An AI investment fund set up or led by Liquid Capital founder Yilihua and his team has published its first research report, centering on U.S.-listed AI infrastructure company Axe Compute. The report says that from late 2025 through the first half of 2026, repeated upward revisions to AI capital expenditure and strong outperformance in semiconductor and data center stocks pushed AI from one market theme to what it described as the main line of global capital markets. Against that backdrop, the fund said Axe Compute has become one of its key names to watch this year.
The note points to Axe Compute’s July 22 announcement of an additional $1.3 billion in AI infrastructure customer contracts as a major reason behind its conviction. If those contracts are executed in an orderly way and later show up in reported financial results, the report says, the company — which it describes as having a market capitalization of less than $100 million at present — could become one of the most undervalued GPU compute gateways in U.S. equities.
From micro-cap biotech to AI compute infrastructure
Before its rebranding in December 2025, Axe Compute was Predictive Oncology Inc. (NASDAQ: POAI), a micro-cap U.S. biotech company. The report describes its biotech phase as unremarkable: revenue remained in the hundreds of thousands of dollars for years, losses persisted, and its market value stayed in the tens of millions of dollars, drawing little attention from the market.
That changed in September 2025, when the company launched its Strategic Compute Reserve and said Aethir’s native utility token, ATH, would sit at the center of the strategy. The report frames that move as an extension of the crypto treasury narrative and an early signal that the company was shifting toward AI and compute infrastructure.
In October 2025, the company completed two concurrent PIPE financings totaling $343.5 million, made up of $50.8 million in cash and ATH with a stated nominal value of $292.7 million. According to the report, the financing moved the company’s balance sheet from negative equity to a positive gain of $47.7 million and gave it 6.348 billion ATH, binding it closely to the Aethir network and creating what the authors describe as a combined AI compute and treasury-company capital markets story.
On Dec. 11-12, 2025, the company completed a rebrand, changing its name from Predictive Oncology Inc. to Axe Compute Inc. and its ticker from POAI to AGPU while remaining listed on Nasdaq.
By the end of the first quarter of 2026, Axe Compute had begun operating as a new cloud services provider. The report says the market was given a clear signal of full transformation through management and board changes, while an Aethir-related party may have become the largest shareholder. On Feb. 9, Charles L. Nuzum became chairman. Christopher Miglino, who had previously been involved in structuring the ATH transaction, formally took over as CEO. The board was restructured in March, and Kyle Okamoto, formerly Aethir’s CTO/GM, became president.
Key 2026 contracts laid out in the report
The report maps out a string of contracts signed or disclosed by Axe Compute in 2026.
On April 1, the company completed enterprise-grade commercial integration with Aethir’s distributed GPU network, which the report says spans more than 400,000 GPU containers, more than 200 locations, and 93 countries. Axe then signed its first batch of enterprise contracts worth about $12 million. Those contracts mainly fell under the Immediate Access Program and were expected to generate about $835,000 in monthly revenue, using a payment structure based on upfront payment plus monthly prepayment. The report adds that this line had only begun to contribute compute revenue, with about $7,000 recognized in the first quarter of 2026.
On April 22, the company disclosed a $260 million dedicated B300 cluster contract, the first order under its Build Program. The core terms described in the report include a 36-month take-or-pay agreement covering delivery of 2,304 NVIDIA B300 GPUs and AI high-speed storage in a U.S. Tier-3 data center, along with 4.8 MW of dedicated power. Payments include a structured deposit, prepayment, and monthly prepayment. Once the project goes live in the third quarter of 2026, quarterly revenue is projected at around $21 million.
On May 27, Axe confirmed receipt of the first $43 million payment tied to the B300 contract. The report calls this the first real contract cash milestone and says it confirmed that the Build model had started on schedule, with hardware procurement and deployment under way.
On June 16, the company added a $25.9 million long-term deployment contract for Blackwell and Grace Blackwell systems, structured as 12 months plus 24 months with an option to renew. Of that amount, $12.9 million had already been prepaid.
Then on July 22, Axe announced another $1.3 billion in AI infrastructure customer contracts. According to the report, those agreements are based on five-year terms with renewal options, require large upfront prepayments, and include provisions to keep upgrading GPUs as new generations come to market. Revenue from those contracts is expected to begin by late Q4 2026, while prepayments are due in Q3 2026. At that point, annual recurring revenue, or ARR, is expected to exceed $384 million. The report says this $1.3 billion order should be the starting point for the broader market to reassess Axe Compute.
