Executives at GPU cloud provider Quake AI (formerly Northern Data) said in an interview with a UK IT outlet, published on October 28, 2026, that their main future customers will be AI agents, not humans. The vision is that AI will secure computing resources autonomously through APIs and use only what it needs. But the largest deal disclosed by parent company RUM Group is a roughly $13.7 billion, six-year contract with an anonymous US cloud provider, and the August earnings materials show no revenue from direct use by AI agents. Simply annualized, that contract comes to about $2.28 billion, or 6.3 times the roughly $360 million obtained by annualizing RUM Group's overall third-quarter revenue outlook. There is a clear gap between the vision of AI agents as the primary customer and the large business-to-business contracts disclosed today.
Quake AI's vision: a future where AI agents buy computing resources
On June 17, 2026, following completion of its acquisition of Germany's Northern Data, Rumble Inc. announced it would reorganize its business into the Rumble video platform and the Quake AI GPU cloud. The next day it renamed the parent company RUM Group, positioning Quake AI, which combines the former Northern Data business with Rumble Cloud, as a new growth pillar. According to the company, Quake AI owns about 22,000 NVIDIA H100/H200 GPUs across nine data centers, and its utilization rate reached about 85% in the second quarter of 2026. It is one of the so-called "neoclouds," providers specializing in GPU computing resources for AI.
RUM Group Chief Financial Officer Michael Masci focused in the interview on how AI agents use services. Humans operate screens and dashboards, while AI agents retrieve information and execute tasks through APIs. What agents need, he said, is not an easy-to-use interface for humans but standardized computing resources with predictable performance and availability, available on demand with pay-as-you-go pricing. Masci pointed out that conveniences built for humans, such as model-serving services, dashboards, and one-click setup, are not necessarily valuable to agents and can even be obstacles.
This vision also appeared in the June reorganization announcement. RUM Group suggested that future business activity may be supported by decentralized AI agents, with transactions conducted on blockchains. It also described its partnership with crypto company Tether as a "trust layer" for ensuring reliability in transactions between AI agents. Whereas in conventional cloud services companies sign contracts to procure computing resources, Quake AI envisions a business model in which AI agents secure them autonomously through APIs.
A $13.7 billion deal with one customer, about 6.3 times RUM Group's annualized revenue outlook
The key to understanding the company's current scale is a large contract signed in August 2026. On August 23, an affiliate of RUM Group entered into a commercial agreement with an unrelated, US-based third-party cloud provider for a data center under development in Maysville, Georgia. The contract covers GPU usage rights and related services, and RUM Group itself also signed a term sheet on warrants with the same customer.
According to the Form 8-K filed with the US Securities and Exchange Commission on August 24, the total order value is about $13.7 billion over six years, with orders split equally across three tranches. At the European Central Bank reference rate of ¥157.25 per dollar on October 30, 2026, the total is equivalent to about ¥2.1543 trillion. However, the obligation for the third order arises only if the customer approves the delivery schedule proposed by the company. The customer will also receive warrants to acquire up to 50,808,408 Class A shares, depending on purchase performance.
The table below converts the main disclosed contracts and the revenue outlook to an annual scale. The annualized figures are simple calculations for comparison and do not represent actual annual revenue.
| Counterparty / comparison | Disclosed amount | Term | Simple annualized amount | Key conditions |
|---|---|---|---|---|
| Anonymous US cloud customer (all 3 tranches) | About $13.7 billion | 6 years | About $2.28 billion | Third tranche conditional on customer approval of delivery schedule |
| Same customer (first and second tranches only) | About $9.13 billion | Assumed 6 years | About $1.52 billion | Excludes the third tranche, which requires delivery schedule approval |
| Tether | Up to $150 million | 2 years after acquisition closes | Up to $75 million | GPUs used for its own AI models and agents |
| Together AI | Undisclosed | Multi-year | Cannot be calculated | Provides NVIDIA HGX B300 computing capacity |
| RUM Group total Q3 revenue outlook | $87 million–$93 million | 1 quarter | $360 million | Four times the $90 million guidance midpoint; consolidated revenue including the video business |
In its second-quarter results on August 10, RUM Group issued formal revenue guidance for the first time, forecasting third-quarter consolidated revenue of $87 million to $93 million. Simply multiplying the $90 million midpoint by four gives $360 million a year. Meanwhile, dividing the $13.7 billion contract evenly over six years gives about $2.28 billion a year, about 6.3 times the annualized consolidated revenue outlook. Using the low and high ends of guidance, the ratio is about 6.1 to 6.6 times, and even the roughly $9.13 billion excluding the third order alone is about 4.1 to 4.4 times as large.
However, this comparison shows the size of the contract and is not a forecast of future annual revenue. Annualizing quarterly revenue does not reflect seasonal variation, and the timing of revenue recognition for the large contract depends on progress in equipment delivery and service provision at the Maysville site, which is still under development. The third order is also conditional. The Tether contract likewise has a cap of $150 million over two years, and its simple annualized amount is only about 20% of RUM Group's consolidated revenue outlook.
The reason neocloud providers value such long-term large contracts is the enormous upfront investment they require. Building out GPUs and data centers takes substantial capital, and long-term purchase agreements are important for projecting how capital spending will be recovered. High customer concentration is also not unusual. For example, major neocloud CoreWeave disclosed in its fiscal 2025 annual report (Form 10-K) that about 67% of its revenue that year came from its largest customer, Microsoft. In 2024, its top two customers accounted for about 77% of revenue.
