Investors see utilisation, inventory, power and the services book on a regular cadence. No vanity metrics. The same rate card the public sees is the one the fleet earns on.
The thesis is simple: hardware is the asset, rental income is the yield. We underwrite each GPU purchase against utilisation we can already forecast, then rack it in a site with a power contract we hold. If raising stopped tomorrow, the cards would keep earning. That is a different bet than funding a cloud bill paid to someone else.
$9.4M has been raised and deployed into inventory. 312 GPUs owned outright across both sites. The close is rolling, not a fictional vintage. Proceeds go to GPUs, colo and power, plus a small research slice part-funded by rental margin. We do not raise to buy market share.
Returns are rental margin minus power, support and research, recycled into the next tray. On-demand, reserved, farm and spot all hit the same cards; reserved and bare metal smooth the floor. Services retainers are a separate P&L so people revenue cannot hide empty racks. Depreciation is modelled before purchase.
Governance is utilisation, inventory, power and the services book on a regular cadence. The public rate card is the one the fleet earns on. Applicable local law. Twenty-one people. Minority positions in operators outside infrastructure sit in a separate sleeve - if what you actually need is GPUs or a retainer, that is Rentals or Services, not this page.
Counts by SKU and site, utilisation, reserved vs on-demand mix, farm share.
Rental margin, retainers, research spend. Separated so one does not hide the other.
Material contracts, power, and any operator minority. No surprise related-party compute.