Go.AI raises $85M selling AI with no token meter

A company just raised $85 million on the argument that the token meter is the problem. Go.AI announced its Series A today: on-premises AI hardware and software for regulated organizations, sold at a fixed fee with no per-token billing. The on-prem AI pricing model is the pitch, and it is worth reading even if you will never buy an appliance.

What actually happened

The $85 million Series A was led by Updata Partners, with existing investors GFT Ventures and LAUNCH participating. Total raised is now $90 million. The Chicago company rebranded from Go Abacus earlier this month.

The numbers in the release: more than 200 customers, upwards of 12.5 million queries a day, annual recurring revenue up more than eightfold year over year, and — the line that stands out in a 2026 AI release — profitable. Demand is concentrated in financial services, healthcare, aerospace and defense, and manufacturing. The products are The Go1, an on-prem hardware and software unit, and Go.OS, the software layer on top.

Carter Griffin of Updata described the thesis plainly in the release: software and hardware that "live inside the customer's secure environment." Per fintech.global, no proprietary customer data leaves for a third party, and the appliance is sold on a fixed fee rather than a per-token meter.

Why fixed-fee AI pricing matters for your business

You are not buying a rack-mounted AI appliance this quarter. That is not the point. The point is that a buyer with money and a compliance department looked at usage-based AI billing and paid a premium to make it stop — and the reasons they did apply to a ten-person shop just as hard.

A token meter makes your cost a function of your customers' behavior. A busy month raises your bill at precisely the moment you have the least attention to spend on it. Worse, it makes the good outcomes expensive: the support agent that handles more tickets, the assistant that drafts more quotes, the pipeline that processes more documents. Every success shows up on the invoice.

Fixed-fee pricing inverts that. You buy a ceiling and then use it. The trade is that you pay for the ceiling during slow months too, and you own the capacity planning.

Here is what to actually do with this, without buying hardware:

Convert your token bill to a fixed number. Take last quarter's AI spend, divide by months, and ask what you would pay for that as a flat rate. If the number makes you flinch, you have been absorbing variance you never priced.

Find your most expensive success. Which workflow costs more the better it performs? That is the one to move to a cheap tier, a cached prompt, or a local model — before volume makes the decision for you.

Ask where your data sits, in writing. "We don't train on your data" and "your data never leaves your environment" are different sentences with different contract implications. If you handle health, financial or client-confidential records, get the second one or get a DPA that reads like it.

Price the exit. On-prem is one answer to vendor lock-in. Portable, self-hostable open-weight models are another, and they cost a lot less than an appliance. The failure mode is having no answer at all.

Sources: Go.AI press release via BusinessWire, fintech.global.