Corvex, Inc. (MOVE) | The Buildout — AI Infrastructure
The Verdict
Corvex runs a neo-cloud: a GPU-accelerated AI cloud platform that sells dedicated compute capacity for AI training and inference. It rents space inside third-party enterprise data centers, buys NVIDIA GPUs, and combines accelerators with networking, power, cooling and systems software to deliver bare-metal or managed Kubernetes clusters to AI model labs, hyperscalers, government-backed AI initiatives and enterprises. Management describes a three-layer strategy: the AI factory compute layer that generates revenue today, a Token Factory inference layer now in closed alpha, and a confidential-computing software layer branded Assured AI that is still being designed. The company was formerly Movano, a consumer-healthcare wearables issuer, and became an AI-infrastructure business through a merger in March 2026; the legacy healthcare assets were disposed of on June 30, 2026.
| Market Cap | — |
| Revenue (TTM) | $4M |
| Revenue Growth | +780.0% |
| EBITDA Margin (TTM) | -525.0% |
| Net Cash | $7M |
| Earnings Beats | 3 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- Definitive agreements are executed for both expansion sites — a doubling of the Mid-Atlantic site and a new Midwest site — with roughly 3,000 additional latest-generation GPUs planned.
- Deferred revenue grew to $4.4 million at March 31, 2026 from $12 thousand at December 31, 2025, reflecting contracted AI compute capacity not yet recognized.
- The ~$33 million PIPE closed September 2, 2026, taking pro forma cash from ~$22 million to ~$55 million, with Goldman Sachs, Morgan Stanley and Oppenheimer as placement agents.
- The legacy connected devices segment, which absorbed $3,210 thousand of operating loss in Q1 2026 on $35 thousand of revenue, was disposed of on June 30, 2026.
- Leadership was rebuilt around capital formation and site development: a new CFO effective June 29, 2026, two new board members, and a data-center sourcing and development lead hired from Cushman & Wakefield.
What We’re Watching
- Both expansion sites are targeted to be 'ready for service' in Q4 2026, with revenue from them anticipated starting Q1 2027 — the first clean test of whether the build converts to the top line.
- The build is conditional on capital availability, customer commitments, equipment delivery, utility energization and local permitting, and no named power or utility counterparty appears in the quoted material.
- Customer concentration: one unnamed customer was 52% of Q1 2026 revenue. Whether that relationship diversifies in the next 10-Q is the key concentration question.
- Supplier concentration is disclosed inconsistently — the 10-K says three suppliers were 63%, 23% and 9% of FY2025 purchases, while the Q1 2026 10-Q says no supplier exceeded 10%. The filings do not reconcile the two.
The thesis is advancing on commitment and financing but remains unproven on execution. Since its debut quarter, Corvex has signed definitive agreements for a five-fold capacity increase, closed a ~$33 million PIPE, disposed of the legacy healthcare segment, converted preferred stock, and reported ~$22 million of contracted annualized recurring revenue on live compute. What has not happened is the part that matters: the new capacity is not yet in service, no multi-year take-or-pay offtake contract has been signed, and no financial guidance has ever been issued. The open question is whether Q4 2026 site readiness and the Q1 2027 revenue start arrive on schedule, and whether the new megawatts get contracted at the pricing the plan assumes.
Earnings
Corvex reported $3.8 million of revenue in the quarter ended June 30, 2026, up from $0.5 million in the March quarter, which included the AI platform for only 12 days. Reported gross margin was −26.1%, and EBITDA was a loss of $11.7 million. The company's standout operating metric was contracted annualized recurring revenue on live compute of approximately $22 million as of August 14, 2026, with $22 million of cash reported at June 30, 2026.
| Metric | Q2 FY2026 | Q1 FY2026 | Q2 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $4M | $0M | $0M | +3700.0% |
| Gross margin | -26.1% | -161.6% | -251.5% | +22540bps |
| EBITDA | −$12M | −$4M | −$3M | +265.6% |
| EPS | $-5.12 | $-0.03 | $-0.04 | +12364.0% |
| Contracted ARR on live compute | $22M | n/a | n/a | — |
We believe AI is driving a once-in-a-generation transformation in global computing infrastructure.— Jay Crystal, Chief Executive Officer, 2026-05-22
Management tone: The only earnings call in the source set is the Q1 2026 debut call on May 22, 2026; the June quarter arrived as a press release with no transcript. On that call, management framed the market bullishly, describing AI as a 'once-in-a-generation transformation,' while leaning on conditional language for the second and third layers — 'upcoming,' 'being designed,' 'intended.' It also volunteered that the company is 'early in our development as a public company.' Management took no analyst questions, citing the limited operating period in reported results, and the source contains no earlier call, so tone cannot be compared with a prior quarter.
