QumulusAI, Inc. Common Stock

QumulusAI, Inc. reported second-quarter results and reiterated its plan to reach 18 megawatts of high-performance computing capacity by year-end 2026, even as it declined to reaffirm its $300 million annual recurring revenue target. The company, in its first earnings call as a public company, said it signed 21 new direct customer contracts worth $169.7 million in the quarter, bringing total signed contract value to $282.5 million across 40 contracts. Revenue rose 118% to $6.7 million, with compute power contributing $5.6 million, while gross margin improved to 66.6% on GPU activations and curtailment credits. The company reported an operating loss of $7.7 million and a net loss of $22.8 million, which included a $19.2 million noncash loss from convertible notes issuance. Management emphasized that growth is constrained by the availability of powered land and shells, not by demand, and noted that all capital expenditures for the 8 megawatts already sold have been ordered and financed.

QumulusAI has signed a seven-year colocation agreement for up to 3.75 megawatts of data center capacity in metropolitan Atlanta, with a right of first offer on as much as 7 megawatts of additional contiguous capacity at the same site. The contracted capacity is expected to support up to 2,048 NVIDIA Blackwell B300-class GPUs, and delivery is conditioned on the provider completing the transition of the premises with the existing occupant, with capacity expected to be ready for service in the fourth quarter of 2026. The agreement addresses the supply side of a demand picture QumulusAI has been building publicly since early June, during which the company has announced customer agreements valued at more than $246 million in aggregate, including a three-year inference agreement valued at more than $71 million and a separate GPU-as-a-Service agreement with DRW. In July, QumulusAI purchased 1,632 NVIDIA Blackwell B300 GPUs to meet that demand, and Atlanta is also the company's home market, where it established its corporate headquarters in Georgia Tech's Tech Square in June.

QumulusAI has signed a GPU-as-a-Service agreement with global trading firm DRW to supply a dedicated Nvidia Blackwell B300 cluster. The deal carries an initial one-year term with three one-year renewal options, potentially extending to four years, and pushes QumulusAI's total announced customer agreements past $246 million since early June. It follows a recent agreement with an agentic hedge fund and a $71.9 million, three-year deal with an AI inference platform provider, both for Nvidia Blackwell capacity. The company targets $300 million in forward annual recurring revenue for fiscal 2026, backed by executed contracts, expected renewals, deposit-backed reservations, and projected signings. However, the DRW contract's renewal structure and the company's history of net losses and capital-intensive growth underscore execution risks.

QumulusAI shares rose after the AI-infrastructure provider announced an unusual agreement to supply Nvidia Blackwell GPU capacity to an agentic hedge fund, combining conventional compute fees with a share of trading profits. The deal gives QumulusAI potential upside beyond fixed cloud-rental revenue without exposing it to the fund's trading losses, but introduces more variability into the company's revenue model. The unnamed hedge fund will pay market rates for compute consumed while QumulusAI also receives a portion of quarterly trading profits above an agreed threshold. Unlike QumulusAI's recently announced take-or-pay contracts, the agreement has no fixed contract value, making the economics potentially more lucrative but less predictable. The deal represents the first use of QumulusAI's reserve Blackwell capacity under this profit-sharing model, as the company aggressively expands its GPU fleet after ordering 1,632 Nvidia Blackwell B300 GPUs in July and signing more than $124 million of three-year inference contracts earlier this year.

QumulusAI has signed a two-year agreement valued at more than $32 million to supply NVIDIA Blackwell B300 capacity to an AI inference platform provider focused on generative AI applications. The customer runs image, video and other generative workloads on dedicated Blackwell B300 clusters from QumulusAI's hyperdistributed AI cloud, with capacity expected to come online in the fall of 2026. The agreement includes renewal options and adds a two-year commitment of more than $32 million to QumulusAI's book of business. Capacity will be served from QumulusAI's U.S. data center footprint, leveraging its demand-led deployment model across a distributed network of colocation and owned facilities.

Nvidia shares are in focus after QumulusAI secured approval as an Nvidia Cloud Partner, highlighting how third-party providers are using Nvidia infrastructure to bring AI compute capacity online quickly as demand increases. The stock has seen a 3.52% decline over the past 30 days but a one-year total shareholder return of 18.77%. Shares now sit about 14% below one intrinsic value estimate and roughly 49% under the average analyst target. The most followed narrative on Simply Wall St estimates fair value at $339.90 versus a last close of $203.28, framing a wide valuation gap for investors to weigh.

QumulusAI has purchased 1,632 NVIDIA Blackwell B300 GPUs to meet accelerating customer demand for high-performance AI compute. The GPUs will be delivered across 204 NVIDIA HGX B300 systems, representing one of the company's largest single capacity expansions to date, and the purchase was funded primarily through financing arrangements with Technology Finance Corporation and USD.ai. The order also includes an additional 192 NVIDIA RTX PRO 6000 Blackwell GPUs, giving customers workload-matched options spanning training, inference and visualization. The expansion supports more than $124 million in three-year inference agreements announced earlier this year and extends a period of rapid growth in which QumulusAI's deployed GPU fleet has expanded by more than 450% from June 2025 to June 2026.

QumulusAI has been approved as an NVIDIA Cloud Partner within the NVIDIA Partner Network. The neocloud infrastructure provider, purpose-built for the AI computing era, can now work with AI-native companies, enterprises, and machine learning teams to deploy NVIDIA AI infrastructure for workloads spanning model training, fine tuning, reinforced learning, and production-scale inference. QumulusAI rapidly expands available capacity by deploying infrastructure across a distributed network of vetted co-location data center sites, bringing high-performance compute online in accelerated timeframes. CEO Michael Maniscalco stated that becoming an NVIDIA Cloud Partner affirms the company's ability to deliver enterprise-grade NVIDIA compute at hyperspeed, meeting customer demand today while building toward a larger inference future.