QumulusAI began trading on the Nasdaq Global Market under the ticker QMLS via direct listing, entering public markets as a distributed, inference-first neocloud. The company has disclosed line of sight to 2.5 gigawatts of capacity by 2027 from a projected year-end 2026 base of 18 MW, with $300 million in ARR guidance, showing the hyperspeed potential of a distributed inference model.
What Is Covered in This Article:
- QumulusAI’s Nasdaq direct listing under QMLS and the megawatt-to-gigawatt capacity roadmap it brings to public markets.
- How QumulusAI’s 2.5 GW line-of-sight capacity and $300 million ARR target compare to the gigawatt-scale trajectories and disclosed financials of CoreWeave and Nebius.
- What QumulusAI’s current revenue-to-megawatt ratio implies for its revenue potential as it scales from 18 MW to multi-gigawatt capacity through 2027.
The News: QumulusAI (Nasdaq: QMLS), an Atlanta-based neocloud infrastructure provider, began trading on the Nasdaq Global Market via direct listing on Thursday, July 16. The company was since approved as an NVIDIA Cloud Partner within the NVIDIA Partner Network, and positions itself around a distributed, inference-first deployment model that brings GPU capacity online in months using sub-50 MW colocation sites rather than multi-year mega-campuses.
QumulusAI reported 3,088 deployed NVIDIA GPUs in its investor presentation, up from 560 in June 2025, with a target of roughly 6,900 by year-end 2026. It operates about 7 MW of active HPC power across five colocation sites and roughly 17 MW including legacy bitcoin-mining load, with a projected 18 MW at year-end 2026 and line of sight to 2.5 GW of capacity by 2027 — a 140x expansion from its near-term operational base. The company also discloses 150 MW of nearer-term colocation power and a stated 250 MW potential across 25-to-125 distributed “AI-XP” sites as intermediate milestones. Compute power revenue was $2.1 million in Q1 2026, up from zero a year earlier, and $4.9 million for full-year 2025; total Q1 2026 revenue was $3.4 million, up 83% year over year. The company has announced $124 million-plus in contracted 2026 agreements on three-year, take-or-pay terms, backed by a financing stack that includes a $90 million ATW Partners convertible facility and a $500 million non-binding USD.AI protocol arrangement using tokenized GPU collateral. Management has guided to approximately $300 million in ARR and roughly 30x ARR growth this calendar year.
“We’ve proven QumulusAI can identify capacity, procure and deploy GPUs, stand up operational environments quickly, and convert infrastructure into customer-ready AI compute,” said CEO Mike Maniscalco. “The AI revolution will be built on infrastructure, and we intend to build the layer that helps more organizations break AI’s biggest barriers, accessing the compute they need faster, more transparently, and with greater flexibility.”
QumulusAI Lists on Nasdaq: The Speed Advantage of a Distributed Path to Multi-Gigawatt Scale
Analyst Take: This listing offers public markets a clean way to track the progress of distributed inference. The company, led by CEO Mike Maniscalco, is designing the future of distributed inference through modular sites that align with available power and scale-out fiber. Instead of committing to multi-year mega-campuses, QumulusAI builds around MW-scale footprints and existing colocation, enabling it to bring new sites online in months rather than years. It pairs that operating model with a financing stack built to scale GPU access as sites come online, including a $500 million non-recourse USD.AI facility that uses tokenized GPU collateral and a $90 million convertible note from ATW Partners. That combination is showing up in expanding engagements with leading GPU marketplaces like Runpod, Hyperbolic, and Shadeform, plus inference-focused customers that want reserved compute instances to contain costs and scale open-source inference. Demand is insatiable, and QumulusAI has been meeting it with additional capacity and customer-specific configurations.
QumulusAI enters public markets three orders of magnitude smaller than the neoclouds it invites comparison to, and the investment question is whether its distributed, inference-first model can convert that gap into a durable revenue ramp. The company is not trying to out-build aspiring hyperscalers on capacity over the next year, but is trying to out-cycle them on speed-to-power with a stated line of sight to 2.5 GW by 2027, up from just 18 MW projected at year-end 2026. The revenue math ultimately reduces to two variables: how many megawatts it can energize, and how much revenue each energized megawatt produces.
The Path to Gigawatt Scale with Hyperspeed Deployment
AI-first clouds like CoreWeave and Nebius have made the gigawatt the unit of ambition. CoreWeave crossed 1.0 GW of active power in Q1 2026, holds 3.5 GW contracted, added six data centers to reach 49 active facilities, and has guided to more than 8 GW by 2030 against a revenue backlog near $99.4 billion. Nebius lifted its year-end 2026 contracted-capacity guidance from 3 GW to 4 GW, announced a 1.2 GW owned Pennsylvania site, and raised 2026 capex guidance to $20 billion to $25 billion. QumulusAI’s disclosed roadmap now includes line of sight to 2.5 GW by 2027 — a figure that, if achieved, would place it in the same order of magnitude as CoreWeave’s active fleet. However, the company’s operational base is projected at just 18 MW by year-end 2026, meaning the 2.5 GW figure represents a roughly 140x scale-up within approximately twelve months. The intermediate waypoints — 150 MW of line-of-sight colocation power and 250 MW of AI-XP potential — suggest a phased ramp, but the gap between 18 MW operational and 2.5 GW line-of-sight is the central execution question investors must underwrite.
