The AI compute assetization wave: Axe Compute may be the most undervalued GPU compute entry point in US stocks

By the end of 2025 to the first half of 2026, as other tracks (some overvalued growth, cyclical, and pure narrative themes) faded, while AI Capex continued to be raised beyond expectations and semiconductor and data-center-related stocks surged significantly, market consensus truly formed—“AI is no longer just one of the themes; it has become the absolute mainline of global capital markets.” With research into undervalued AI companies in the US stock market, Axe Compute has become a key company we focus on this year. A recent announcement on July 22 added a $1.3 billion AI compute capacity contract, greatly strengthening our confidence in an investment in Axe. If the contract can land smoothly and data can genuinely be reflected in future financial reports, we believe that “Axe Compute, a company with a current market cap of less than $100 million, may become the most undervalued GPU Compute entry point in US stocks.”

I. A glamorous pivot from pharma to AI compute infrastructure

Before it changed its name in December 2025, Axe Compute’s predecessor was Predictive Oncology Inc. (NASDAQ: POAI), a typical US micro-cap biotech company. As a classic “small-cap biotech,” POAI delivered mediocre performance in the drug development phase: revenue stayed for a long time at the hundreds of thousands of dollars level, with continued losses, and a market capitalization that hovered in the tens of millions of dollars range for years. It received very little attention from the capital markets.

In September 2025, the company suddenly launched a Strategic Compute Reserve, clearly making Aethir’s native utility token (ATH) as the core, continuing the crypto treasury strategy narrative, implying the company would pivot toward AI narrative and compute business.

In October 2025, it completed two simultaneous PIPE financings, totaling $343.5 million in cash, using a hybrid structure funded by $50.8 million cash plus $292.7 million in nominal value of ATH. Through this financing, the company’s balance sheet moved from negative equity to positive $47.7 million in gains, obtaining 6.35B ATH, formally creating a deep binding between the company and the Aethir network. This enabled a capital operation model combining an AI compute narrative and a treasury company overlay; from then on, the company entered our observation scope.

On December 11–12, 2025, the company rebranded. Its name changed from Predictive Oncology Inc. → Axe Compute Inc., and its ticker from POAI → AGPU, continuing to trade on Nasdaq.

By the end of Q1 2026, Axe Compute officially began operating as a new cloud services provider. Aethir-related crypto project entities may become its largest shareholder, sending a signal of a full-scale transformation to the financial market:

On February 9, Charles L. Nuzum became Chairman of the Board; Christopher Miglino (who had previously participated in the ATH transaction structure design) officially became CEO. In March, the board was reorganized, and Kyle Okamoto (former Aethir CTO/GM) became President.

On April 1, the company completed enterprise-grade commercial onboarding of Aethir’s distributed GPU network (400,000+ GPU containers, 200+ locations, 93 countries), signing its first batch of approximately $12 million enterprise contracts. The contracts mainly come from the Immediate Access Program, contributing expected revenue of about $835k per month. The payment terms are prepayment plus monthly prepayment. It has already started contributing a small amount of compute revenue (Q1 actual recognition was about $7,000).

On April 22, 2026, it disclosed a $260 million B300 dedicated cluster contract (Build Program’s first deal). The key contract terms: a 36-month take-or-pay agreement, delivering 2,304 NVIDIA B300 GPUs plus AI high-speed storage (US Tier-3 data centers, 4.8 MW dedicated power). Structured deposit + prepayment + monthly prepayment arrangements. After going live in Q3 2026, the quarterly revenue is expected to be about $21 million.

On May 27, 2026: confirmed receipt of the first payment of $43 million under the B300 contract. This is the first true contractual cash milestone, confirming that the Build model has started as planned, with hardware procurement and deployment underway.

On June 16, 2026: the company landed a $25.9 million long-term deployment contract for Blackwell / Grace Blackwell (12 months + 24 months, extendable), of which $12.9 million has been prepaid.

On July 22, 2026: it announced an additional $1.3 billion in AI infrastructure customer contract additions. These agreements are based on five-year contracts and can be renewed; large upfront prepayments are required. They also include terms for continuous GPU upgrades as new-generation GPUs are launched. Revenue is expected to start generating by the end of Q4 2026, and prepayments will be made in Q3 2026. At that time, annual recurring revenue (ARR) will exceed $384 million. This $1.3 billion big order should become the true starting point for the whole market to rethink Axe.