Management Guidance
The company has issued no formal financial guidance — no revenue, margin, EBITDA or capex outlook — on the Q1 2026 call or in the releases since. Its forward numbers are operating targets presented under forward-looking-statement caveats: capacity of roughly 1.5 MW rising to roughly 8 MW by year-end 2026, about 3,000 additional GPUs, both expansion sites targeted ready for service in Q4 2026, and revenue from those expansions anticipated starting Q1 2027. The 12.5 MW right-of-first-refusal block is anticipated ready for service in Q3 2027. Management states these are subject to capital availability, customer commitments, equipment delivery, utility energization and local permitting, among other conditions.
Trajectory
Reported revenue moved from $0.5 million in the March 2026 quarter — which included the AI platform for only 12 days — to $3.8 million in the June quarter. The EBITDA loss widened in dollars, from $4.5 million to $11.7 million, as data-center costs scaled ahead of contracted revenue, while margin percentages improved sharply as the mix shifted away from legacy healthcare. Deferred revenue reached $4.4 million at March 31, 2026 from $12 thousand at December 31, 2025, reflecting contracted AI compute capacity not yet recognized. The company dates revenue from its new capacity to the first quarter of 2027.
The Model
The model projects FY+1 revenue of $14.8 million with EBITDA of −$37 million (−251.4%), and FY+2 revenue of $73.5 million with EBITDA of −$19 million (−25.3%). The near term is anchored on capacity already contracted: the Mid-Atlantic and Midwest sites are targeted ready for service in Q4 2026 with revenue anticipated from Q1 2027. FY+2 depends on that capacity being energized, utilized and priced near plan, and on the Token Factory and software layers beginning to contribute. Dispersion across the model's five runs is 11% on FY+1 revenue and 72% on FY+2 revenue.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $0M | $15M | $74M |
| YoY Growth | — | +3600.0% | +396.6% |
| EBITDA | −$15M | −$37M | −$19M |
| EBITDA Margin | -3825.0% | -251.4% | -25.3% |
Projections are the median of 5 independent model runs.
The company has issued no formal financial guidance — no revenue, margin, EBITDA or capex outlook — on the Q1 2026 call or in the releases since. Its forward numbers are operating targets presented under forward-looking-statement caveats: capacity of roughly 1.5 MW rising to roughly 8 MW by year-end 2026, about 3,000 additional GPUs, both expansion sites targeted ready for service in Q4 2026, and revenue from those expansions anticipated starting Q1 2027. The 12.5 MW right-of-first-refusal block is anticipated ready for service in Q3 2027. Management states these are subject to capital availability, customer commitments, equipment delivery, utility energization and local permitting, among other conditions.
What Could Go Right — and Wrong
- A signed multi-year take-or-pay GPU-as-a-service contract — the structure management says it is targeting — would convert the ~$22 million contracted ARR into a disclosed revenue base.
- Both expansion sites ready for service in Q4 2026 and revenue starting Q1 2027 would validate the brownfield-with-live-power thesis and the dated milestone.
- Customer diversification — a second or third disclosed AI customer, or the 52% relationship falling below half of revenue — would reduce the concentration exposure.
- The Token Factory exits closed alpha and reaches general availability with named users, beginning to support the recurring, asset-light revenue mix management describes.
- A named confidential-computing or Assured AI licensing customer, particularly a federal or regulated one, would validate the software layer and its funnel into AI factory deployments.
- Energization or permitting delays push the Q1 2027 revenue start out; because revenue is recognized straight-line over committed terms, a late start moves the revenue quarter by quarter.
- The new capacity energizes but goes uncontracted — eight megawatts of idle compute is a cost, not a revenue stream.
- Loss of, or pricing pressure on, the 52% customer, with no named contract disclosed to cushion it.
- A further financing that is materially dilutive, or a shortfall, since both the build and the ROFR block are conditioned on capital availability and pro forma cash of ~$55 million is small against a multi-megawatt plan.
- Goodwill impairment: $519.3 million of goodwill on a $604 million asset base, under a preliminary allocation, on an acquiree determined to be a variable interest entity that could not self-finance.
Looking Ahead
Over the next twelve months the story turns on two dated events: both expansion sites targeted ready for service in the fourth quarter of 2026, and revenue from them anticipated starting in the first quarter of 2027. In parallel, the company must file a PIPE resale registration statement by October 2, 2026, finalize the merger purchase-price allocation by about March 19, 2027, and move the Token Factory out of closed alpha. Whether any multi-year take-or-pay offtake is signed has no stated date and carries an explicit no-assurance caveat.
