QumulusAI signs DRW deal for NVIDIA Blackwell B300 GPU capacity under GPU-as-a-Service plan

QumulusAI, a company positioning itself as infrastructure for the AI computing era, said it has signed a GPU-as-a-Service agreement with DRW for access to NVIDIA Blackwell B300 GPU capacity. The announcement was made on August 11, 2026.

Under the GPU-as-a-Service arrangement, QumulusAI provides customers with GPU capacity through a managed service model rather than requiring them to buy, install, and operate their own hardware. The deal is described as being linked specifically to NVIDIA Blackwell B300 capacity, a reference to NVIDIA’s B300-class data center GPUs used to run AI training and inference workloads.

QumulusAI said the agreement is annually renewable for up to four years. That means the relationship is structured to be extended on a yearly basis, with a ceiling on the maximum contract term of four years as described in the announcement.

The announcement also frames DRW as the counterparty, describing the firm as a diversified global trading company. Beyond identifying DRW and the general nature of the GPU capacity arrangement, the company did not provide additional operational specifics in the reported item, including the size of the capacity, pricing, or the expected workload mix.

While the details are limited in the disclosed report, the structure fits a broader pattern in AI infrastructure spending where data center capacity is increasingly accessed through service contracts. For GPU suppliers and ecosystem partners, these arrangements can help translate hardware demand into longer-duration capacity commitments, at least from the provider’s perspective.

NVIDIA’s role in such deals is typically tied to its data center GPU platform, which is aimed at accelerating AI workloads across training and inference. In this case, the reported agreement is explicitly linked to the Blackwell B300 platform, indicating that QumulusAI is aligning its service offering with NVIDIA’s newer data center generation.

Still, key commercial terms were not disclosed in the reported announcement. The item did not say how many GPUs are included, what performance levels are guaranteed, what portion of the capacity is reserved versus shared, or whether the agreement covers additional services such as networking, software stack management, or support at the application level.

For investors and industry watchers, the next question is whether the agreement results in measurable customer demand for QumulusAI’s managed AI compute services and whether similar capacity contracts are announced as the company scales. The annual renewal feature means subsequent disclosures may also emerge around whether both parties extend the contract after the first year.