$QMLS: a $198M GPU landlord with $246M in announced agreements and a market full of trust issues

I think the market has QMLS wrong! Here me out:

At 8:00 a.m. ET on August 6, \[QMLS traded at $6.04\](https://www.nasdaq.com/market-activity/stocks/qmls). The \[final prospectus\](https://www.sec.gov/Archives/edgar/data/2084026/000143774926023622/quma20260714\_424b4.htm) says 32,867,931 basic shares would be outstanding after registration. That puts the basic equity value near $198.5 million.

Since June 11, the company has announced more than $246 million in signed (signed!), multi-year customer agreements.

My whole thesis sits inside that gap (you can skip to the bottom now if you want). The market sees a tiny former infrastructure operator with a short public history, heavy financing needs, and promises taller than its income statement. but \~I\~ see a GPU landlord that may have found tenants before Wall Street finished reading the lease.

\## The receipts

On June 11, QumulusAI announced \[more than $124.4 million of three-year agreements\](https://www.qumulusai.com/articles/qumulusal-signs-more-than-124-million-in-ai-inference-infrastructure-agreements) with Hyperbolic and another AI inference platform. Those agreements cover 1,280 NVIDIA Blackwell GPUs and include almost $21.9 million in upfront customer commitments.

Then came an \[18 million dollar, two-year take-or-pay agreement\](https://www.qumulusai.com/articles/qumulusai-signs-18-million-two-year-take-or-pay-nvidia-blackwell-b300-agreement-with-marketplace-partner) on July 22, a \[32 million dollar, two-year agreement\](https://www.qumulusai.com/articles/qumulusai-signs-32-million-two-year-nvidia-blackwell-b300-agreement-with-ai-inference-platform-provider) on July 23, and a \[71.9 million dollar, three-year agreement\](https://www.qumulusai.com/articles/qumulusai-signs-719-million-three-year-agreement-with-ai-inference-platform-provider) on July 28.

Divide each announced agreement by its stated term and the simple annualized amount comes to more than $90.4 million. That is rough contract math. GAAP revenue, margin, cash flow, and deployment timing live in separate columns. Put the crayons down for a moment.

At the current basic market cap, the stock trades around 2.2 times that rough annualized contract amount. The company still has to deploy the clusters, bill the customers, collect the cash, fund the hardware, and protect its margins. That execution gap is exactly why the opportunity exists.

\## The hardware is starting to follow the demand

On July 20, QumulusAI announced the \[purchase of 1,632 NVIDIA Blackwell B300 GPUs\](https://www.qumulusai.com/articles/qumulusai-purchases-1632-nvidia-blackwell-b300-gpus-amid-strong-customer-demand) across 204 HGX B300 systems, plus another 192 RTX PRO 6000 Blackwell GPUs. The company said Technology Finance Corporation and USD.ai primarily funded the purchase.

The June agreement block calls for 160 Lenovo and Supermicro bare-metal servers, with Cisco Nexus One networking for the cluster fabric. Those are vendor roles. The customer and channel side includes Hyperbolic, Shadeform, and RunPod.

In May, QumulusAI and Shadeform announced \[two H200 clusters totaling 680 GPUs\](https://www.qumulusai.com/articles/qumulusai-and-shadeform-deploy-two-nvidia-h200-clusters-totaling-680-gpus-for-leading-ai-inference-platforms) for a two-year Kansas City deployment. In July, QumulusAI \[joined the NVIDIA Partner Network as an NVIDIA Cloud Partner\](https://www.qumulusai.com/articles/qumulusai-joins-the-nvidia-partner-network-as-an-nvidia-cloud-partner).

That NVIDIA status confirms participation in the partner ecosystem. NVIDIA gives no guarantee of QumulusAI's economics. The part I like is the sequence: customer commitments, upfront cash, equipment financing, hardware orders, and stated deployment windows.

The racks are servers when occupied. They are very expensive space heaters when empty. QMLS has at least started showing receipts for the tenants.

\## Forward ARR, with adult supervision

Management is guiding to \[$300 million of forward ARR by December 31, backed by 18 MW of capacity\](https://www.qumulusai.com/articles/qumulusai-issues-fiscal-year-2026-guidance-300-million-in-forward-arr-backed-by-18-mw-of-capacity). At today's basic equity value, that is about 0.66 times the target.

The company's definition includes executed contract revenue, expected renewals, deposit-backed reservations, and projected contract signings. Management also warns investors to keep ARR separate from GAAP revenue. Actual Q1 2026 revenue was $3.42 million, gross profit was about $1.28 million, and operating loss was $5.53 million.

This stock looks cheap only if the announced demand turns into activated capacity and collected revenue. The forward number is management's scoreboard. The public filings are the scoreboard that gets audited.

\## Why $6.04 has my attention

The SEC filing says QumulusAI issued about 2.66 million private-placement shares at $10.80 from July 2025 through March 9, 2026. It also says the company began issuing private-placement shares at $23.15 in March.

The current quote sits about 44% below $10.80 and 74% below $23.15...The filing warns that the latest private price may have little relationship to public-market demand. Fair warning - I view this comparison only as evidence that private investors funded the company at much higher prices. It proves nothing about today's fair value.

Public markets are allowed to disagree. They have disagreed with enthusiasm.

\## The bear case has teeth

Q1 revenue was still tiny. The company reported a $49.6 million net loss for the quarter, largely driven by a $54.6 million accounting loss when the fair value of a new convertible note exceeded the cash proceeds. The operating loss was $5.53 million. QumulusAI also reported $102.4 million of total liabilities against $120.1 million of assets, and $16.1 million of cash at March 31. The prospectus identifies material weaknesses in financial controls and says management had to address conditions that initially raised substantial doubt about the company's ability to continue as a going concern.

RunPod produced 71.9% of GPUaaS revenue in Q1 2026, down from 85.7% after the Cloud Minders acquisition in 2025, but still a serious concentration. Several large new customers remain unnamed. The full commercial agreements are not public, so we cannot inspect every termination right, credit condition, or deployment milestone.

The direct listing raised no primary capital. Expansion depends on leases, equipment financing, customer deposits, and convertible notes. The prospectus covers 32.87 million expected basic shares and registers up to 39.47 million shares for resale, including shares tied to financing arrangements. Dilution already lives in the paperwork.

Small caps keep receipts in both directions.

\## My thesis (the good bit)

QMLS looks undervalued because the current quote appears to price in a large execution failure while the visible demand signals have improved: more than $246 million in announced multi-year agreements, $21.9 million of upfront customer commitments in the June block, a take-or-pay deal, a major Blackwell order, expanding marketplace relationships, and NVIDIA Cloud Partner status.

There's no bs squeeze fan fiction here. The bet lives or dies on clusters going live, contracts becoming GAAP revenue, customer concentration falling, and financing staying manageable.

If QumulusAI executes, $6.04 looks like somebody divided by the wrong column. If it misses, this becomes a warehouse full of hot rectangles and a cap table that needs its own weather radar. (Lol)

Heres whqat I want - I want to see the announced summer, fall, and Q3 deployments activated. I want the first public quarterly reports to show the revenue ramp, reasonable gross margin, clean collections, and no ugly financing surprise. If those fail thesis breaks. imo. A working rack beats diamond hands every time.

I am bullish, with a small-cap execution-risk label attached, Rip the math apart.

This is a high-risk stock thesis, ~not~ personalized financial advice. Verify the filings and size your own risk, my regards.