On August 12, Nebius, a leading company on the Neo Cloud赛道, rose 28% in a single day.
Interestingly, just a few days ago, Michael Burry, who gained fame for shorting subprime mortgages, disclosed a short position in this company, with nearly 30% of the float being shorted. After the earnings report, short sellers were forced to cover their positions, marking Nebius's largest single-day gain since September 2025.
The day before, CoreWeave, another leader in the Neo Cloud sector alongside Nebius, reported its second-quarter earnings: revenue of $2.8 billion, more than doubling year-over-year; unfulfilled order backlog surged to $104 billion, over three times the level from a year ago; full-year guidance was raised accordingly, and the stock rose 19% the following day. IREN, Applied Digital, and Core Scientific in the same sector also rallied. That same week, Roundhill launched an ETF focused on the Neo Cloud theme.
In this August tech stock rebound, Neo Cloud is one of the most significant and strongest-performing sectors among this year's tech stocks.
Over the past year, the market has been debating who will build the next large model; now, an increasing amount of capital is asking a more practical question: if models continue to grow larger, who will run them?
The answer isn't just NVIDIA. For a GPU to truly become sellable computing power, it must be connected to a data center, power supply, cooling, network infrastructure, operations and maintenance, and a customer order with upfront payment. The GPUs are still on the way, but transformers may already have a six-month waitlist; model companies are rushing to scale up, yet available space in data centers is growing increasingly scarce.
As a result, a new wave of companies specializing in building AI infrastructure for clients is gaining increasing attention in the capital markets. They don’t just sell a standard cloud account—they provide an entire environment capable of running large models: clients don’t need to buy their own hardware, secure power supply, or build a data center team from scratch. These companies are now being referred to as Neo Cloud.
In addition to well-known names like CoreWeave and Nebius, Nasdaq is home to another company that has recently rebranded and is emerging from the competition.
It's called Axe Compute, stock symbol AGPU.
On August 14, Axe Compute released its second-quarter financial report. The report disclosed that the cumulative contracted value as of 2026 has surpassed $3.2 billion; if all these projects are fully deployed, the annualized revenue run rate is expected to exceed $696 million.
It is clearly not a company on the same scale as CoreWeave. But for a company that only changed its name at the end of last year and only began selling computing power this year, these disclosed figures are worth examining.
What kind of new company is Axe Compute?
Axe Compute is the new brand and strategy following the rebranding of Predictive Oncology.
The company previously focused on oncology drug discovery, shifted to AI computing infrastructure at the end of 2025, officially changed its name on December 11, 2025, and began trading on Nasdaq as AGPU the following day. Starting in February 2026, the management team was fully replaced, with Christopher Miglino appointed as CEO, Kyle Okamoto joining as President in April, and Jeremy Yaukey-Witter promoted to CFO in May.
Axe Compute management team members
In terms of business lines, Axe Compute currently has two.
The first business line is Axe Compute Access, which orchestrates already-online third-party GPU capacity to provide customers with immediately available computing power. This line primarily connects to global third-party data center resources through the Aethir distributed compute network, which spans 93 countries and over 200 nodes, offering access to more than 435,000 GPUs.
The second business line is Axe Compute Build, which designs, deploys, holds, and operates dedicated GPU clusters for enterprise clients. Clients do not need to purchase hardware, secure facilities, or assume ownership and operational responsibilities for the equipment. The ownership of the GPU infrastructure remains with Axe Compute, and after the contract ends, these GPUs can theoretically continue serving the next client.
Clearly, the Build business better reflects the company's ambition.
In April this year, Axe Compute signed its first landmark Build contract: a 36-month agreement worth approximately $260 million, planning to deploy 2,304 NVIDIA Blackwell B300 GPUs and accompanying high-speed storage in a Tier 3 data center in the United States, with a target launch in Q3. If successfully operationalized, this contract could generate approximately $20 to $21 million in quarterly revenue.
Axe Compute built its first data center outside Duos Corporation
On the other end of the contract, the facility is located in Columbus, a city in the southeastern United States. Originally an unnamed spot on the U.S. data center map, it has recently seen a surge in AI data center deployments due to saturated power capacity and long grid connection wait times at primary hubs.