II. Multiple AI compute solutions with high elasticity: Axe’s business model broken down like “Coreweave”

Axe Compute Inc. is a technology company focused on providing high-performance computing infrastructure specifically for artificial intelligence (AI) workloads. It obtains large-scale GPU capacity from hardware manufacturers and infrastructure providers, then deploys it to enterprise customers through long-term service agreements. Service scope includes hardware procurement, data-center colocation, networking, storage, and financing. Axe also retains a cancer drug R&D solutions business, but it is not currently the company’s main business.

1. Axe’s business is divided into two product lines:

(1) Immediate Access Program

For customers that need to go live quickly and expand flexibly. Relying on Aethir’s distributed network’s existing GPU inventory, deployments can be completed in as fast as 48 hours, covering more than 200 global nodes. Suitable for inference, fine-tuning, and small-to-medium-scale training. Customers pay monthly based on reserved capacity.

(2) Build Program / AI Factory

For ultra-large-scale, long-cycle dedicated compute needs. Axe is responsible for end-to-end architecture design, data-center site selection and power negotiations, hardware financing arrangements, and ultimately enterprise-grade SLA (Service Level Agreement) operations—“design-deploy-own-operate.”

The $260 million three-year mega-deal landed in April 2026 is a flagship example of this model. The company plans to procure from a US third-tier data-center facility 2,304 NVIDIA B300 GPU cards to build a dedicated cluster and AI-dedicated high-speed storage infrastructure, supported by 4.8 MW dedicated redundant power. The customer designates deployment locations and service standards, with the deployment work plan completed in Q3 2026. Using a structured payment arrangement, the initial $43 million has already been received. Over the 36-month service period, the company will recognize approximately $21 million in revenue each quarter.

In June 2026, the company also signed a $25.9 million long-term deployment contract for Blackwell and Grace Blackwell, covering two major scenarios: inference infrastructure and a simulation platform. $12.9 million has already been received in the form of a prepayment.

In July 2026, the Build business line again secured five-year AI infrastructure long-term contracts totaling over $1.3 billion across the US and Europe, far exceeding the company’s full-year $1 billion signing target. Prepayments will be received in Q3 2026, and starting at the end of Q4 2026 the company will formally confirm continuing revenue. Once all clusters are deployed and operate steadily, the corresponding annual recurring revenue will exceed $384 million. Company management said that current market demand is strong, and related revenue will be counted toward 2027 annual recurring revenue, continuously opening up long-term growth space.

2. Re-understanding Axe’s Build compute business: the best comparison object is CoreWeave—one is the leader in centralized training, and the other is a new global hybrid compute force:

CoreWeave takes the route of heavy-asset, centralized, and deeply focused on training scenarios. It operates 49 large AI data centers across North America and Europe, with about 250k high-end GPUs. It builds single-room “tens of thousands of cards” training clusters through an InfiniBand high-speed interconnect network and Kubernetes-native orchestration. The extreme performance of ultra-large-scale distributed training is ideal for trillion-parameter training by top AI labs such as OpenAI, Meta, and Microsoft. It listed on Nasdaq in March 2025, and received a $2 billion strategic investment from Nvidia in January 2026, becoming a benchmark company for dedicated AI compute cloud (Neo-Cloud). However, because its data centers are all concentrated in North America and Europe, cross-continent transmission brings 80–150 milliseconds of network latency. Combined with data residency and compliance requirements in each country, CoreWeave is largely kept out of many regional markets such as Asia-Pacific, the Middle East, and Latin America.

Axe Compute, in contrast, follows a hybrid, distributed, globally covered route. On one hand, it integrates global third-party data-center resources via Aethir’s distributed compute network. It deploys more than 200 compute nodes across 93 countries, providing access to a total of over 435,000 GPUs. On the other hand, it is aggressively expanding a new cloud business centered on assets totaling $1 billion+ in value. This enables it to enter the large-scale custom compute market, serving all types of GPU buyers and AI companies.

