Axe Compute Inc. (AGPU) | The Buildout — AI Infrastructure
The Verdict
Axe Compute is a neocloud: it buys NVIDIA GPUs, places them in data centers, owns them, and rents the resulting dedicated clusters to enterprises that need AI training and inference capacity. Two lines carry the business — Access, an immediate-inventory service where customers can start in as little as 24 to 48 hours, and Build, where Axe co-engineers, deploys, owns, and operates large-scale clusters on three- to five-year terms. Management calls it "a virtual data center owner, that doesn't own the buildings, but we own the equipment globally." It makes no chips, models, or software.
| Market Cap | — |
| Revenue (TTM) | $3M |
| Revenue Growth | +220.0% |
| EBITDA Margin (TTM) | -950.0% |
| Net Cash | $20M |
| Earnings Beats | 1 of 6 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- The 2026 signed contracted book reached more than $3 billion, after more than $2.8 billion of new contracts were signed in Q2 FY2026 alone.
- Customers have paid up front: $43 million received in May 2026, more than $317 million by August, and more than $184 million reported again in September.
- The $260 million landmark deal is a 36-month contract for 2,304 NVIDIA B300 GPUs and 4.8 MW of dedicated N+1 power, expected to pay about $21 million per quarter once live.
- Q2 FY2026 revenue grew more than 90x sequentially, the first full quarter of compute operations.
- The Helomics sale closed 2026-09-11, completing the transition to a pure-play neocloud and removing the declining legacy drug-discovery segment.
What We’re Watching
- The $260 million cluster is targeted to deploy in Q3 2026 and had not gone live in the source material; a slip defers the ~$21 million per quarter.
- The $4.3 billion pipeline is unsigned — 45 qualified prospects — and management says "not all of them will close."
- One customer was about 79% of accounts receivable at 3/31/26, and counterparties are unnamed.
- No financing plan is disclosed for the owned-hardware builds, and the 49% Duos equity stakes sit on nonbinding term sheets.
The evidence shows a company whose forward indicators have scaled far ahead of its reported financials. Signings, prepayments, and one large committed flagship deal are dated and real; the reported P&L is still a lagging artifact of a business that barely existed a year earlier. Near-term promises — the May installment, the CFO transition, the drug-discovery exit — were kept, which supports the case that management does what it says. The thesis stays unproven until the flagship cluster is live, recognized revenue follows, and a financing structure is disclosed. The open question: can the balance sheet fund the owned-hardware build fast enough to convert the signed book before concentration or capital becomes binding?
Earnings
In Q2 FY2026 (quarter ended 2026-06-30), the first full quarter of compute operations, revenue was $3.2 million and gross margin was 6.3%. The company said revenue grew more than 90x sequentially from Q1 FY2026's $35,311. The quarter carried a net loss of $17.2 million, and the company reported signing more than $2.8 billion in new contracts during the period.
| Metric | Q2 FY2026 | Q1 FY2026 | Q2 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $3M | $0M | $0M | — |
| Gross margin | 6.3% | -427.6% | -579.4% | +58570bps |
| EBITDA | −$4M | −$3M | −$2M | +60.0% |
| EPS | $-0.87 | $-0.36 | $-0.23 | +278.3% |
| 2026 signed contracted value | >$3.0B | n/a | n/a | — |
| Expected annualized run rate (full deployment) | >$696M | n/a | n/a | — |
In less than eight months, we have gone from our first revenue-generating compute contract to more than $3 billion in signed contract value.— Christopher Miglino, CEO, 2026-08-14
Management tone: Only one call transcript exists in the source material — the Q1 FY2026 call (2026-05-18) — and it is a scripted presentation with no analyst Q&A, so a quarter-over-quarter tone comparison is not possible. There, management was unusually declarative: the CEO said the first deal "opened a floodgate of opportunities for us." The President attached an explicit caveat that the pipeline is "not signed contracts." The CFO was direct that the quarter was not representative and that the $4.3 million loss was noncash. No Q2 FY2026 call appears in the record.
