The neocloud business model fits in one line: buy GPU servers with debt, lease colocation space, sell GPU-hours. Profit is the hourly rate minus depreciation, power and facility cost, and interest. What the one-liner hides is how the costs behave. Almost all of them are fixed. Depreciation runs whether the machines are earning or idle. Interest is contractual. Colocation is leased by the rack. The marginal cost of selling one more GPU-hour is close to zero, which means the entire profit-and-loss statement is levered to a single operating variable: utilization.
A reference case makes the shape visible. Take a leveraged 1,024-GPU H100-class cluster at typical 2026 rates and costs. At 55% utilization it burns about $333,000 a month. At 70% it breaks even. At 85% it generates about $336,000 a month. The distance between distress and health is thirty points of utilization on the same hardware, the same contracts, and the same balance sheet. No other input — rate, power price, staffing — moves the outcome anywhere near as fast within a quarter.
This is why the reported profitability of GPU clouds deserves a suspicious read. The largest operator discloses an adjusted EBITDA margin around 60%, and still reports a net loss of over a billion dollars once roughly $2.45 billion of depreciation and $1.2 billion of interest are counted. Neither number is wrong. They are answers to different questions. Adjusted EBITDA answers whether the operations cover their cash operating costs. Net income, and more precisely cash after debt service, answers whether the business survives its own capital structure. For a lender only the second question matters, because the loan is repaid from cash after debt service, not from adjusted anything.
Depreciation policy is where the two answers are quietly reconciled, and it is a solvency lever hiding in an accounting footnote. Public operators book GPU servers over four, five, or six years depending on the house view. On a $300,000 server, six-year straight-line is $50,000 a year; four-year is $75,000. Same machine, same market, 50% more annual cost. An operator on a six-year schedule can show equity and EBITDA that a four-year schedule would erase. The underwriting rule that follows is mechanical: restate every borrower's economics on a four-year schedule before believing them. If the equity disappears under restatement, the six-year book was hiding it.
The second rule follows from the utilization bridge itself. If breakeven sits near 70%, then debt service coverage must be tested at or below 70%, not at the borrower's forecast. A projection showing 90% utilization is a sales document. The credit question is what happens at 65% with the anchor customer renewing late. This is also why utilization telemetry belongs in the covenant package rather than in a quarterly PDF: per-GPU utilization, uptime by asset, and revenue-linked usage, verified continuously. Lenders who accept cluster averages self-reported at quarter-end are underwriting a spreadsheet.
Utilization interacts with price in one more way that matters for India. Selling more hours at any price is not the same as selling financeable hours. An operator can hold 85% utilization by dumping capacity into subsidised or spot tiers at the bottom of the price waterfall, and the utilization chart will look excellent while the margin per hour quietly falls below the level that services debt. The honest metric is utilization at contracted rates, with the contract quality inspected separately. High utilization on bad revenue is how a cluster looks busy all the way into default.
For Indian operators preparing to raise debt, the practical checklist is short. Know your true breakeven utilization on a four-year depreciation lens. Keep verifiable per-GPU telemetry from day one, because lenders will want months of history and it cannot be reconstructed. And when the model is stress-tested, let the stress case use the rate at the bottom of your realistic waterfall, not the rate on your public price card. The operators who clear those three tests are rarer than the market believes, which is precisely why they get financed.