Alpha Compute's $47 Million Seller Financing Buys Cheap Power, Not a Funded Data Center

In August, Alpha Compute announced a binding term sheet to buy roughly 1,800 mineral acres of Marcellus shale gas rights in northern Pennsylvania for $55 million — the land under a planned 200-megawatt data center campus that could one day expand to a gigawatt. The detail that got the headline is that $47 million of that purchase is seller financing. Read that number the right way and it tells you almost everything about what Alpha Compute is buying, and what it is not.

Seller financing sounds like money raised, but it is the opposite: it is money deferred. Instead of paying the landowner $55 million today, Alpha Compute pays about $8 million in cash and converts the seller into a creditor who waits for the rest. This is not a funded project — it is a down payment that spreads the tab for the cheapest and least expensive part of the whole venture. A seller willing to hold that paper is effectively betting alongside the buyer, which is a mild vote of confidence. But it is also leverage sitting on the balance sheet of a company that has almost no revenue to service it, and it buys only the dirt — not the data center.

That dirt is worth noticing, because the economics of the underlying asset are genuinely attractive. The deal couples the gas rights with on-site power generation and targets a power offtake of $0.0585 per kilowatt-hour — under six cents, far below the ten-to-fifteen-cent power costs that constrain most inland AI builds. In an era when the bottleneck for AI is no longer chips but the grid, cheap, self-produced energy is the scarce asset. Securing a Marcellus site and generating electricity on your own land sidesteps the multi-year utility interconnection queues that are the industry's binding constraint. This is the part of the story that is real, and it is why a $500 million AI campus planned to start operating in the third quarter of 2027 is more than vapor.

The problem is everything that comes after the land. The $47 million in seller financing covers the real-estate transaction, not the build. Alpha's own disclosure puts the remaining construction cost — the data center plus electrical connections — at roughly $500 million, spread over several years. That is where the actual capital requirement lives, and on that, there is no committed financing disclosed. The deal remains subject to due diligence, permitting, and final agreements, and the company itself frames the purchase as an option it expects to close after a 60-day diligence period.

So put the numbers side by side. To build 200 megawatts by late 2027, Alpha Compute needs the better part of $500 million in construction capital. Its trailing revenue for the last fiscal year was $97,000. It recorded a net loss of $38.6 million, and that loss widened substantially from the prior year. Total assets stood around $79 million at the start of June, against $44 million of liabilities — $34 million of it already tied up in GPU and data center leases — leaving about $35 million of equity. The seller financing does not fund the project; it refinances the cost of the dirt with the very same seller whose land it is buying.

This is the judgment that separates Alpha Compute's claims from its delivered results — and this is where I hold the line. The company's only operating asset is a 504-GPU NVIDIA B200 deployment in a hydro-powered Canadian facility, and even that site is not yet self-funding: it took in about $1.1 million in cash while spending roughly $1.25 million. The Pennsylvania campus is a signed term sheet, which is a claim about what might be built, not a result about what has been delivered. I apply that standard because in this market every micro-cap with a narrative is promising a gigawatt; the ones worth owning are the ones that can cite revenue, utilization, and positive unit economics from hardware already turned on. Alpha Compute, so far, has cited the contract it signed and the land it optioned.

None of this makes the deal a fraud, and none of it makes it a certainty. Cheap power is a real moat in the current cycle, and if Alpha Compute can convert an under-six-cents energy position into a 200-megawatt campus that it then rents out at GPU-service economics, the optionality here is substantial — the stock has already swung from pennies into the low single digits this year on far less. But the $47 million headline is deferred payment, not a war chest, and the decisive gap is the $500 million it will take to build what the land only promises. I would treat ownership of ALP as ownership of that specific option: real, early, and priced on the expectation that a company with $97,000 of trailing revenue and a six-figure cash burn at its only live site finds the capital and the customers to finish what it started. Watch whether the construction financing arrives and whether the Canadian pilot turns cash positive — those are the two facts that would tell us this campus is a bet being executed rather than a bet being repeated.