On August 12, Nebius, a leading company in the Neo Cloud track, saw a 28% surge in a single day.
Interestingly, just a few days ago, Michael Burry, who became famous for shorting the housing market, revealed a short position against this company, with nearly 30% of the float being shorted. Following the financial report release, short sellers were forced to cover their positions, resulting in Nebius's largest single-day gain since September 2025.
The day before, after-market close, CoreWeave, another key player in the Neo Cloud track along with Nebius, also released its second-quarter earnings report: with a revenue of $25.8 billion, more than doubling year-over-year, and an order backlog surging to $104 billion, over three times what it was a year ago. The full-year guidance was also raised, leading to a 19% increase in stock price the next day. Companies like IREN, Applied Digital, and Core Scientific in the same track also followed suit. Additionally, Roundhill launched a Neo Cloud-themed ETF during the same week.
In this August tech stock rebound, Neo Cloud emerged as a crucial and most vigorous track, displaying one of the most significant rebounds in the tech sector this year.
Over the past year, the market has been discussing who can develop the next big model; now, more and more funds are starting to ask a more realistic question: if models continue to grow in size, who will run them?
The answer is not just NVIDIA. For a GPU to truly become sellable computing power, it needs to be accompanied by a data center, power supply, cooling system, network infrastructure, operations and maintenance, and a pre-order from a willing customer. While GPUs are still in transit, transformers might have been on backorder for half a year; model companies are eager to expand, yet data center vacancies are decreasing.
As a result, a group of new companies specializing in building AI infrastructure for clients are gaining more attention in the capital market. They are not selling a regular cloud account but an entire environment capable of running large models: customers do not need to buy their own machines, compete for power supply, or build their own data center teams from scratch. These companies have a new name, called Neo Cloud providers.
Apart from the well-known names like CoreWeave and Nebius, there is another company on the Nasdaq exchange that has recently rebranded and is making a mark in the race.
It is called Axe Compute, with the stock symbol AGPU.
On August 14, Axe Compute released its second-quarter earnings report. The report revealed that the cumulative contracted value for 2026 has surpassed $3.2 billion; if all these projects are fully deployed, the projected annualized revenue run rate is expected to exceed $696 million.
It is clearly not a company on the same level as CoreWeave. However, for a company that only changed its name at the end of last year and truly started selling computing power this year, these disclosed data are worth looking into.
What Kind of Company is Axe Compute?
Axe Compute is the new brand and strategy after the renaming of Predictive Oncology.
The company previously focused on cancer drug discovery, switched to AI computing infrastructure by the end of 2025, officially rebranded on December 11, 2025, and began trading on NASDAQ the next day using AGPU. In February 2026, the management underwent changes, with Christopher Miglino becoming CEO, Kyle Okamoto joining as President in April, and Jeremy Yaukey-Witter being promoted to CFO in May.
Specifically regarding its business lines, Axe Compute now has two.
The first business line is Axe Compute Access, scheduling third-party GPU capacity that is already online to provide customers with immediate computing power. This line mainly accesses global third-party data center resources through the Aethir distributed computing network, with a network covering 93 countries and over 200 nodes, providing access to over 435,000 GPUs.
The second business line is Axe Compute Build, designing, deploying, owning, and operating dedicated GPU clusters for enterprise customers. Customers do not need to buy machines, find locations, or take on hardware ownership and maintenance responsibilities. The ownership of the GPU infrastructure remains with Axe Compute, and theoretically, these GPUs can continue to serve the next customer after the contract period ends.
Evidently, the Build business line better reflects the company's ambitions.
In April of this year, Axe Compute secured its first landmark Build contract: valued at approximately $260 million, with a term of 36 months. The plan is to deploy 2,304 NVIDIA Blackwell B300 GPUs with accompanying high-speed storage at a Tier 3 data center in the United States, aiming to go live in the third quarter. If successfully entering the service period, this contract could bring in revenue of around $20 to $21 million per quarter.
At the other end of the contract, it landed in a city called Columbus in the southeastern United States. This place was originally nameless on the U.S. data center map. However, due to the overloaded power capacity and interconnection schedule of first-tier hubs, AI data centers have been migrating in large numbers to these second-tier cities in the past two years.