Two product lines: Immediate Access and Build
The report defines Axe Compute as a technology company focused on high-performance computing infrastructure for AI workloads. It sources large-scale GPU capacity from hardware manufacturers and infrastructure providers, then deploys that capacity to enterprise clients through long-term service agreements. Its service scope includes hardware procurement, data center hosting, networking, storage, and financing. Axe still retains a cancer drug discovery solutions business, but the report says that legacy business is no longer the core operation.
The company’s business model is broken into two product lines.
Immediate Access Program
This line targets customers that need rapid onboarding and flexible scaling. Built on Aethir’s existing distributed GPU inventory, it can deploy in as little as 48 hours and reaches more than 200 global nodes. The report says it is suitable for inference, fine-tuning, and small- to mid-sized training workloads, with customers billed monthly based on reserved capacity.
Build Program / AI Factory
This line is aimed at very large, long-duration dedicated compute demand. Axe handles architecture design, data center site selection, power negotiations, hardware financing, and the final enterprise-grade SLA operation. The report characterizes it as a design-deploy-own-operate model.
Its showcase example is the $260 million three-year order signed in April 2026. Axe plans to procure 2,304 NVIDIA B300 GPUs from a U.S. Tier-3 data center facility to build a dedicated cluster, along with AI-optimized high-speed storage infrastructure and 4.8 MW of dedicated redundant power. The client specifies the deployment location and service standards, and deployment is scheduled to be completed in Q3 2026. The structured payment package already delivered an initial $43 million, and the report says Axe should be able to recognize about $21 million in revenue each quarter over the 36-month service period.
In June 2026, the company followed that with a $25.9 million long-term deployment contract covering Blackwell and Grace Blackwell systems for inference infrastructure and simulation platforms, with $12.9 million already received as a prepayment.
By July 2026, the report says, the Build line had secured more than $1.3 billion in five-year AI infrastructure contracts across the United States and Europe, surpassing the company’s full-year target of $1 billion in signed business. Those projects are expected to receive prepayments in Q3 2026, begin recognizing recurring revenue in late Q4 2026, and, once fully deployed and running steadily, push ARR above $384 million. Management said demand remains strong and that this revenue would feed into 2027 ARR.
The CoreWeave comparison in the report
The report spends considerable time comparing Axe with CoreWeave. It describes CoreWeave as a heavy-asset, centralized operator focused on training workloads. CoreWeave runs 49 large AI data centers across North America and Europe and has about 250,000 high-end GPUs, according to the report. With InfiniBand interconnects and Kubernetes-native orchestration, it can build single-site training clusters with tens of thousands of cards, making it well suited for extremely large distributed training jobs run by AI labs such as OpenAI, Meta, and Microsoft.
The report notes that CoreWeave listed on Nasdaq in March 2025 and received an additional $2 billion strategic investment from Nvidia in January 2026, establishing itself as a benchmark in the purpose-built AI cloud or neo-cloud segment. But because its data center footprint is concentrated in North America and Europe, the report says cross-continent latency of 80 to 150 milliseconds and data residency rules in different jurisdictions can limit its reach in parts of Asia-Pacific, the Middle East, and Latin America.
Axe Compute, by contrast, is presented as a hybrid, distributed, globally covered model. On one side, it taps Aethir’s distributed compute network to aggregate third-party data center resources around the world. The report says that network spans more than 200 compute nodes in 93 countries and can provide access to more than 435,000 GPUs. On the other side, Axe is trying to grow a new asset-centric cloud business valued at more than $1 billion, allowing it to address large customized compute demand from a broad base of GPU buyers and AI companies.
Financial review: tiny Q1 revenue, large contracts not yet recognized
The report cites the company’s financial results for the quarter ended March 31, 2026. Axe held $6.9 million in cash and cash equivalents, $20.2 million in ATH digital assets, representing about 2.83 billion tokens, and $9.4 million in current digital asset receivables, for a total liquidity pool of about $36.5 million. Management said that should be enough to support operations through fiscal 2026 and beyond.
Revenue in the first quarter of 2026 came in at $35,000, down from $110,000 in the first quarter of 2025. Most of that first-quarter revenue still came from the legacy drug discovery services segment, with compute services contributing only $7,000. The report stresses that the $43 million first payment for the B300 order, received in May, and the additional $25.9 million long-term Blackwell-series contract announced in June had not yet flowed into the income statement as reported revenue.
