Guides for owners
- Do you own what your AI tools produce?
As of 4 November 2026, your vendor's terms and US copyright law answer two different questions. The terms say the vendor claims nothing in the output. Copyright protects the human expression in a work, not material a system generated without enough human control, so some of what you produce carries no copyright at all.
- Why do most AI pilots never reach production?
As of 3 November 2026, the published research points at one cause: pilots are scoped to be shown, not to be run. Tools get evaluated far more often than they get deployed, budgets follow the functions that are easiest to measure, and projects built with an outside partner reach production about twice as often as ones built in-house.
- Buy GPUs or rent them? What the decision turns on
As of 2 November 2026, rent while your demand is still moving and buy once a customer has signed for the capacity. Rented capacity is published by the hour and stops costing you the hour you stop using it. An owned cluster carries a fixed payment every month, and lenders will not underwrite on-demand rates.
- Who lends against GPUs, and which kind of lender fits your cluster
As of 1 November 2026, GPU clusters are financed by banks, private credit funds, equipment lessors, small-cluster programs, bridge lenders and onchain credit. Order size and whether a customer has signed decide which one will look at you. Under about $20 million, start with the small-cluster programs, the equipment lessors, and banks that run a smaller equipment book.
- AI-data buyer programs compared: micro1, Scale AI, Handshake AI, Mercor
As of 29 October 2026, micro1, Scale AI, Handshake AI and Mercor each publish a program for company operating records. Stated ranges run from about $10,000 to over $4 million, with minimums of 20 to 40 employees. Each range is that program's advertisement, not an agreed price, and the right buyer depends on the company's data, rights and plans.
- Can you license data your customers created? A rights checklist
As of 29 October 2026, being able to export a record does not mean you may license it. Customer contracts, employee privacy, the terms of the systems you export from and any earlier licence decide it. This checklist works through each check in the order a buyer's lawyers will read them, and names the public sources behind each.
- AI-data buyer tracker, November 2026 edition
As of 29 October 2026, four AI-data buyers publish per-company programs. Ranges run from about $10,000 to over $4 million, with minimums of 20 to 40 employees. One buyer announced a $1 billion, 12-month commitment on 9 October 2026. This note tracks what each page says this month and what has changed.
- What goes in a data room for licensing company data to AI labs
As of 29 October 2026, buyers of company operating data ask for records from the systems you already run on, plus evidence you hold the rights to license them. A data room therefore needs three parts: an inventory of systems and volume, proof of rights and consent, and a de-identification plan agreed before any sample leaves your systems.
- How lenders underwrite a GPU cluster loan: what they check first
As of 29 October 2026, lenders financing a GPU cluster read the customer contract first, then your own cash, then the quotes. They set terms against those: an advance of 50 to 80 percent of the equipment cost, a debt-service coverage covenant of 1.15 to 1.25 times, and a rate priced by the customer's credit.
- Lender terms seen in filings this month: October 2026
As of 29 October 2026, this month's 8-K filings and public releases show a range of lender terms: senior revolvers at SOFR plus 1.25 percent, warehouse facilities routed through bankruptcy-remote subsidiaries, large term loans with JPMorgan as agent, and a receivables-backed small loan at a fixed fee. Each term below links to the filing it came from.
- Licensing your company's data to AI labs: what buyers say they pay
As of 29 October 2026, several AI-data buyers publish programs to license a company's operating records. Their stated ranges run from about $10,000 to over $4 million per company, usually for firms with 20 to 40 or more employees. These are advertised ranges, not agreed prices, and every deal turns on whether you hold the rights to license the data.
- Loan, recap or sale for a 25 to 75 person company: a side-by-side
As of 29 October 2026, a 25-to-75-person company has three main capital paths: a loan, a minority recap that keeps the owner on, and a sale of the business. Each reads the same contracts, concentration and owner-dependence points differently, and the right one depends on what you want next and when.
- Loan or recap for a 25 to 75 person company: what lenders read
As of 29 October 2026, lenders to small services companies lend against receivables, contracts and equipment, and check that cash flow covers the payments. Minority investors who keep the owner on look at the same concentration and owner-dependence points as buyers. Frontier Desk tracks hundreds of lenders and builds each profile from the firm's own published criteria.
- Who buys cybersecurity services firms, and what they check
As of 29 October 2026, 25 of the 365 buyers Frontier Desk tracks with a completed services deal in the last 180 days bought cybersecurity firms, counted from their own announcements. They range from private-equity platforms to services firms that buy others. Each marks a firm down for customer concentration, owner dependence and resale-heavy revenue.
- Selling a services firm AI is changing: the two kinds of buyer
As of 29 October 2026, buyers of AI-exposed services firms split in two. Some want a company that software already runs and will pay for the lower costs once a full year of earnings shows them. Others want to buy it as it is and make the change themselves, so they price the work people still do.
- Who buys MSPs and cybersecurity services firms, and what they check
As of 29 October 2026, buyers of MSPs and cybersecurity services firms fall into three groups: private-equity-backed platforms rolling up regional providers, AI-enabled roll-ups buying firms whose service desks still run on people, and lenders that fund against managed services agreements. Each reads the same contracts and prices them differently.
- Forward-deployed engineers: what the work is, how it is measured
As of 28 October 2026, a forward-deployed engineer works inside your company and automates specific processes, such as request triage, document review, scheduling, billing and reporting. Frontier Desk starts with a fixed-fee assessment in weeks one and two that measures each process, ranks changes by margin effect and prices the first build, then measures the result against that baseline.
- How AI changes what an IT services, MSP or accounting firm is worth
As of 28 October 2026, AI changes a services firm's value through its contracts. On per-user, per-device or monthly fees, hours software takes off the work stay as margin. On hourly work the saving passes to the customer unless the contract changes. Buyers then price it differently: cash-flow buyers pay only for earnings that show it.