The first B300 cluster from Axe Compute is housed in this Tier 3 data center campus in Columbus, operated by Duos Edge AI. With 288 nodes, each equipped with 8 B300 GPUs and 288GB of HBM3e memory, the total adds up precisely to the 2,304 cards specified in the contract. Nodes are interconnected via InfiniBand Quantum-X800, delivering 6.4 Tb/s bandwidth per node, alongside over 20PB of high-speed storage provided by WEKA.
Inside the Duos Edge data center
Interestingly, on the Q2 earnings call on August 17, the company streamed directly from this data center. The first words spoken were: "Today we're broadcasting live from Columbus, Georgia, the location of the B300 cluster announced in April."
Axe Compute CEO Christopher Miglino and Duos CEO Doug Recker
Duos’s CEO, Doug Recker, was seated in the audience himself, describing the construction progress this way: “What has been accomplished in these 45 days is incredible—if you were to pan a camera around the site, you’d see just how astonishingly fast the progress has been.” He added, “We won’t tell you we can build 100 megawatts in three weeks, but everything you’ve seen here was completed within 45 days.”
The cluster is expected to go live within weeks, with plans to triple its scale over the coming months. Meanwhile, another cluster is under construction in Boden, Sweden, featuring 256 B300 nodes with spare parts, RoCEv2 networking, and WEKA storage, scheduled to launch by the end of 2026.
After this, contract amounts began to flood in starting in July.
On July 22, the company announced over $1.3 billion in new contracts in the United States and Europe, surpassing its $1 billion annual contract target set in May; five days later, on July 27, it unveiled another five-year contract worth over $1.5 billion to deploy a dedicated cluster with more than 9,200 B300 GPUs in the United States. Both contracts include extension options. By mid-August, the total contracted value for 2026 had exceeded $3.2 billion, with over $2.8 billion coming from the three Build contracts signed in July.
In three months, we tripled our annual goal.
Completed $3.2 billion in orders in six months
From Axe Compute's Q2 financial report, the most obvious change is in revenue.
The company's revenue for the quarter was $3.215 million, compared to approximately $35,000 in the previous quarter, representing nearly a 90-fold sequential increase; for comparison, the same period last year was $3,000. Operating cash flow for the first half of the year was positive at $17.4 million, with cash on hand at quarter-end totaling $21.9 million, up from $6.9 million at the end of the first quarter.
The cost of revenue for the second quarter was $3.013 million, resulting in a gross profit of approximately $200,000 and a gross margin of 6.3%.
Is 6.3% high or low? It’s impossible to tell just by looking at Axe Compute—we need to compare it with the leaders in this space.
In the same quarter, CoreWeave reported revenue of $2.58 billion and adjusted EBITDA of $1.51 billion, with an EBITDA margin of 59%. Although the two companies' revenue scales differ by a factor of 800, making direct comparison irrelevant, the disparity in gross profit structures still reveals significant insights.
The difference in gross margin structure is not caused by scale, but by the nature of the business.
Axe Compute's second-quarter revenue primarily came from Access, which is essentially a resale business. Axe Compute purchases already-online capacity from third-party networks and resells it to customers at a markup, earning the spread between purchase and sale prices. Since it doesn’t own the hardware or the data centers, its pricing power is limited. The gross margin ceiling for this type of business is inherently low. In contrast, CoreWeave and Nebius sell GPUs that they purchase, install, and operate themselves; customers sign multi-year, take-or-pay contracts, and these companies earn returns on their assets rather than commissions from matchmaking.
This is also why Axe Compute began shifting its core focus to Build starting in the second quarter.
Perhaps Axe Compute realized that the Access model had reached its limits. While these figures might be respectable for a company just starting to sell computing power, they fall short of Axe Compute’s true ambitions. Axe Compute needed its own GPU infrastructure—and so far, it has secured over $3.2 billion in contract orders, with management projecting gross margins for the Build project between 28% and 44%.
We all know that the second quarter ended on June 30. Therefore, the three Build contracts signed in July, totaling over $2.8 billion, are explicitly marked as subsequent events in the company’s financial report. This means that the Q2 financial report currently seen by the market is effectively the last one before Axe Compute’s transformation.
In the July 27 announcement of the $1.5 billion contract, the company stated that it expected to receive cumulative advance payments exceeding $534 million within 30 days. On the August 17 conference call, Miglino announced the first payment: over $317 million in advance payments has already been received for the expanding cluster.