The first B300 cluster of Axe Compute is installed in Duos Edge AI's Tier 3 data center campus located in Columbus. With 288 nodes, each node has 8 B300 cards with 288GB of HBM3e memory, totaling the specified 2304 cards in the contract. The nodes are interconnected with InfiniBand Quantum-X800, with a single-node bandwidth of 6.4 Tb/s, and equipped with over 20PB of high-speed storage provided by WEKA.
Interestingly, during the earnings call for the second quarter on August 17, the company directly broadcasted from this data center. The first sentence of the opening was: "Today we are live from Columbus, Georgia, this is where the B300 cluster announced in April is located."
Duos' CEO, Doug Recker, was also present on-site, and he described the construction progress as follows: "What has been achieved in these 45 days is incredible. If you set up a camera and walk around the site, you will find that the progress is astonishingly fast." He added, "We're not going to tell you we can do 100 megawatts in three weeks, but what you see here was all done within 45 days."
The cluster is expected to go live in the coming weeks and is planned to triple in size over the next few months. At the same time, another cluster is being built in Boden, Sweden, with 256 B300 nodes plus spares, RoCEv2 networking, and WEKA storage, scheduled to go live by the end of 2026.
After that, the volume of signings surged like a tide starting from July.
On July 22, the company announced over $1.3 billion in new contracts in the United States and Europe, surpassing the $1 billion annual contract target set in May; five days later on July 27, they unveiled another contract, exceeding $1.5 billion over five years, to deploy a dedicated cluster in the United States with over 9,200 B300s, both contracts including extension options. By mid-August, the cumulative contract value for 2026 had surpassed $3.2 billion, with over $2.8 billion coming from three Build contracts signed in July.
In three months, they tripled the annual target.
Securing $3.2 Billion in Orders in Six Months
According to the Axe Compute Q2 financial report, the most noticeable change is in revenue.
The company's quarterly revenue was $3.215 million, compared to just about $35,000 in the previous quarter, showing an almost 90-fold increase quarter-over-quarter; as a comparison, this figure was $3,000 for the same period last year. The operating cash flow for the first half of the year was positive $17.4 million, with $21.9 million in cash at the end of the quarter, compared to $6.9 million at the end of the first quarter.
The revenue cost for the second quarter was $3.013 million, roughly estimated, with a gross profit of only about $200,000, resulting in a gross margin of 6.3%.
Whether 6.3% is considered high or low, we cannot judge just by looking at Axe Compute; let's compare with the leader in this field.
In the same quarter, CoreWeave recorded revenue of $2.58 billion, adjusted EBITDA of $1.51 billion, and an EBITDA margin of 59%. Although the revenue scales of the two companies differ by a factor of eight hundred, making them incomparable, the difference in gross profit structure still sheds some light on certain issues.
Because the gap in gross profit structure is not due to scale but rather the nature of the business.
Most of Axe Compute's revenue in the second quarter mainly came from Access, which is essentially a resale. Axe Compute purchases available capacity from third-party networks and resells it to customers at a markup, earning the price difference in the middle. After all, they don't own the hardware, the data centers, and their pricing power is limited. The gross profit ceiling of this type of business is inherently low. In contrast, CoreWeave and Nebius sell GPUs they purchase, install, and operate themselves, with customers signing multi-year pay-or-don't-negotiate contracts. They earn returns on assets, not matchmaking commissions.
This is also why Axe Compute shifted its focus to the Build track starting in the second quarter.
Perhaps Axe Compute realized that the Access model had reached its limit. While these numbers may be impressive for a company just starting to sell computing power, they are not enough to support Axe Compute's true ambitions. Axe Compute needs its own GPU infrastructure, so far, Axe Compute has secured over $3.2 billion in contract orders, and the management team has estimated a gross margin of 28% to 44% for the Build project.
We all know that the deadline for the second quarter is June 30. Therefore, the three Build contracts signed in July, totaling over $2.8 billion, are explicitly marked as post-quarter events in the company's financial report. It can be said that the second quarter financial report the market is currently seeing is actually Axe Compute's last statement before its transformation.
