QumulusAI, Inc. Common Stock Q3 FY2026 earnings call

• Strategic Pivot to Hyper-Speed: CEO Mike Maniscalco emphasized a deliberate shift away from traditional four-year data center builds toward 'hyper-speed' deployment into existing pockets of power (1-50 MW). This strategy allows for capacity activation in months rather than years, addressing immediate market demand.
• GPU Fleet Expansion: The GPU fleet grew by 224% from Q1, increasing from 952 units to 3,088 units. Most of this growth came from compute assets that were already sold but are now coming online.
• Commercial Momentum: The company signed 21 new direct customer contracts totaling $169.7 million in the quarter. Total signed contract value is now $282.5 million across 40 contracts with a weighted average term of 2.2 years.
• Customer Base Transition: Direct customer relationships now account for over 96% of recurring revenue, a significant shift from less than 10% a year ago. The company has moved away from dependence on single marketplaces toward direct multi-year agreements.
• FACTS Framework: Management highlighted five key value drivers: Flexibility, Access, Trust, Speed, and Security. They noted that 'Speed' is currently the primary decision factor for customers who need compute immediately.
• NVIDIA Partnership: Two weeks after its direct listing, QumulusAI was approved as an NVIDIA Cloud Partner, securing access to the supply chain necessary to source GPUs rapidly.

Total revenue for the second quarter of 2026 was $6.7 million, representing a 118% year-over-year increase from $3.1 million in the prior year period. The primary growth driver was compute power revenue, which surged 328% to $5.6 million, accounting for 84% of total revenue (up from 43% in the same period last year). Legacy lines, including mining hosting and cryptocurrency mining, contributed approximately 16% of revenue, down significantly from 39% in Q1 and 57% a year ago. Gross profit reached $4.5 million with an expanded gross margin of 66.6%, up from 55.1% previously.

• Capacity Targets: Management reaffirmed the guidance issued on July 14, 2026, targeting 18 MW of HPC capacity by year-end. This includes 8 MW already leased/sold and 10 MW to be developed. Recent announcements include a 3.75 MW colocation agreement in Metro Atlanta, part of that 10 MW development plan.
• Revenue per Megawatt: New Blackwell-based contracts are generating between $18 million and $20 million of annualized revenue per megawatt. The blended figure across the installed base is approximately $16 million per megawatt.
• ARR Guidance Status: While a $300 million ARR target was mentioned in previous materials, management did not explicitly reaffirm or revise this specific ARR figure during this call, focusing instead on capacity constraints and operational execution.
• Full Year Revenue Expectation: All 8 MW of currently leased capacity is expected to begin generating revenue within the current fiscal year.

• Supply Chain Constraints: The primary constraint on growth is not demand, but the availability of land, power, and shell infrastructure. Securing these physical assets remains the bottleneck for scaling operations.
• Capital Market Dependencies: The business model relies heavily on accessing capital markets for both equity and debt financing to procure GPUs and fund deployments. Changes in the cost of capital or availability of funding could impact growth speed.
• Operational Execution Risk: Closing the gap between signed demand ($282.5 million) and recognized revenue requires rapid deployment. Any delays in getting GPUs online due to technical or logistical issues could impact financial performance.
• Contract Renewal Risk: Unlike competitors locking in long-term rates, QumulusAI’s shorter contract terms mean exposure to fluctuating GPU prices and market conditions upon renewal, though management views this as a benefit in a rising price environment.

Q: Michael Donovan asked about the economic structure of the AI-XP initiative, specifically CapEx contributions and revenue sharing with site partners, and the timeline for energizing the initial sites.

A: Mike Maniscalco clarified that the AI-XP sites are distinct from their high-growth third-party colocation strategy. In the AI-XP model, partners provide the land, power, and shell, while QumulusAI handles rack installation and GPU deployment once connectivity is ready. The first site in Wichita is currently being focused on, with additional sites further out in the pipeline.

Q: Bill Papanastasiou inquired about customer profiles, the GPU financing environment post-IPO, and whether the $300 million ARR guidance remained valid given demand trends.

A: Maniscalco stated that near-term growth is driven by 'AI natives' (VC-backed inference companies) alongside enterprise verticals. He noted that financing options have increased since becoming public, allowing better alignment of capital to assets. Regarding guidance, he declined to reaffirm the $300 million ARR figure but emphasized that demand is not the constraint; rather, the lack of land and power is the primary limiter to growth.

Q: Allen Klee sought details on the competitive landscape against neoclouds pursuing gigawatt-scale campuses and asked about GPU financing amortization periods.

A: Maniscalco explained that QumulusAI differentiates itself by targeting sub-50 MW pockets of power for rapid deployment, whereas competitors focus on long-term, large-scale projects. He cited their NVIDIA partnership and OEM relationships as key advantages. Scott Krosnowski added that most GPU financing arrangements are structured over three years, aligning with contract terms, and they aim to lower the cost of capital as they scale.