In other words, the $3.2 billion contract and the $300 million down payment already received can be considered more significant than revenue in the Q2 financial report.
One of the most counterintuitive aspects of the Neo Cloud space is that long-term, non-cancelable contracts locking in future cash flows can be used as collateral. CoreWeave’s $104 billion backlog of orders essentially functions as a risk-mitigation mechanism, securing future revenue before asset depreciation occurs, enabling the company to borrow at lower costs. In March 2026, CoreWeave completed an approximately $8.5 billion debt financing backed by these contracts, becoming the first company in the industry to secure an investment-grade-rated GPU-backed loan. Nebius is following the same path: over 70% of its new contracts in Q2 included upfront payments, with over $9 billion expected to be received this year.
Large companies are willing to act as "prepaid clients" because purchasing compute power from Neo Cloud allows them to expense the cost over the contract term, rather than capitalizing it upfront as they would with building their own data centers—which would strain their already pressured free cash flow. Thus, the order has been completely reversed. In the past, suppliers invested heavily to build infrastructure first and then waited for clients to migrate gradually; now, clients lock in capacity and pay upfront, enabling suppliers to use the contracts and prepayments to procure GPUs, reserve data center space, and secure financing.
$317 million in cash inflow directly reduces the external financing required for this deployment, allowing the company to immediately initiate equipment procurement and secure data center capacity without first negotiating a round of equity or debt. The fact that customers are willing to transfer funds before the cluster is even powered on serves as the most direct validation of the order’s authenticity and commitment; for downstream project financiers, this also provides a verifiable cash receipt for due diligence.
Axe Compute has also outlined its revenue realization timeline for the coming period: an annualized revenue run rate of $37 million by the end of Q2, increasing to approximately $139 million after the Columbus cluster goes live, and exceeding $696 million upon full deployment of all signed contracts—nearly doubling its previous guidance of $385 million, with the window targeting Q4 2026 to Q1 2027.
Looking ahead, the company disclosed a qualified sales pipeline of $5.9 billion across 98 opportunities, nearly double the value of current signed contracts. Miglino said on the August 17 earnings call: “We remain confident in signing an additional $2 billion in contracts before the end of this year.”
From accessing computing power to controlling electricity, Axe Compute aims to go deeper.
The most prominent term in AI infrastructure is GPU, but what's truly hard to acquire on short notice is often electricity and space.
After a B300 arrives at a data center, it still requires high-density power supply, liquid cooling, network infrastructure, and a sufficiently stable facility environment. While chips can be manufactured on schedule, securing the physical space and grid connection often takes much longer. Therefore, whoever secures land and power capable of supporting the next-generation GPU gets closer to landing the next major contract.
This is also the significance of Axe Compute's recent collaboration with Duos Technologies.
On August 17, the two parties announced an additional AI data center capacity arrangement of up to 55 MW, with projects distributed across multiple locations in the United States, with expected total payments exceeding $500 million. As previously mentioned, prior to this, the two parties had already advanced a 10 MW project in Columbus, Georgia, which Duos is currently delivering. The initial project readiness targets for the additional capacity are set to begin by the end of 2026 and extend into early 2027.
But the more critical provision in this agreement is not about capacity, but about the change in identity.
Both parties have signed the term sheet; Axe Compute will make a minority equity investment in the relevant project entity, with an expected ownership stake of approximately 49%, participating in the ownership and financing of certain facilities through a special-purpose entity.
In other words, Axe Compute is no longer just renting space in other data centers—it now holds equity in the building and power infrastructure that hosts its own customers. Ownership provides long-term control over capacity and costs, represents a tangible asset behind multi-year customer contracts, and grants the right to scale at its own pace, without having to wait in line for others’ resources.
At a time when power resources have become the most critical bottleneck in AI infrastructure, the value of this is easy to understand. GPUs can be ordered, data centers can be rented, but whether a site can secure sufficient power and when it can be connected to the grid are schedules that money cannot buy. Locking in 55MW of capacity in advance turns what would otherwise be uncertainty around deployment due to location, power, or construction timelines into a guaranteed certainty.
Initially, Axe Compute accessed online computing power through third-party networks; later, it began building dedicated GPU clusters for its clients; now, it is attempting to enter the data center and power projects that host these clusters. It aims to move from a role of “purchasing capacity and reselling to customers” toward one of “jointly owning and operating infrastructure.”