Once the $260 million dedicated cluster goes live in Q3, the report says Axe could recognize about $21 million in compute revenue in a single quarter, roughly 600 times its total first-quarter revenue. If the $1.3 billion order begins ramping in Q4, quarterly revenue could increase by another $65 million to $86 million, implying quarter-on-quarter growth of more than 400%. On that basis, the note argues that the company is at the edge of a step change from quarterly revenue measured in tens of thousands of dollars to revenue measured in hundreds of millions.
The company posted a net loss of $7.7 million in Q1 2026. Included in that figure was a $4.3 million non-cash mark-to-market loss on its ATH digital asset holdings. Accounts receivable stood at $659,000 as of March 31, 2026, up from $32,000 on Dec. 31, 2025. The report says both accounts receivable and contract liabilities rose sharply during the quarter, reflecting monthly prepayments due from compute services customers after projects began going live near quarter end.
The note also cites a statement from Axe Compute CEO Christopher Miglino: “Our goal this year was to sign $1 billion worth of contracts, and the July contract puts us well beyond that target... We believe signing another $2 billion worth of contracts this year is not out of reach, and that would help lift next year’s annual recurring revenue.” Combined with his earlier public remarks this year, the report says Axe currently has a potential business pipeline of more than $4 billion, has already signed more than $1 billion in contracts, and is targeting $3 billion in signed contracts for the year.
Valuation work: two models in the note
The report uses two valuation approaches.
Model 1: FY2026E forward price-to-sales
First, it estimates that already announced and confirmed orders imply about $125 million in secured FY2026 revenue. It then notes that three Wall Street analysts forecast average 2026 revenue for AGPU at $163,935,524, with a low estimate of $157,505,455 and a high estimate of $168,752,872. For 2027, the average estimate cited is $254,372,663, with a low of $244,405,017 and a high of $261,853,600. The authors use $125 million as their conservative baseline.
Using a forward P/S multiple of about 3.88x for CoreWeave, the report calculates an implied equity value for Axe of $485 million based on $125 million in revenue. With total shares outstanding of 11.385 million and a current share price of $6.85, that works out to an implied value of about $42.60 per share, or roughly 6.21x the current price.
Model 2: P/ARR scenario analysis
The second model uses price-to-annual recurring revenue, a metric the report says is standard in compute infrastructure because it fits long-term contracted business models better than near-term sales. It uses CoreWeave’s July 2026 P/ARR valuation center of about 2.4x as a benchmark for a mature compute services provider.
On that basis, and using the report’s figure of $384 million in long-run ARR from Axe’s Build contracts currently in hand, the implied equity value comes to $921.6 million. Dividing that by 11.385 million shares yields an implied target price of about $80.94 per share, or about 11.8x the current price.
Putting the two together, the report says Axe shares could have 6x to 11x upside. It also acknowledges that the calculations do not apply a discount for differences in scale and maturity between the two companies, so a lower fair-value center is possible in practice.
The note also includes a relative comparison. It says AGPU’s market capitalization is only about $80 million at present, while guided ARR based on long-term contracts already signed is $384 million, implying a P/ARR of only 0.2x. By comparison, the report lists P/S on ARR multiples of 6.9x for Nebius, 2.4x for CoreWeave, 4.0x for IREN, and 10.4x for WhiteFiber. Even after allowing for AGPU’s early commercialization stage and the fact that revenue recognition has not yet fully caught up with contracts, the report says the stock still screens well below the group.
The “Compute + Treasury” structure tied to ATH
Beyond the operating business, the report argues that AGPU has an additional flywheel built around its ATH treasury strategy. Unlike treasury companies that simply hold BTC or ETH, the report says, ATH sits inside a company whose operating business can produce related cash flow. Compute orders can drive ATH demand and settlement activity, while gains in the treasury can feed back into compute expansion.
What Aethir and ATH do in the structure
The report says the Aethir network was developed by DCI Foundation, a Panamanian foundation company, and is a decentralized physical infrastructure network. It aggregates enterprise-grade GPUs contributed by independent data centers, businesses, and other hardware owners into a globally distributed network for AI training and inference, cloud gaming, and other virtualized compute workloads. In many cases, the report says, the service can be cheaper than centralized cloud providers.
Three roles support the network: “containers,” which execute the compute; “checkers,” which test and monitor containers to verify integrity and performance; and “indexers,” which match users with appropriate compute resources. Buyers obtain compute through the network.