3. Financial analysis:

Axe Compute’s first-quarter financial performance as of March 31, 2026

As of March 31, 2026, the company held $6.9 million in cash and cash equivalents, $20.2 million in ATH digital asset holdings (about 835k tokens), and $9.4 million in current digital asset receivables, totaling about $36.5 million in a liquidity pool. Management believes this is sufficient to support the company’s operations through fiscal year 2026 and beyond.

Revenue in Q1 2026 was $35k, while Q1 2025 was $110k. Q1 2026 sales mainly came from the traditional drug discovery services segment, while the compute services segment contributed only $7,000. According to the company’s disclosures, the first payment of the $43 million B300 deal arrived in May, and in June it added another $25.9 million long-term contract for the Blackwell series; neither has been converted into income statement revenue yet.

Once the $260 million dedicated cluster goes live in the third quarter, the compute revenue can be recognized at about $21 million per quarter—equivalent to 600 times total revenue in the first quarter. Assuming the $1.3 billion order goes live in the fourth quarter, quarterly revenue could further increase by $65 million to $86 million, with quarter-over-quarter growth of more than 400%. The company is standing right at the explosive inflection point jumping from $100,000-level quarterly revenue to billion-level quarterly revenue, and current market pricing still fails to fully reflect the certainty of this stepwise surge in revenue.

Axe Compute’s net loss in Q1 2026 was $7.7 million. The net loss includes a non-cash mark-to-market loss of $4.3 million on the company’s ATH digital asset holdings. As of March 31, 2026, accounts receivable were $659k, versus $320,000 on December 31, 2025. Both accounts receivable and contract liabilities increased significantly this quarter, reflecting that after projects go live at the end of Q1, Compute Services customers need to pay monthly prepayments.

Axe Compute CEO Christopher Miglino said: “Our goal this year is to sign $1 billion worth of contracts. The July contract landing has already put us far beyond that goal… We believe it’s not out of reach to sign another $2 billion worth of contracts this year, which will help increase next year’s annual recurring revenue (ARR).” Combined with its public statements in the first half, Axe Compute’s current potential business orders exceed $4 billion; it has already signed $1 billion+ in contracts, and the company’s target is to sign $3 billion in contracts in total this year.

4. Valuation analysis:

Model 1: FY2026E forward P/S (Forward Price-to-Sales)

Estimated annual revenue

Below are official confirmed orders already released. Based on these contracts, the confirmed revenue for FY2026 is about $125 million.

Three Wall Street analysts predict AGPU’s average revenue in 2026 at $163,935,524. The lowest forecast is $157,505,455 and the highest forecast is $168,752,872. In 2027, this value reaches $254,372,663, with the lowest revenue forecast at $244,405,017 and the highest at $261,853,600. Currently, using our conservative estimate based on confirmed revenue of about $125 million.

CoreWeave Forward PS is about 3.88x. Axe Compute’s 2026 confirmed revenue is about $125 million; total shares outstanding are 250k; current price is $6.85.

Axe market cap = $125 million × 3.88 = $485 million

Implied target share price: $2.83B ÷ 35k shares ≈ $42.60 per share

Upside multiple vs. current price: $42.60 ÷ $6.85 ≈ 6.21x

Model 2: P/ARR (forward scenario estimation)

P/ARR (Annual Recurring Revenue) is a steady-state valuation metric commonly used across the compute infrastructure industry: it is the ratio of total market capitalization to annual recurring revenue. It fits a business model centered on multi-year locked compute contracts and reflects the intrinsic value of long-term stable cash flows in enterprises more effectively. Here, we use the industry leader CoreWeave’s July 2026 P/ARR valuation “center” of about 2.4x as the fair pricing benchmark for a mature compute services provider.

As of now, all Build business long-term orders on hand can reach annualized recurring revenue (ARR) of up to $384 million on a forward steady-state basis.

Axe forward fair total market cap = $384 million × 2.4 = $921.6 million

Implied target share price = $921.6 million ÷ 110k shares ≈ $80.94 per share

Upside space vs. current price: $80.94 ÷ $6.85 ≈ 11.8x

Based on the combined calculations, Axe’s share price has 6–11x upside potential. The current market value is seriously undervaluing it. These estimates do not apply valuation discounts for differences in business scale and development maturity between the two companies; the actual reasonable valuation “center” may have room to move downward.