In the announcement of the $1.5 billion contract on July 27, the company stated that it expects to receive cumulative prepayments of over $534 million within 30 days. During the August 17 call, Miglino announced the first payment: over $317 million in prepayments for the expanding cluster had been received.
In other words, the $3.2 billion in contracts and the $300 million prepayment that has already been received are perhaps more critical data in the Q2 report than revenue.
This is the most counterintuitive aspect of the Neo Cloud track: securing future cash flows locked by long-term paid-until-not-discussed contracts can be used as collateral. The $10.4 billion order backlog at CoreWeave is essentially a risk mitigation mechanism that locks in future income before asset depreciation, allowing the company to borrow at a lower cost. In March 2026, CoreWeave used these contracts as collateral to secure a loan of around $8.5 billion, becoming the industry's first to receive an investment-grade-rated GPU collateralized loan. Nebius is following the same path, with over 70% of its new deals in the second quarter including prepayments, expecting to receive over $9 billion in the full year.
Major players are willing to act as the "prepaid counterparty." One key motivation is that procuring computing power from Neo Cloud can be categorized as operating expenses over the contract period, rather than a one-time capital expenditure as with a self-built data center, which would strain the already stressed free cash flow. Thus, the order has been completely reversed. In the past, the supplier would invest money to build the infrastructure first, then wait for the customer to gradually migrate; now, customers lock in capacity and pay first, while suppliers use the contracts and prepayments to purchase GPUs, reserve data centers, and secure financing.
A cash inflow of $317 million means a significant portion of the external financing required for this deployment has been directly compressed, allowing the company to immediately initiate equipment procurement and secure data center capacity without needing to pursue an equity or debt round first. The fact that customers are willing to send money before the cluster is up and running is the most direct endorsement of both the authenticity of the order and its stickiness. For downstream project funders, this cash receipt can also serve as evidence for due diligence.
Axe Compute has also provided a glimpse of its monetization trajectory in the coming period: with an annualized revenue run rate of $370 million by the end of the second quarter, rising to around $1.39 billion post the Columbus cluster going online, and surpassing $6.96 billion after the deployment of all contracted agreements, nearly doubling the previous $3.85 billion guidance, with the window pointing to the fourth quarter of 2026 to the first quarter of 2027.
Furthermore, the company disclosed a qualified sales pipeline scale of $5.9 billion, with 98 opportunities, close to twice the size of the currently signed contracts. Miglino stated during the performance call on August 17, “We still feel confident about signing an additional $2 billion in contracts by the end of this year.”
From Accessing Compute Power to Controlling Power Supply, Axe Compute Aims for Deeper Integration
In the AI infrastructure realm, while GPUs are the most visible term, the truly scarce resources are often power and physical space.
Once a B300 arrives at a data center, it still requires high-density power supply, liquid cooling, networking, and a sufficiently stable data center environment. Chips can be manufactured, but securing locations and grid connections often involve longer wait times. Therefore, whoever secures land and power capable of running next-gen GPUs first is closer to landing the next major contract.
This is where Axe Compute's recent collaboration with Duos Technologies comes into play.
On August 17, the two parties announced an additional allocation of up to 55MW of AI data center capacity, with projects spread across multiple locations in the U.S., expecting total payments to exceed $500 million. As mentioned earlier, the two had already been advancing a 10MW project in Columbus, Georgia, with Duos in the delivery phase. The preliminary project readiness target for the additional capacity is set to commence by the end of 2026, extending into early 2027.
Yet, the most critical aspect of this agreement lies not in the capacity but in a change of role.
Both parties have signed a term sheet, with Axe Compute set to make a minority equity investment in the relevant project entities, expecting to hold approximately 49% equity, participating through special purpose entities in the ownership and financing of certain facilities.
In other words, Axe Compute is no longer just renting someone else's data center, but rather holding ownership of the building and power infrastructure that supports its own clients. Ownership brings long-term control over capacity and costs, representing a tangible asset behind multi-year customer contracts, as well as the right to scale at its own pace without having to wait in line each time.
In the current environment where power resources have become the major bottleneck of the entire AI infrastructure, the value of this move is easy to understand. GPUs can be ordered, data centers can be leased, but whether a site can receive sufficient power and when it can be grid-connected is a scheduling timeline that money can't buy. By securing 55MW of capacity in advance, the deployment for clients is de-risked from site, power, or construction delays, turning it into a certainty ahead of time.
