Nebius Q2 revenue surges 454% as AI compute prices climb to $50 million per megawatt

Organized & Compiled by Deep潮 TechFlow

Speaker: Shay Boloor, Chief Market Strategist at Futurum Equities

Podcast source: FE Earnings Edge (Futurum Equities)

The Nebius (NBIS) Earnings Rally, Explained

Broadcast date: August 12, 2026

Disclosure: The speaker explicitly stated that Nebius is among the top ten holdings in their family portfolio and significantly increased their position during a market correction several months ago; the portfolio also holds Bloom Energy, mentioned in the video. This article presents a shareholder’s interpretation of the earnings report, not an independent analysis, and does not constitute investment advice.

Nebius (NBIS) released its second-quarter earnings pre-market on August 12: revenue reached $582 million, a 454% year-over-year increase; AI revenue surged over 500% to approximately $575 million; adjusted EBITDA margin rose significantly to 41% from 32% in the prior quarter, with the core AI cloud business achieving a profit margin of up to 50%. On the earnings day, the stock price climbed from the August 11 closing price of $193.23 to an intraday high of $243.89, a gain of about 26%, forcing short sellers—whose short interest was nearly 25%—to cover their positions, further amplifying the rally.

Speaker Shay believes that a more significant signal than revenue lies in profit margins: the shortage of computing power is simultaneously driving higher prices, faster payback periods, and improved financing terms. Emergency computing power (with delivery in 3 to 6 months) is selling for $40 million to $50 million per megawatt—twice the price of standard contracts (1- to 3-year terms, at $20 million to $25 million per megawatt); the average payback period for contracts signed in Q2 is approximately 1 year and 10 months; 70% of contracts include upfront payments, with over $9 billion in customer prepayments expected to be received by 2026, covering 50% to 60% of related capital expenditures. A week before the earnings report, Michael Burry, the real-life inspiration for "The Big Short," disclosed a short position in Nebius; the stock’s surge on earnings day made this long-short battle especially prominent.

Regarding this earnings report

• This earnings report proves one thing: AI computing power shortages are translating into better pricing, faster payback periods, higher gross margins, and improved financing terms. (Shay)
• For shareholders, it’s crucial that a company spending tens of billions of dollars on infrastructure can demonstrate that these assets can generate such profit margins. (Shay)

Regarding pricing power

• "If you're an AI company about to train your next-generation large model, getting GPUs six months from now is essentially useless, because AI evolves exponentially. You need them today—and that means paying a premium." (Shay)
• The winning bid for the first hashpower auction was 15% higher than Nebius’s previous all-time high—this is true price discovery. (Shay)

Regarding the financing structure

• The customer funds half of the data center, and the bank then provides a loan based on contracted revenues, leaving Nebius’s stake as the last dollar, not the first. (Shay)
• If the lightweight asset model works, Nebius can generate profits from infrastructure it doesn’t need to fully fund, significantly improving capital efficiency. (Shay)

Regarding the next steps

• Management says that, based on the current contract economics, they could sell out all of their 2027 capacity today—but they’re choosing not to. (Shay)
• Demand is no longer the issue; the next challenge is execution: turning the 5GW pipeline into grid-connected, revenue-generating computing power. (Shay)

Key insight: The most important takeaway from this earnings report is not that both metrics exceeded expectations, but that compute shortages are becoming economic.

Shay immediately set the tone: Nebius’s earnings report beat market expectations on revenue and EPS per share, but what truly matters is that the company provided concrete evidence that AI compute shortages are simultaneously driving higher prices, faster payback periods, improved gross margins, and better financing terms—all while it is still aggressively expanding capacity for 2027.

On a numerical level: Revenue reached $582 million, a 454% year-over-year increase (approximately 5.5 times), slightly above the market expectation of around $570 million; AI revenue grew by over 500%, reaching approximately $575 million. However, what Shay is truly focused on is profitability: the group’s adjusted EBITDA margin rose to 41%, up significantly from 32% in the previous quarter; the adjusted EBITDA margin for the core AI cloud business reached 50%.

“A company spending tens of billions of dollars to build AI infrastructure must demonstrate that these assets can generate such profit margins—that’s decisive for shareholders,” he said, explaining why the stock surged more than 20% after earnings: “The improvement in fundamentals far exceeds the actual revenue or profit numbers.”

Core insight: Hash rate is splitting into two price tiers—one for "need it now" and one for "can wait."

The financial report for the first time disclosed the actual selling prices of capacity. Standard contracts (1 to 3 years) generate $20 million to $25 million per megawatt annually; however, if customers require delivery within 3 to 6 months, the price jumps directly to $40 million to $50 million per megawatt, and sometimes higher.