From a horizontal peer comparison, AGPU’s current market valuation is significantly mismatched with its business scale and growth potential. As of now, the company’s market cap is only about $80 million, while based on ongoing long-term contract estimates from already landed agreements, its guided ARR has reached $384 million, implying a P/ARR of only 0.2x. By comparison, peers Nebius, CoreWeave, IREN, and WhiteFiber have P/S on ARR of 6.9x, 2.4x, 4.0x, and 10.4x, respectively. Even considering that AGPU is still in an early commercialization stage and its revenue recognition cadence has not fully been released, its valuation remains far below the industry average. As the B300 dedicated cluster and subsequent $1 billion+ contracts gradually contribute revenue in the second half of 2026, AGPU’s ARR could quickly be realized in the next year. The extremely low valuation multiple provides investors with significant margin of safety and upside elasticity.

III. Capital-model design for “AI x Crypto”: a “Compute + Treasury” dual-wheel driving model

Besides the Compute business with astonishing data, AGPU has another flywheel with huge imagination space: the ATH Treasury strategy. Unlike a traditional treasury company that just holds BTC or ETH, ATH is placed inside a company whose associated business generates positive cash flow. Compute orders directly drive ATH demand and settlement; Treasury appreciation in turn supports Compute expansion. The two are mutually causal and resonate together, forming a self-reinforcing positive flywheel.

1. What are ATH and Aethir?

The Aethir network is a decentralized physical infrastructure network developed by DC I Foundation (abbreviated as “DCI”), a company under the Panama Foundation. The Aethir network aggregates enterprise-level GPUs contributed by independent data centers, enterprises, and other hardware owners into a global distributed network. The network is designed to provide on-demand GPU compute resources for AI training and inference, cloud gaming, and other virtualized computing workloads, and in many cases its pricing is lower than that of centralized cloud providers. Within the network, three roles jointly ensure the availability, suitability, and quality of compute resources: “containers,” which are the actual execution containers for compute; “verifiers,” which test and monitor containers to validate their integrity and performance; and “indexers,” which match compute resource users with the appropriate containers. Buyers of compute resources use the network’s compute resources.

ATH, as a proxy unit for GPU compute capability, is the trading medium and incentive unit for Aethir network participants. To become a provider of compute resources, network participants need to obtain ATH and stake it as collateral, so they can contribute GPU resources and become eligible to process compute requests. After compute resources are delivered and verified, ATH flows from compute users to resource providers as payment and rewards. Resource providers earn ATH through “Proof of Capacity” rewards (to maintain availability and readiness) and “Proof of Delivery” rewards (to complete workload tasks), as well as service fees paid by compute users. Service providers can re-stake, hold, lend, or sell the ATH they receive. The Aethir treasury manages protocol fees and the allocation of ATH to protocol development, while the blockchain settlement layer records transactions and facilitates ATH transfers.

2. Axe and ATH: capital design

ATH Treasury’s capital architecture is not simply “buying coins and holding them.” Through a two-layer design, it tightly binds Axe Compute’s operating entity with the Aethir (ATH) ecosystem, forming a three-in-one closed loop of business-capital-token. The core advantage of this design is: each time a Compute order is executed, it converts into incremental ATH demand and value capture—not like a traditional treasury company that relies only on external market liquidity and sentiment.

(1) Axe’s Access business is built on the Aethir network

The Access model (immediate access) is the core of AGPU’s light-asset expansion, and its underlying foundation fully relies on the Aethir distributed GPU network (400,000+ GPU containers, 200+ locations, covered across 93 countries). After customers place orders through AGPU’s Access platform, compute tasks are sent directly into the Aethir network for execution. Each invocation of orders consumes or stakes ATH, creating real demand.

Each Access order = directly drives ATH demand + generates positive cash flow (prepaid revenue). This design makes AGPU’s Access business a “natural demand engine” for ATH. Linking business growth with the ATH price—more orders mean more ATH consumption/staking, strengthening price support.