This is also the difference between Axe Compute and CoreWeave.
CoreWeave is already a scaled AI cloud company, with advantages in large-scale GPU clusters, mature deployment capabilities, software systems, and a substantial number of operational customer contracts. Operating dozens of large AI data centers across North America and Europe, it builds single-facility, ten-thousand-GPU training clusters using InfiniBand high-speed interconnects and native Kubernetes orchestration to serve the ultra-large-scale training needs of leading AI labs.
Axe Compute does not yet have such scale; its approach is better suited to its current stage—starting with Access to meet immediate demand, leveraging Aethir’s distributed network for global coverage and rapid reach, then progressing to Build projects for customized, asset-intensive deliveries. By combining lightweight outreach with heavyweight customization in parallel, it can enter market segments and serve more diverse, underserved customer needs that CoreWeave cannot reach.
It must be objectively acknowledged that Axe Compute is still far behind CoreWeave.
It does not have the scale already achieved by CoreWeave, where a $10.4 billion order backlog is more than thirty times its $320 million contracted value; it also lacks the years of accumulated power assets typical of traditional data center companies, and the 55 MW equity arrangement still requires completion of formal documentation, closing conditions, and mutual approvals. The adjusted EBITDA for the second quarter was negative $4.9 million, indicating the company remains in its early investment phase; management has projected Build project gross margins of 28% to 44% and EBITDA margins of 62% to 76%, but these are only estimates based on mature project operations.
This certainly doesn't mean AGPU can yet be measured on the same scale as CoreWeave or Nebius. The market's low valuation essentially prices in the early delivery risk: contracts must first become advance payments, advance payments must become operational clusters, and only then do they turn into revenue on financial statements. But the mismatch between "large orders and a small company" is precisely what makes this firm most worth tracking going forward.
Price-to-Sales Ratio Comparison Chart
To make the contrast more intuitive using valuation language more familiar in capital markets, as of August 17, the market capitalization of AGPU was approximately $90 million; however, according to the company’s guidance, once all signed projects are fully deployed, the annualized revenue run rate could exceed $696 million. Based on this forward projection, Axe Compute trades at an implied price-to-sales ratio of approximately 0.13x.
Compared to industry leaders like CoreWeave and Nebius, which have market capitalizations in the tens of billions of dollars, their price-to-sales ratios are 5.5x and 24x respectively. This stark contrast highlights Axe Compute’s potential—a company with substantial order volume but a relatively small market capitalization.
After all, its advantage lies precisely in its lightweight start and rapid agility. The Aethir network enables it to provide callable computing power without owning its own data center, generating its first revenue right away; the Build contracts and the accompanying upfront payments then provide an entry point toward heavy-asset delivery, without requiring it to burn through a round of equity to build its first data center.
CoreWeave has shown capital markets that AI computing power can be turned into a business intertwined with long-term contracts, equipment financing, and infrastructure operations. Axe Compute is still far from reaching that point.
From the first $260 million contract in April, to the $317 million advance payment received in August, to the upcoming cluster in Columbus set to go live, Axe Compute has at least carved out its own entry path.
Over the coming months, key items to watch include: whether Columbus’s B300 cluster will come online as scheduled; whether the prepaid amounts already received can continue to be converted into equipment deployment and revenue; and whether Miglino’s statement on the earnings call about “securing an additional $2 billion in contracts by year-end” will actually materialize.
A $3.2 billion contract value has brought Axe Compute into the spotlight, and a price-to-sales ratio of 0.13 highlights its potential—a company with a large order book and a small market capitalization.
Whether this misalignment can be resolved going forward will also depend on whether the contracts can be successfully deployed as operational clusters and recognized as confirmed revenue. Therefore, over the coming months, we can continue to monitor: when Axe Compute’s Columbus cluster will go live and when customers will begin to be billed; and how much of Miglino’s $2 billion year-end target will ultimately be achieved.
Click to learn about the open positions at BlockBeats
Welcome to the official BlockBeats community:
Telegram subscription group: https://t.me/theblockbeats
Telegram group: https://t.me/BlockBeats_App
Official Twitter account: https://twitter.com/BlockBeatsAsia