Rumble and Together AI sign multi-year NVIDIA Blackwell compute deal

Rumble (NASDAQ: RUM) and Together AI have signed a multi-year agreement under which Together AI will purchase dedicated GPU cloud capacity from Rumble, powered by liquid-cooled NVIDIA HGX B300 systems. The deal is positioned by the companies as a commitment to building AI infrastructure outside the traditional hyperscaler ecosystem, giving AI-native developers more choice in where they run compute-intensive workloads.

The agreement covers large-scale inference, fine-tuning and training workloads, though neither party disclosed contract value, committed spend, or the precise number of GPUs involved. Rumble noted it has received multiple non-dilutive GPU financing offers from unaffiliated third parties and said the deal provides "long-duration revenue visibility" as it scales its cloud infrastructure.

The deal

Chris Pavlovski, Chairman and Chief Executive of Rumble, described the agreement as a validation of the company's strategy to build what he termed "sovereign, high-performance AI compute as a Service outside the hyperscaler stack." Together AI's founder and CEO Vipul Ved Prakash said access to reliable Blackwell-class capacity is critical for customers training and deploying demanding AI models, and that the partnership expands his company's global GPU footprint.

Together AI serves AI-native builders — startups and enterprises building directly on foundation models — and has positioned itself as an open-platform alternative to hyperscaler-hosted model APIs. The company offers access to a wide range of open-source models and has attracted notable usage among developers seeking fine-tuning and custom-inference capabilities.

Market context

The GPU compute supply crunch at the frontier — particularly for NVIDIA's Blackwell generation — has pushed AI companies to diversify their infrastructure providers beyond the three major hyperscalers. Independent GPU cloud providers, including CoreWeave, Lambda Labs and Crusoe Energy, have grown rapidly on the back of this demand, and Rumble's move into AI compute represents a significant strategic pivot for a company that began as a video-streaming platform.

NVIDIA's HGX B300 systems, based on the Blackwell architecture, represent the current performance ceiling for large-scale AI training and inference. Liquid-cooled deployments are increasingly the norm at this tier: rack-level thermal density has outpaced what air-cooled facilities can handle cost-effectively, and hyperscalers and specialist providers alike are retrofitting or building new facilities with direct liquid cooling as standard.

The independent GPU cloud sector remains under scrutiny on questions of reliability, security certification and long-term financial stability — enterprise and AI-native buyers routinely request SOC 2 Type II evidence and redundancy guarantees before committing to non-hyperscaler providers. Rumble has not disclosed what compliance certifications its cloud infrastructure currently holds, nor the locations of its data centres, which may be a factor for customers with data-residency requirements.

Competitive positioning

For Together AI, diversifying Blackwell capacity across independent providers reduces concentration risk at a time when demand continues to outstrip supply. For Rumble, the deal signals an attempt to establish credibility in the enterprise AI infrastructure market — a segment where brand recognition, financial strength and operational track record weigh heavily in procurement decisions.

The press release references a potential Northern Data business combination in its forward-looking risk disclosures, suggesting a broader consolidation play may be under way, though no deal terms were disclosed. Investors will look for further named enterprise customers and published infrastructure specifications as the next evidence of commercial traction.