Management Guidance
The company does not publish formal financial guidance and reports no non-GAAP adjusted figures. On its Q1 FY2026 call, management said the $260 million flagship deal should generate about $21 million per quarter once live, with deployment targeted for Q3 2026 and revenue recognized ratably over the 36-month term. Management reaffirmed a goal of closing around $1 billion in 2026 transactions and said it expected the model to begin showing in the financial statements in Q2 and Q3 2026. It also said that if 3 or 4 of the pipeline deals closed in 2026, that would add "hundreds of millions of dollars" in incremental contracted revenue.
Trajectory
Reported revenue had stayed under $1 million a quarter for years, drifting down with the legacy drug-discovery business. That changed with the compute pivot: Q1 FY2026 revenue was $35,311, and Q2 FY2026 revenue was $3.2 million, which the company described as more than 90x sequential growth in the first full compute quarter. Q2 gross margin was 6.3%, and hardware depreciation has barely begun — compute-segment depreciation and amortization was $0 in Q1, when total operating costs were $3.5 million. The forward numbers are gated on deployment rather than demand: management points to more than $3 billion of signed 2026 contracted value and an expected annualized run rate of more than $696 million upon full deployment.
The Model
The model projects FY+1 revenue of $40.15 million with an EBITDA loss of $1 million (-2.3%), and FY+2 revenue of $370.0 million with EBITDA of $109 million (29.5%). The near term is anchored by deployment: the flagship cluster's roughly $21 million per quarter once live, plus the first recognition from the signed 2026 book. FY+2's step-up to $370.0 million implies the wider signed-contract book deploying at scale, carrying the model to positive EBITDA. The projections carry wide dispersion — the FY+1 revenue spread across the model's runs is 19%, and the FY+2 spread is 65%.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $0M | $40M | $370M |
| YoY Growth | — | +40050.0% | +821.5% |
| EBITDA | −$28M | −$1M | $109M |
| EBITDA Margin | -27700.0% | -2.3% | 29.5% |
Projections are the median of 4 independent model runs.
The company does not publish formal financial guidance and reports no non-GAAP adjusted figures. On its Q1 FY2026 call, management said the $260 million flagship deal should generate about $21 million per quarter once live, with deployment targeted for Q3 2026 and revenue recognized ratably over the 36-month term. Management reaffirmed a goal of closing around $1 billion in 2026 transactions and said it expected the model to begin showing in the financial statements in Q2 and Q3 2026. It also said that if 3 or 4 of the pipeline deals closed in 2026, that would add "hundreds of millions of dollars" in incremental contracted revenue.
What Could Go Right — and Wrong
- The $260 million cluster goes live in Q3 2026 and roughly $21 million per quarter of revenue recognition begins on schedule.
- The more than $3 billion signed book and the $4.3 billion pipeline convert into deployed clusters with named customers.
- A disclosed financing structure funds the owned-hardware build without straining the balance sheet.
- The compute segment shows the model's true economics — revenue, hardware depreciation, and margin at scale — and deals prove cash-flow positive as management claims.
- Customer concentration narrows as more of the book converts and counterparties are disclosed.
- The Q3 2026 flagship deployment slips, deferring the roughly $21 million per quarter of recognition.
- The pipeline fails to convert, and signed value does not become recognized revenue on the implied timeline.
- The build requires capital the company must raise or borrow at undisclosed terms, pressuring the balance sheet.
- A large customer is delayed or lost, given one customer was about 79% of accounts receivable.
- The Aethir token reverses again, producing further noncash losses and testing the "additional cash" framing.
Looking Ahead
The next twelve months turn on execution. The $260 million flagship cluster is targeted to go live in Q3 2026, and the Duos 55 MW program's initial readiness is targeted for late 2026 into early 2027. The 49% Duos equity stakes and the financing of builds remain on nonbinding term sheets or undisclosed. Management expects the financial statements to begin reflecting the model in Q2 and Q3 2026. What would settle the case is recognized revenue, hardware depreciation, and margin at scale from a live cluster — none of which is in the source material so far.
