WhiteFiber, Inc. (WYFI) stock intrinsic value fair value estimate & owner earnings analysis

WhiteFiber builds and runs data centers for AI computing and rents out NVIDIA GPU capacity as a cloud service. It was carved out of Bit Digital and listed in August 2025; Bit Digital still holds a majority of the votes. Its pieces today: (1) a GPU cloud business, mostly hosted in Iceland, that produced $68.8M of 2025 revenue but lost its largest customer (70.7% of 2025 revenue), whose contract was terminated in 2026 with a one-time $12.3M payment booked in the second quarter; (2) small Montreal colocation sites (MTL-1 and MTL-3, the latter leased to Cerebras at about CAD 1.4M a month for five years from November 2025), $8.9M of 2025 revenue; and (3) NC-1, a former textile plant in Madison, North Carolina, bought for $45M and converted into a 40 MW AI campus leased to Nscale under a ten-year contract worth about $865M including escalators and installation fees, with power passed through. Billing on NC-1 began in the third quarter of 2026 after switchgear delays. Earnings record. The business is too young to have a normal earnings level. Owner earnings (operating cash flow, less working-capital swings, less stock compensation, less upkeep spending) were about $1.2M in 2024 ($0.04 a share) and a loss of $24.3M in 2025 (-$0.77 a share). Because nearly all of the $268M spent on property in 2025 (and $345M more in the first half of 2026) was building new capacity, upkeep spending was set equal to depreciation rather than counting all construction as a cost of standing still. 2025 operating cash flow of $45.7M looks healthy only because customers paid $48.7M in advance; we strip that out. There is no history to anchor on, so value has to be built contract by contract, and only contracts that are signed are counted. What is counted. NC-1: about $62M a year of cash after site costs, rising with the contract escalators, for the ten-year term; after 2036 only a re-let of the shell at a much lower rent is assumed, because the Nscale contract is the only evidence of demand for it. MTL-3: the Cerebras rent for its five-year term, then a reduced rate. New cloud contracts signed in 2026 (Baseten in Ontario, about $165M over three years; Prime Intellect, $108M over three years from 2027; a five-year Paris deployment over $160M; a five-year Iceland renewal of $87.5M): each requires acquiring new GPUs, so the cash credited is what is left after the hardware is paid for, and because GPUs are worn out or obsolete within about five years, these contracts are treated as ending with no assumed renewal. Corporate overhead of about $50M a year (2025 administrative costs were $52.5M including stock compensation) is charged against all of it. Not counted: the 1,500 MW development pipeline, the additional Duke Energy power at NC-1 beyond 40 MW, MTL-2 (on hold) and any future cloud contracts not yet signed; none is secured. The company also needs roughly $150M more in the second half of 2026 to finish NC-1, which every case charges. What it owes. $230M of 4.5% convertible notes due February 2031 (holders can demand repayment in February 2029; conversion at $25.91 is not assumed), about $83M of equipment and bridge loans added in mid-2026, against $56M of cash, plus roughly $100M of services already paid for in advance by customers that must still be delivered without further cash. About $357M of claims rank ahead of shareholders. The January 2026 note issue was paired with a $120M forward purchase of 5.9M of its own shares, so we count about 32.9M shares rather than the 38.8M outstanding. Base case: NC-1 runs to contract, the 2026 cloud contracts deploy on schedule and run their terms, and overhead stays near today's level. The business is worth about $348M today; after $357M of claims shareholders are left with roughly nothing, about -$0.28 a share. The near-term years look strong (business cash of $140-155M in 2028-29) but they are the payback on GPUs acquired in 2026-27, and that cash disappears when the contracts end in 2029-31. Bear case: AI-compute rental prices fall as they did for older GPU generations through 2024-25 (the company already had a cloud contract terminated in December 2025 and lost its largest customer in 2026), one of the new cloud customers - all young AI companies - fails or walks away midway, and NC-1 project financing closes on expensive terms or not at all, forcing dilution. Cash from the cloud book roughly halves and overhead persists. The business is worth less than nothing in that path, and the company would owe about $540M more than it is worth (-$16.35 a share); in practice that ends in a restructuring of the 2029 convertible put. Bull case: everything delivers on time, NC-1 gets its second phase of power from Duke Energy and is leased at similar economics (adding about $55M a year after a ~$360M build), and overhead is held lower. The business is worth about $696M; after claims about $339M, or $10.29 a share. What would make this wrong. The value hinges on two things the record cannot yet show: whether WhiteFiber can earn more on its GPUs than they cost before they become obsolete, and whether NC-1 can be expanded and re-let profitably beyond Nscale. The 2026 contracts are its first test of the first; a second large NC-1 lease would materially lift every case. Previously we valued the company on a projected path to steady profitability; the figures are not comparable. Margin-of-safety price. We are not confident of any earnings level: there is not one full year of profitable operation, the largest customer is gone, every customer is an early-stage AI company, and the hardware it rents out loses most of its value in about five years. The worst plausible outcome - a collapse in GPU rental rates while the 2029 note put falls due - is one the balance sheet would not survive without shareholders being largely wiped out, and a majority owner (Bit Digital) controls the outcome of any recapitalisation. If we are wrong about cloud economics in the favourable direction, shares may be worth $10; if wrong in the other, nothing. The business pays no dividend and burns cash through 2027 in the meantime. With the median outcome near zero, there is no price at which a 20% annual return can be called likely, so no margin-of-safety price is set.