Rumble eyes $3B AI compute run rate from 250MW power pipeline

Rumble's Quake AI unit is targeting more than $3 billion in annual revenue from 250 megawatts of grid-connected power by 2027, as the video platform operator pivots into AI compute infrastructure.

"The company's primary financial opportunity is expected to come from Quake AI," Mike Massey, group CFO at Rumble, said at a Canaccord Genuity event with senior analyst Kingsley Crane.

Quake AI operates 22,000 Hopper-generation GPUs across Europe serving more than 50 customers across inference, training, pre-training and QLoRA workloads. GPU utilization improved from below 20 percent in mid-2025 to more than 83 percent in the first half of this year. The unit's largest contract is a $270 million multiyear deal with Together AI for NVIDIA B300 GPUs.

The 250MW pipeline includes 180MW in Atlanta, a Pittsburgh site for AI-native customers, 50MW in Sweden and 20MW in Norway. Data center capital expenditures typically range from $8 million to $12 million per megawatt, and the Atlanta project alone could require 40,000 to 50,000 NVIDIA Vera Rubin GPUs.

Quake AI's GPU Fleet and Utilization Gains

Massey attributed the utilization improvement to new management and a renewed focus on execution and customer credibility following the Northern Data acquisition. He said demand continues to exceed supply across AI compute markets, and pricing for Hopper capacity has strengthened.

Quake's current Hopper GPU estate generates roughly $6 million to $7 million of annual revenue per megawatt, Massey said. Latest-generation Blackwell systems have been priced at more than $11 million per megawatt annually, while future Vera Rubin deployments are expected to command a premium.

Massey argued that older GPU generations could remain useful longer than some investors expect, as customers use different hardware generations for different tasks such as retraining and recursive inference.

Power Pipeline and Capital Strategy

The Atlanta location has use permits, a Georgia Power CES agreement, an installed substation and transformers already in place, Massey said. The company is engaging with multiple hyperscalers as it seeks a partner for the site.

Massey said management is less concerned about chip supply than about construction-related constraints, such as skilled labor, steel, uninterruptible power systems and chillers. The Atlanta market offers a strong labor pool, and several long-lead items including transformers, a substation and generators are already available.

The company expects that a majority of construction capital could be financed when supported by long-term, take-or-pay customer agreements. Tether, which owns nearly 50 percent of the combined company, provides a tailwind in discussions with partners because of its capital base and international reach, Massey said.

Quake AI intends to remain focused on infrastructure rather than expanding into software or platform-as-a-service offerings, Massey said. The company wants to avoid competing with customers that provide higher-level AI services.

Rumble shares trade on NASDAQ under the ticker RUM. Massey said Quake AI's financial results are expected to "far outstrip" those of the video platform over time, and that the broader market has not yet fully recognized RUM Group's role in AI compute as a service. The company is also exploring potential AI-training applications for its video data after receiving outreach from robotics companies interested in spatial and temporal video datasets.

This article is for informational purposes only and does not constitute investment advice.