- Q3 2026Flagship cluster go-live — Tests whether ~$21M/quarter recognition begins on schedule.
- Q3 2026Compute model in financials — First read on revenue, hardware depreciation, and margin at scale.
- Late 2026–early 2027Duos 55 MW readiness — Subject to construction, commissioning, and performance testing.
- OngoingPipeline conversion — Whether the unsigned $4.3B pipeline becomes signed contracts.
- Through December 2028Aethir locked-token vesting — $9.4M current and $5.9M noncurrent receivable at 3/31/26.
- ~September 2027Helomics lock-up expiry — 12-month lock-up on 636,328 DataMeds shares and the note.
Financials
Annual Summary
| Metric | FY2024 | FY2025 | TTM | YoY |
|---|---|---|---|---|
| Revenue | $1M | $0M | $3M | -90.0% |
| Gross Margin | -46.7% | -140.5% | 6.3% | 9,378bps |
| EBITDA | −$10M | −$28M | −$30M | -174.3% |
| EBITDA Margin | -1010.0% | -27700.0% | -950.0% | 2,669,000bps |
| Net Income | −$13M | −$233M | −$254M | -1735.4% |
| Free Cash Flow | −$13M | −$10M | −$5M | — |
| Net Cash | — | — | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)6.3%
- EBITDA Margin (TTM)-950.0%
- Net Margin (TTM)-7921.9%
- ROIC-178.0%
- SBC / Revenue21.9%
The Company
Axe Compute provides dedicated enterprise GPU compute — the physical AI factory that enterprises rent when they cannot get capacity elsewhere. It sells two things: Access, an immediate-access inventory service, and Build, co-engineered, owned, and operated clusters typically on three- or five-year terms. The company makes no chips, models, or software; its AI exposure is the financing-and-ownership layer of GPU infrastructure.
Axe is deliberately not a data-center owner of buildings. The CEO frames it as "a virtual data center owner, that doesn't own the buildings, but we own the equipment globally" — though the August 2026 Duos agreements shift toward taking minority equity stakes, expected at 49%, in the entities holding the buildings and power. Customers pay monthly in advance, usually with a 15% to 30% prepayment upfront. Axe arranges the financing of each build, sometimes through SPVs, and at the end of a term owns and redeploys the hardware. The flagship is a 36-month, $260 million build: 2,304 NVIDIA B300 GPUs and 4.8 MW of N+1 power at an unnamed U.S. Tier 3 site.
Business Segments
Competitive Landscape
Axe's filings describe a market where demand is driven by a small number of large technology companies and GPU supply has at times been constrained. The 10-K states that "a substantial portion of current and projected demand for GPU compute infrastructure is driven by a small number of large technology companies — including Google, Amazon, Microsoft, Meta, and Apple." Axe positions itself as a full partner rather than a vendor — handling sourcing, placement, financing, and operations — and at the end of a term it owns the hardware and can redeploy it. It owns no differentiated silicon, model, or building, and the source's criticality read is that customers would switch to alternative GPU cloud providers if it could not deliver.
- AmazonNamed in the 10-K among the large technology companies that drive GPU compute demand and form the competitive backdrop; not discussed individually.
- GoogleNamed in the 10-K among the large technology companies that drive GPU compute demand and form the competitive backdrop; not discussed individually.
- MicrosoftNamed in the 10-K among the large technology companies that drive GPU compute demand and form the competitive backdrop; not discussed individually.
- MetaNamed in the 10-K among the large technology companies that drive GPU compute demand and form the competitive backdrop; not discussed individually.
- AppleNamed in the 10-K among the large technology companies that drive GPU compute demand and form the competitive backdrop; not discussed individually.
Supply Chain
Axe Compute sits one layer up from the silicon: it buys NVIDIA GPUs, places them in third-party and increasingly part-owned data-center and power entities, and rents the resulting clusters. No neighbor in the source material names Axe Compute.