(2) Axe’s treasury strategy: holding ATH and forming a strategic reserve

AGPU’s Treasury strategy is an upgraded version of BTC/ETH treasury. Companies like MicroStrategy passively hold BTC as “digital gold,” with returns depending on external Bitcoin halving cycles and market cycles, lacking endogenous cash-flow support. AGPU’s ATH is “placed inside” a compute business that generates positive cash flow. ATH is not just a reserve: the Aethir network uses ATH staking and settlement to run it, and Axe’s Access business operates on top of it, naturally forming a closed loop.

3. How the positive flywheel between AGPU and ATH works

(1) AGPU and ATH business flywheel: an “order-demand-value appreciation” cycle driven by the Access model

When orders land, Access orders grow — ATH demand rises — endogenous ATH value appreciation — Axe’s balance sheet expands — AGPU value increases — more AI compute orders are obtained

(2) AGPU and ATH capital flywheel: a “performance-funds-increasing holdings” cycle driven by Treasury value appreciation

Orders land — company performance improves, available funds increase — purchase more ATH, expanding ATH holdings — ATH appreciates externally — Axe’s balance sheet expands — AGPU value increases — more AI compute orders are obtained

The business flywheel provides endogenous demand and cash flow (Access orders directly “feed” ATH demand). The capital flywheel provides leveraged value appreciation and asset expansion (Treasury buying amplifies the ATH price effect). Together, they form two deeply nested driving models—performance and the ATH price together push up both AGPU’s share price and performance expansion. This model may become a new paradigm for “AI x Crypto” capital models.

IV. Potential risk variables for Axe

Axe Compute’s story has huge room for imagination, and current market pricing does not reflect optimistic expectations about future contract execution and the value of ATH reserves. However, as with any high-elastic growth stock, the story leads ahead of financial delivery. Valuation is more based on the delivery of future GPU contracts and the performance of ATH price—not on already confirmed revenue and profits. Historical revenue remains at a very low base, and the true conversion of large orders and financial statement validation still take time. The following are the main risk variables that investors need to evaluate carefully.

1. Contract execution and delivery risk

The Build Program is the key for AGPU’s transition from a light-asset Access model to a semi-heavy-asset bespoke cluster model. The $260 million B300 dedicated cluster (after a Q3 2026 go-live, quarterly revenue of about $21 million) and the subsequent $1.3 billion global customer contracts have been signed. But there are still execution risks in hardware procurement, data-center coordination, power deployment, and enterprise-grade SLA onboarding. If the project cannot go live on time or customer acceptance is delayed, revenue recognition will be postponed, and cash flow and market confidence may be impacted.

2. Revenue conversion and financial report validation risk

In Q1 2026, revenue was only $35,000 (compute services contribution is minimal), while the company has signed contracts totaling about $1.6 billion that have not yet been largely converted into revenue. Wall Street analysts’ average revenue expectation for 2026 is about $164 million, but those forecasts include conversion assumptions for the second half. If order execution lands slower than expected, actual revenue could be far below consensus. Non-cash ATH mark-to-market losses will continue to fluctuate, and the increase in accounts receivable and contract liabilities also reflects potential bad-debt risks under a prepayment model.

3. Macro and market valuation risk

If AI Capex is scaled back due to economic slowdown or technical iteration, order demand may be affected. Tightening in GPU supply, energy costs, and data-center compliance requirements could raise execution costs. Both Forward P/S and P/ARR calculations are based on assumptions that financial data can be realized. In practice, the reasonable valuation “center” may face valuation discounts due to differences in scale and maturity; and current high elasticity also means that volatility is amplified.

Overall, Axe Compute’s story is ahead of financial delivery. The cadence of revenue recognition and the next quarter or half-year financial reports will be the key validation window. The risk variables above are not exhaustive. Investors should conduct their own due diligence, fully understand relevant risks, and make independent decisions based on their own risk tolerance.

Overall summary: In less than a year, AGPU has completed a glamorous pivot from traditional biotech into an AI GPU compute entry point. Its business model covers a hybrid AI compute solution combining a light-asset “Access model” and large-scale cluster build-and-rental “Build model,” and it has achieved astonishing $1.6 billion in order contracts data. Combined with the “Compute + Treasury” dual-wheel model, and an integrated assessment of its business model, asset reserves, and an unpriced valuation, relative to the current stock price, AGPU has 6–11x growth potential as a highly elastic stock worth重点 attention amid the wave of AI compute assetization.