the record of how work is done inside firms — and that companies will sell it.
What to watch is whether any named enterprise signs publicly, and whether a customer whose records were included objects.