- By Oct 2, 2026PIPE registration filing — Resale registration statement for the ~$33M PIPE must be filed by Oct 2, 2026.
- Q4 2026Expansion sites ready — Both Mid-Atlantic and Midwest sites targeted 'ready for service.'
- Year-end 20268 MW capacity target — Critical IT capacity targeted to rise from ~1.5 MW to ~8 MW.
- Q1 2027Expansion revenue start — Revenue from the new capacity anticipated to begin; the first clean test.
- By ~Mar 19, 2027Purchase-price allocation — Merger purchase-price allocation to be finalized within the 12-month window.
- Q3 202712.5 MW ROFR readiness — Right-of-first-refusal capacity anticipated ready for service.
Financials
Annual Summary
| Metric | FY2024 | FY2025 | TTM | YoY |
|---|---|---|---|---|
| Revenue | $1M | $0M | $4M | -63.6% |
| Gross Margin | -681.1% | -1002.1% | -46.6% | 32,097bps |
| EBITDA | −$24M | −$15M | −$23M | +36.5% |
| EBITDA Margin | -2190.9% | -3825.0% | -525.0% | 163,409bps |
| Net Income | −$24M | −$18M | −$28M | +22.8% |
| Free Cash Flow | −$23M | −$11M | −$20M | — |
| Net Cash | — | — | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)-46.6%
- EBITDA Margin (TTM)-525.0%
- Net Margin (TTM)-629.5%
- ROIC-3.8%
- SBC / Revenue279.5%
The Company
Corvex operates a neo-cloud: a GPU-accelerated AI cloud platform that sells dedicated compute capacity for AI model training and inference. It combines high-performance AI accelerators, networking, power, cooling and systems software, and sells to AI model labs, hyperscalers, government-backed AI initiatives and enterprises. Customers buy compute and storage capacity and platform services as a stand-ready obligation, and revenue is recognized over time on a straight-line basis across the committed contract term. Management describes a three-layer strategy: the AI factory compute layer that generates revenue today, a Token Factory inference layer in closed alpha, and a confidential-computing software layer, branded Assured AI, that is designed to license onto third-party infrastructure and act as a lead source for future AI factory deployments.
Corvex operates its platform in a third-party Tier III data center in Delaware, its only named operating compute facility at the 10-K date, with a principal office at 3401 North Fairfax Drive in Arlington, Virginia. It does not build its own data centers: the expansion plan is brownfield, placing clusters inside existing enterprise data centers with live utility power, which the company says can shorten deployment timelines and reduce development risk relative to greenfield construction. Clusters are designed around published vendor reference architectures and offered as bare metal or with a managed Kubernetes orchestration layer. The company buys NVIDIA GPUs — the B200 is named, and the product text also references the H200 — and states that it has no long-term supply contracts guaranteeing capacity.
Business Segments
Competitive Landscape
The 10-K names the hyperscalers — Amazon Web Services, Google Cloud Platform and Microsoft Azure — as representing 'the most significant competitive threat,' and describes neoclouds and specialized providers such as CoreWeave and Nebius as focusing on 'GPU-optimized infrastructure with flexible provisioning and competitive pricing.' Management says the strategy is differentiated by investing in 'higher value layers of the AI stack rather than participating solely as a provider of commodity compute infrastructure,' with confidential computing aimed at workloads where security rather than price is the buying criterion. The supply-chain intelligence scan adds a blunt external read: MOVE is not yet visible in the supply chain, and no named competitor treats it as a peer.
- Amazon Web ServicesNamed in the 10-K within the traditional hyperscale cloud platforms described as 'the most significant competitive threat.'
- Google Cloud PlatformNamed in the 10-K within the traditional hyperscale cloud platforms described as 'the most significant competitive threat.'
- Microsoft AzureNamed in the 10-K within the traditional hyperscale cloud platforms described as 'the most significant competitive threat.'
- Named as a neocloud focused on 'GPU-optimized infrastructure with flexible provisioning and competitive pricing.'
- Named as a neocloud focused on 'GPU-optimized infrastructure with flexible provisioning and competitive pricing.'
Supply Chain
Corvex sits downstream of GPU and data-center equipment suppliers and upstream of the AI labs and enterprises that rent its compute. It buys NVIDIA GPUs and rents space inside third-party data centers; no neighbor in the source set names the company.