Why would someone pay double? Shay’s explanation is straightforward: If you’re an AI company preparing to train the next generation of large models, receiving GPUs six months from now is essentially meaningless, because AI evolves at an exponential pace. What you’re buying is a scarce resource—compute power that exists precisely when you need it. And the current reality is that this supply is extremely scarce.

This quarter, Nebius also conducted its first-ever compute auction. Shay believes this is the best indicator of supply-demand imbalance: the highlight wasn't management’s statements, but customers actively bidding—final prices exceeded Nebius’s previous all-time high by 15%. "This gave me real price discovery, showing that the market value of immediate compute is still rising."

Core insight: The biggest weakness of AI cloud companies is fundraising—Nebius is turning this into a moat.

The point where Nebius was most heavily attacked by short sellers was that "NeoCloud must continuously raise funds, leading to equity dilution and costly growth for shareholders." Shay acknowledged this as the Achilles' heel of the entire sector, emphasizing that choosing the right horse matters. However, this earnings report presents a different capital structure:

Of the contracts signed in the second quarter, approximately 70% include advance payments. The company expects to receive over $9 billion in customer advance payments for the full year of 2026, covering 50% to 60% of the related capital expenditures. Building on this, the company has secured $775 million in asset-backed financing, using its contracted infrastructure and customer cash flows as collateral to borrow at relatively favorable interest rates.

“The customer pays for half the construction, and the bank disburses loans according to the contract—Nebius’s stake becomes the last dollar, not the first,” said Shay. This shift in priority means the capital structure is beginning to revert to the norm of the financial world before the AI era. Furthermore, if the “light-asset model” succeeds, other investors could fund and own the data centers while Nebius contributes only its cloud platform, software, customer relationships, and demand—allowing it to profit from infrastructure it doesn’t fully finance.

Key insight: Management signals on the supply side are more aggressive than financial report figures.

The management has raised the contracted power target for the end of 2026 to 5 GW, marking the ninth consecutive quarter with an increase of approximately 1 GW each quarter. Shay said that power is the lifeline of this arms race, and the ability to continuously raise targets speaks volumes.

Even more aggressively, another statement says: At current contract economics, Nebius could sell out its entire planned capacity for 2027 today—but it deliberately chooses not to. Management believes that reserving some capacity for "spot demand" will allow it to command higher prices and generate greater returns.

However, Shay also cautioned that only about 800 MW to 1 GW will be truly grid-connected by the end of 2026, and revenue generation can only begin after installing GPUs, networking, cooling, and software. "This earnings report is strong, but my core argument for the next phase is turning this massive pipeline into operational infrastructure quickly enough. Concerns about demand can be put to rest—now it’s execution risk that matters."

Core judgment: In the long term, Nebius cannot merely be a landlord of scarce GPUs.

Shay repeatedly emphasized that NeoCloud investors must focus on one thing: whether these companies can evolve from merely renting scarce GPUs to selling platform value, since GPU scarcity won’t last forever. Nebius’s answer lies in the software layer—Token Factory and the broader software ecosystem—that retains customers through the platform, software, and services, not just through GPUs that can’t be found elsewhere.

In his view, a 50% AI cloud EBITDA margin may be an early sign of operating leverage: Nebius isn’t simply buying expensive hardware and renting it out at thin margins—it’s already achieving economies of scale. Additionally, the New Jersey project switched to Bloom Energy’s fuel cells because they provide reliable on-site power without delaying the schedule. Such details exemplify his focus on execution. “Permits, power, and construction delays push revenue to the right—no matter how strong demand is.”

Core judgment: All current evidence points to demand growing faster than supply, but this investment hinges on one assumption.

At the close, Shay made it clear: he is a holder who significantly increased his position during the market correction months ago and is now very pleased. The scarcity of computing power not only creates demand but also generates pricing power, financing rights, and the "right to choose when to monetize capacity." Clients are willing to pay $40 million to $50 million per megawatt for immediate computing power, achieving payback in less than two years, with customers prepaying half of the capital expenditures. The first auction saw a 15% premium, and management says all 2027 capacity could be sold out today.

Music is playing. Of course, risks remain: Nebius is deliberately betting that scarcity will persist—and I believe it will last many years, far longer than the market fears. All current evidence points to demand growing faster than supply, with customers paying higher prices for immediate capacity, while Nebius secures cheaper financing for each megawatt. This truly has all the ingredients of the next major new AI economy.