For Quake AI, securing the funds needed to fulfill the large contract is also a challenge. In its August 8-K, RUM Group said it does not currently have sufficient funds to cover the expenditures associated with the contract. At the same time, the contractual obligations are not conditioned on successful financing. Liquidity at the end of June was $220.5 million. According to a transcript of the earnings call, Masci explained that for contracts like the one with Together AI, the company signs after it has a clear path to securing the necessary equipment and computing capacity.
No agent-driven revenue in the earnings materials
While Quake AI positions AI agents as its main future customers, neither the August 2026 second-quarter earnings release nor the 8-K from the same month shows revenue or usage counts from direct agent use. The earnings release mentions agents only in comments from CEO Chris Pavlovski and in the company description, neither backed by concrete results. The 8-K describing the large contract has no references to agents, APIs, or on-demand use. The earnings call transcript also discusses demand growth from the spread of agentic AI, but does not disclose corresponding revenue or transaction counts.
As of November 1, 2026, Quake AI's official website advertises on-demand use of H100/H200 GPUs and reserved clusters. However, the contact point shown is a sales email address, and no published guidance could be found on a dedicated API for AI agents to purchase computing resources autonomously. The June reorganization announcement said details of products and the development roadmap would be shared over the coming months, but no specific launch timing or revenue targets for agent-oriented services have been given.
That said, the absence of agent revenue from earnings materials does not mean such revenue does not exist. The materials available are mainly those through August 2026, and as of November 1 neither third-quarter results nor the quarterly report (Form 10-Q) had been released. The second-quarter 10-Q was also not part of this review, and the earnings call transcript was prepared by a third party and is not an official record. Still, the three main deals whose amount or counterparty is known are with companies: the anonymous US cloud provider, Tether, and Together AI.
Accordingly, what can be confirmed now is that Quake AI's disclosed contracts center on large enterprise deals, and that the revenue scale of direct agent use has not been disclosed. It is impossible to judge whether the vision of AI agents as the main future customer is already generating meaningful revenue.
Agent demand may be arising through large customers
Large business-to-business contracts and AI agent demand are not necessarily at odds. For example, if the cloud provider that procured GPUs from Quake AI uses those resources to offer inference services for agentic AI, agent demand is indirectly supporting Quake AI's revenue. At the earnings call, CEO Pavlovski also said demand for computing resources continues to expand due to the spread of AI inference and agentic AI, and that there are no signs supply will catch up with demand in the near future.
Technologies for AI agents to secure computing resources directly have already appeared. For example, io.net's "Agent Cloud," according to its product documentation, has a mechanism for agents to secure GPUs via a Model Context Protocol (MCP) server. If credits run short, an agent can also pay with the USDC stablecoin in response to an x402-style payment request and retry the process. However, authentication uses an API key issued by a human, so agents are not acting as fully independent contracting parties. The company has also not disclosed the scale of usage through this mechanism.
On the web, too, automated AI access is increasing. In a tally covering TollBit's partner sites, the frequency of AI bot visits rose from one for every 200 human visits in the first quarter of 2025 to one for every 31 by the end of that year. But what is measured there is mainly website visits by bots used for retrieval-augmented generation (RAG) and similar purposes, not purchases of GPU computing resources. The rise in automated AI access cannot be directly linked to growing agent purchases of cloud services.
These examples show the technical possibility of AI agents automating the use of and payment for computing resources. Meanwhile, it has not been disclosed what purposes the anonymous cloud provider that is the counterparty to Quake AI's large contract will actually use the GPUs for. So while agent demand may indirectly be supporting Quake AI's revenue, its scale cannot be confirmed. What matters is not whether demand from AI agents exists, but how much of Quake AI's revenue comes from direct use by agents.
Agent-facing APIs and revenue disclosure will be the test
To evaluate Quake AI's business vision, it will be important to see how far future earnings reveal actual agent usage. The first thing to watch is whether RUM Group discloses Quake AI's API-based revenue and distinguishes use by AI agents from ordinary use by humans and companies. If agent use accounts for 10% or more of Quake AI's revenue, that would be a basis for judging that the vision has reached a certain business scale. However, the 10% level is not a target set by RUM Group but an analytical benchmark for judging contribution to the business.
Information that API revenue has grown does not by itself confirm growth in agent demand. Because Quake AI also accepts on-demand use by humans and companies, data that separates the two is needed. Whether the company formally releases an agent-oriented API with clearly stated pricing will also be a gauge of how concrete the vision has become.
In the existing business, the focus is whether the $13.7 billion contract is fulfilled as planned. If delivery of the first tranche is delayed, or the customer approval of the delivery schedule needed for the third order is not obtained, the revenue outlook for the large contract supporting the current business could change. For neoclouds that require huge capital investment, what matters is not only winning contracts but also building out GPUs and data centers and delivering services on schedule.
Similar businesses are spreading in Japan as well. On May 25, 2026, SoftBank announced that it would begin offering, from October, a GPU cloud service as a neocloud business, including "Inference as a Service," which provides large language model (LLM) inference via API. However, its service does not limit users to AI agents, and providing computing resources through APIs must be distinguished from a business model with AI agents as the main customer.
Mechanisms providing a technical basis for the future Quake AI envisions, in which AI agents autonomously purchase computing resources, are beginning to appear. But the revenue base that can be confirmed from the company's current disclosures remains centered on large enterprise contracts. Whether agent-facing APIs become concrete, and whether actual usage and revenue show up in earnings materials, will be the key to judging the viability of this vision.