Distributed Model Doubles Revenue per Megawatt
QumulusAI achieves $16.5 million in revenue per megawatt, a figure that reflects the premium unit economics of its inference-first model and the high utilization rates characteristic of take-or-pay contract structures on a smaller base. By comparison, CoreWeave’s Q1 2026 revenue of $2.08 billion against 1.0 GW of active power implies approximately $8.3 million of annualized revenue per active megawatt — a run-rate figure derived by annualizing the quarter ($2.08 billion × 4 = $8.32 billion) and dividing by the roughly 1,000 MW active fleet at quarter-end. That figure is conservative, as CoreWeave added more than 400 MW during Q1, so average active power during the period was below the exit number, meaning true revenue-per-average-MW was likely higher. Nebius is targeting a similar average in 2026, based on its $7-9 billion exit ARR target with up to 1 GW of active power. Based on this approximate comparison, QumulusAI’s per-megawatt yield exceeds leading AI-first clouds’ fleet-wide average by up to 2x, suggesting that the distributed model either commands higher pricing, runs at tighter utilization, or benefits from a contract mix skewed toward committed capacity rather than spot. As megawatts multiply, maintaining $16.5 million per MW becomes a function of GPU density, contract retention, and the company’s ability to keep fleet utilization near capacity across dozens of distributed sites rather than a handful of hyperscale campuses.
$20B+ Revenue Potential for 2027
At QumulusAI’s demonstrated $16.5 million per megawatt, the revenue envelope scales rapidly with each incremental megawatt energized — but the figure also signals something deeper about the repeatability of its customer relationships and the maturity of its vendor partnerships. Each new site QumulusAI stands up is not a greenfield experiment; it is a replication of a proven deployment playbook built on tight integrations with best-of-breed partners including NVIDIA for GPU compute, VAST Data for high-performance storage, and Cisco for networking fabric. That partner stack gives QumulusAI a reference architecture that can be stamped across dozens of distributed AI-XP sites without re-engineering the environment at each location.
As these partnerships scale, procurement cycles compress, volume pricing improves, and time-to-revenue per site shortens — reinforcing the unit economics that already yield 2x the per-megawatt revenue of larger peers. On the customer side, the 30x ARR growth guidance depends on existing marketplace and enterprise relationships expanding their committed capacity as new sites come online. The three-year, take-or-pay contract structure transforms each customer renewal into a predictable capacity reservation that travels with QumulusAI’s footprint expansion, making revenue growth a function of site replication rather than net-new customer acquisition alone.
Looking further out, the 2.5 GW line of sight transforms the revenue envelope entirely. At QumulusAI’s current $16.5 million per megawatt, 2.5 GW implies roughly $41 billion of annual revenue — a theoretical ceiling. At a blended mid-case of $8 million to $10 million per megawatt — reflecting some normalization toward CoreWeave-like rates as the fleet scales — 2.5 GW implies $20 billion to $25 billion. The path from 18 MW to 2.5 GW requires capital, power contracts, GPU supply, and execution at a pace that no neocloud has yet demonstrated from a comparable starting point. Progress toward the $300 million ARR target this year will calibrate investor confidence in the far larger 2027 ambition.
What to Watch:
- What capital commitments, power purchase agreements, and site activations convert line-of-sight capacity into contracted and then energized megawatts
- How fast the fleet shifts to NVIDIA Blackwell GPUs and sheds legacy mining load
- Whether the distributed model improves the profitability of the company at scale
- How best-of-breed partners across compute, networking, storage, software, and power support gigawatt ambitions
- Whether growth for key GPU marketplace customers drives gigawatt-scale demand
Read the full press release here.
Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.
Other Insights From Futurum:
Apache Spark 4.2: A Leap Forward for AI and Analytics Integration
Knowit’s Leadership Transition Signals Strategic Shift Amid AI Integration
Jacobs Takes a Strategic Step in Germany’s Energy Transition
Author Information
Brendan is Research Director, Semiconductors, Supply Chain, and Emerging Tech. He advises clients on strategic initiatives and leads the Futurum Semiconductors Practice. He is an experienced tech industry analyst who has guided tech leaders in identifying market opportunities spanning edge processors, generative AI applications, and hyperscale data centers.
Before joining Futurum, Brendan consulted with global AI leaders and served as a Senior Analyst in Emerging Technology Research at PitchBook. At PitchBook, he developed market intelligence tools for AI, highlighted by one of the industry’s most comprehensive AI semiconductor market landscapes encompassing both public and private companies. He has advised Fortune 100 tech giants, growth-stage innovators, global investors, and leading market research firms. Before PitchBook, he led research teams in tech investment banking and market research.
Brendan is based in Seattle, Washington. He has a Bachelor of Arts Degree from Amherst College.