ATH functions as the proxy unit for GPU computing power and as the transaction and incentive token for participants. To become a compute resource provider, participants must obtain ATH and stake it as collateral before contributing GPU resources and qualifying to process jobs. After compute is delivered and verified, ATH moves from compute users to resource providers as payment and reward. Providers can earn ATH through proof-of-capacity rewards, proof-of-delivery rewards, and service fees paid by users, then choose to re-stake, hold, lend, or sell those tokens. The Aethir treasury manages protocol fees and ATH allocated for protocol development, while the blockchain settlement layer records transactions and facilitates token transfers.
How the report describes Axe’s capital design
The report says Axe’s ATH treasury architecture is not just a “buy and hold” strategy. Instead, it has two layers that bind Axe Compute’s operating company and the Aethir ecosystem together into a business-capital-token loop.
The first layer is that Axe’s Access business runs on the Aethir network. The report says AGPU’s Immediate Access model depends entirely on Aethir’s distributed GPU network, covering more than 400,000 GPU containers, more than 200 locations, and 93 countries. When enterprise clients place orders through Axe’s Access platform, the workloads are executed directly through Aethir. In the report’s framing, every Access order therefore consumes or requires staking of ATH, creating token demand while also bringing in prepaid operating cash flow.
The second layer is the treasury strategy itself. The report calls this an upgraded version of the BTC or ETH treasury model. It contrasts the approach with companies such as MicroStrategy, which it says hold BTC passively as “digital gold” and depend mainly on external market cycles. By contrast, the report says AGPU’s ATH is embedded inside a compute business that generates positive cash flow, while the Aethir network uses ATH for staking and settlement. Axe’s Access business operates on top of that network, creating a tighter operating loop.
Two flywheels in the report’s framework
The report describes the interaction between AGPU and ATH through two flywheels.
- A business flywheel: orders land, Access activity grows, ATH demand increases, ATH appreciates internally, Axe’s balance sheet expands, AGPU’s value rises, and the company can win more AI compute orders.
- A capital flywheel: orders improve operating performance and available funds, the company buys more ATH and expands holdings, ATH appreciates externally, Axe’s balance sheet grows, AGPU’s value rises, and more AI compute orders follow.
In the authors’ view, those nested loops make both operating performance and ATH pricing relevant to AGPU’s market value and growth trajectory, and could represent a new AI-and-crypto capital model.
Risks highlighted by the report
The note closes with a detailed risk section and says investors should assess the name carefully.
1. Contract execution and delivery risk
The report says the Build Program is central to Axe’s move from a lighter Access model toward more customized, semi-heavy-asset cluster deployments. Although the company has signed the $260 million B300 cluster and the later $1.3 billion global customer contracts, there are still execution risks tied to hardware procurement, data center coordination, power deployment, and enterprise SLA delivery. If projects are delayed or customer acceptance slips, revenue recognition could be pushed back, affecting cash flow and market confidence.
2. Revenue conversion and financial verification risk
The note emphasizes that Q1 2026 revenue was only $35,000 and that compute services contributed very little, even though signed contracts now total nearly $1.6 billion. The average 2026 revenue forecast cited from Wall Street is about $164 million, but those numbers assume conversion in the second half of the year. If contract ramp-up is slower than expected, actual revenue could come in far below consensus. The report also says non-cash ATH mark-to-market losses may keep introducing volatility, while rising accounts receivable and contract liabilities show the credit and collection risks that can still exist even under prepayment-based structures.
3. Macro and valuation risk
The report adds that if AI capital expenditure is cut because of slower economic growth or faster technology shifts, customer demand may weaken. Tighter GPU supply, higher energy costs, and stricter data center compliance rules could all raise execution costs. The forward P/S and P/ARR frameworks in the report both depend on the assumption that the financials eventually catch up with the contract pipeline, while fair valuation may still require a discount because of differences in size and maturity versus larger peers.
The report says Axe Compute’s narrative is still ahead of financial realization. The pace of revenue recognition, together with the next quarter or the next half year of reported results, will be the key verification window. It also says the listed risks are not exhaustive and that investors should conduct their own due diligence and make decisions based on their own risk tolerance.
Bottom line of the report
The research note’s conclusion is that AGPU completed its shift from traditional biotech to an AI GPU compute entry point in less than a year. It says the company now combines a light-asset Access model with a large-scale cluster build-and-lease model, and has already accumulated about $1.6 billion in contract value. Combined with the “Compute + Treasury” structure, the report argues that AGPU offers 6x to 11x upside from the current share price and should be watched closely as a high-beta name in the broader trend of AI compute assetization.