QumulusAI signs US$240.9M GPU contracts with December capacity handoff
Two unnamed inference customers extend earlier deals, with QumulusAI targeting capacity turnover in December 2026 and early 2027.
Julian Okafor
US$240.9 million in signed GPU contracts faces its first delivery test in December 2026, followed by a second in early 2027. QumulusAI Inc. NASDAQQMLS, the Atlanta provider of rented AI computing capacity, said on Sept. 29 that it signed two three-year take-or-pay agreements covering more than 2,000 Blackwell B300 GPUs for unnamed inference customers.
Shares rose 7.58 per cent to US$6.53 in Nasdaq trading, according to NASDAQQMLS at 4 p.m. EDT on Oct. 2. The Sept. 29 release did not report an upfront payment under either agreement.
The larger agreement carries a stated value of US$211.9 million, with capacity targeted for customer turnover in early 2027. QumulusAI put the second at US$29.1 million and expects to turn over that cluster in December 2026. Both amounts are the company's reported contract values over three years; neither is a statement of revenue already earned.
The company's Aug. 25 results recorded US$6.7 million in revenue and a US$22.8 million net loss for the quarter ended June 30. The Sept. 29 commitments therefore set up a delivery test later in 2026 and in early 2027, after that reported quarter.
Both Customers Had Earlier Contracts
The larger customer had signed a US$71.9 million, three-year agreement on July 28, according to the Sept. 29 release. Its combined commitments with QumulusAI now exceed US$283 million. The smaller customer signed a US$32 million, two-year agreement on July 23, lifting that account's stated commitments past US$61 million.
The Sept. 29 release also lists extension options separately from the initial contract value. The larger deal has two one-year options worth US$70.6 million apiece, while the smaller has a two-year option worth US$19.4 million. QumulusAI says exercising every option would lift the new agreements' potential value to US$401.6 million over five years; the options may never be exercised.
"Inference platforms are planning capacity years out now, and they are backing those plans with cash. That lets us place GPUs against signed contracts rather than against forecasts," QumulusAI chief executive Michael Maniscalco said in the Sept. 29 release.
The Sept. 29 announcement does not specify a payment schedule or quantify any advance payments under these two agreements. It also does not identify the customers or attach their contracts, leaving the disclosed take-or-pay label and aggregate value as the public account of their terms.
A September Handoff Activated An Earlier Deal
QumulusAI said on Sept. 10 that it had completed a Philadelphia deployment and turned the capacity over to a customer on Sept. 1. That handoff activated an earlier US$18 million, two-year take-or-pay agreement announced July 22. The new December and early-2027 targets need the same change from contracted to operating capacity.
The Aug. 25 results put the deployed fleet at 3,088 GPUs on June 30, up from 952 at the start of the quarter. They also reported eight megawatts of capacity under executed lease and colocation agreements at quarter end. The Sept. 29 announcement does not name a site or give a power requirement for its two new clusters.
In a July 20 analysis, Brendan Burke, research director at Futurum, called the gap between QumulusAI's operating base and its much larger capacity roadmap the central execution question for investors. He said the operating test was how much power the company could energize and how much revenue each energized megawatt produced. His assessment predates the Sept. 29 contracts, but their planned handoffs give that test nearer dates.
GPU Financing Has A Peer Comparison
SharonAI Holdings Inc. NASDAQSHAZ, an Australian AI cloud operator, said on Oct. 1 that it entered a US$356 million debt facility secured against GPUs and associated cash flows. Its stated use is compute infrastructure dedicated to customer contracts. Mugglehead's earlier report on SharonAI's facility examined how that borrowing is structured.
Read more: SharonAI Holdings enters US$356M GPU facility in two tranches
QumulusAI's Sept. 29 release does not disclose a GPU purchase price or a financing plan for the newly contracted clusters. Its own risk statement names equipment delivery delays, the availability and cost of power and network connections, and access to financing as factors that could change the expected outcome.
The first dated test is December 2026, when QumulusAI expects to turn over capacity under the smaller US$29.1 million agreement. The larger US$211.9 million cluster is targeted for early 2027; subsequent reports will show whether those customer handoffs produce revenue.