Initially, Axe Compute accessed online computing power through third-party networks; subsequently, it began building dedicated GPU clusters for clients; now, it is attempting to enter data center and power projects that host these clusters. It is moving from a role of "purchasing capacity to resell to clients" toward a position of "co-owning and operating infrastructure together."
This is also the difference between Axe Compute and CoreWeave.
CoreWeave is already a scaled AI cloud company, with strengths in large-scale GPU clusters, mature deployment capabilities, software systems, and a large number of operational customer contracts. Operating dozens of large AI data centers in North America and Europe, it builds single-room multi-thousand-GPU training clusters using InfiniBand high-speed interconnects and Kubernetes native orchestration, serving the massive-scale training needs of top AI labs.
Axe Compute does not yet have this scale; its starting point is more suitable for its own situation. It first captures immediate demand through Access, using the Aethir network's distributed resources for global coverage and rapid access, then embarks on Build projects for customized heavy asset delivery. By combining light asset access with heavy asset customization in parallel, it can tap into regional markets and more diversified customer needs that CoreWeave does not cover.
It must be objectively acknowledged that Axe Compute is still far from CoreWeave.
It does not have the scale that CoreWeave has already achieved; the $104 billion order backlog is over thirty times its $3.2 billion contract value. It also lacks the power assets accumulated over many years by traditional data center companies; the 55MW equity arrangement still needs to go through formal documentation, closing conditions, and approvals from both parties. The adjusted EBITDA for the second quarter was negative $4.9 million, indicating the company is still in the early investment stage; the management's provided gross margin for Build projects is 28% to 44%, EBITDA margin 62% to 76%, but these are only estimates once the projects are maturely operational.
Of course, this does not mean AGPU can be compared on the same scale as CoreWeave or Nebius. The market's undervaluation of AGPU is essentially pricing in the early delivery risk: contracts need to turn into prepayments, prepayments need to turn into live clusters, and only then will it show up as revenue in the financial reports. However, the "big orders, small company" mismatch is precisely where this company's future trajectory is most worth tracking.
If we translate orders into valuation language more familiar to the capital markets, Axe Compute's contrast becomes more intuitive. As of August 17th, AGPU's market capitalization is around $90 million; whereas the company's projection is that once all signed projects are deployed, the annualized revenue run rate is expected to exceed $696 million. Based on this forward-looking estimate, Axe Compute corresponds to a market-to-sales ratio of about 0.13.
By comparison, at the same time, market leaders like CoreWeave and Nebius, with market capitalizations in the hundreds of billions of dollars, have market-to-sales ratios of 5.5 and 24, respectively. The significant difference highlights Axe Compute's potential as a company with big orders but still a small market value.
After all, its strength lies precisely in its light start and quick pivot. The Aethir Network allows it to provide on-demand computing power even without its own data center, enabling it to secure its first revenue stream. Build contracts and the subsequent prepayments have provided it with a path to pivot to asset-heavy delivery without having to first burn through a round of equity to build the first data center.
CoreWeave has shown the capital markets one thing: AI computing power can be turned into a business intertwined with long-term contracts, equipment financing, and infrastructure operations. Axe Compute has not yet reached that stage.
However, from the initial $260 million contract in April, to the $317 million prepayment received in August, and now to the Columbus cluster that is about to go live, Axe Compute has at least found its own way onto the field.
In the coming months, key points to watch will be: whether the B300 cluster in Columbus can go live as scheduled; whether the received prepayments can continue to translate into equipment deployment and revenue; and whether the promised "additional $2 billion contract by the end of the year" mentioned by Miglino in the conference call can truly materialize.
A $32 billion contract value has already brought Axe Compute into the spotlight, and a market-to-sales ratio of 0.13 also makes us aware of its potential as a company with big orders but still a small market value.
Whether this misalignment can be corrected will also depend on whether the contract can successfully transition to the online cluster and realize the expected revenue. Therefore, in the coming months, we can continue to monitor: when will Axe Compute's cluster in Columbus go live and when will customers start being billed; Miglino's year-end target of $20 billion, how much will actually be achieved.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
