Nebius stock jumped on Thursday after the AI cloud provider revealed another round of price increases for access to its computing infrastructure, giving investors a surprisingly direct signal that demand for Nvidia-powered capacity remains stronger than available supply. Beginning October 1, Nebius plans to raise pay-as-you-go pricing on selected Nvidia GPU instances by roughly 17% to 21%, while some CPU pricing is also moving higher. It is the company’s second increase in only a few months, and that detail may matter more than the percentage itself. AI infrastructure companies normally face a long-term threat from falling hardware costs and increasingly abundant computing capacity, yet Nebius is currently doing the opposite of discounting: it is asking customers to pay considerably more.
The market immediately understood the implication. Nebius shares surged more than 9% in premarket trading before moderating after the open, extending a remarkable year in which the stock has already risen more than 150%. The bullish interpretation is straightforward: if Nebius can increase prices while simultaneously bringing huge amounts of new GPU capacity online, revenue per unit of infrastructure could rise at the same time as the company expands the number of units it can sell. That is a much stronger economic proposition than merely building increasingly expensive data centers and hoping utilization eventually follows. But the price increase also arrives as Nebius commits billions of dollars to infrastructure, signs enormous contracts with customers such as Meta and Microsoft, and raises substantial amounts of debt. The crucial question for Nebius stock is therefore not simply whether AI compute remains scarce today, but whether that scarcity lasts long enough for the company to earn attractive returns on the massive capacity it is building for tomorrow.
A 20% Price Increase Says More About AI Demand
The latest pricing changes are unusually revealing because they affect real compute already being offered to customers rather than capacity promised several years into the future. Reuters reported that Nebius will raise rates on selected Nvidia GPU instances by between 17% and 21% from October 1, marking the second time in three months that the company has increased AI cloud pricing. Investing.com reported that H100-based services are rising by roughly 17%, while pricing for Nvidia’s newer B300 GPU instances is increasing by around 21%. AMD EPYC Genoa CPU rates are reportedly rising about 25% as well.
That pricing behavior challenges one of the most persistent bearish arguments around companies such as Nebius and CoreWeave. Critics have warned that AI cloud providers are spending enormous amounts on GPUs that depreciate quickly, while growing industry capacity should eventually push rental prices lower. Nvidia launches new architectures at a rapid pace, so older hardware such as the H100 theoretically should become less valuable when customers can rent newer chips with better performance. Yet Nebius is raising prices even on H100 instances, despite the chip having launched years ago. Barron’s noted that the move complicates the argument that older GPUs must rapidly lose their economic value as each new generation arrives.
The reason appears to be simple: customers still want more compute than the market can comfortably supply. Nebius is not alone in seeing that dynamic. Reuters reported Thursday that competitor CoreWeave has also been signing contracts at higher prices, reinforcing the idea that the pricing environment is being driven by broader AI infrastructure scarcity rather than something unique to Nebius. For investors, that makes Thursday’s price increase potentially more important than another announcement about planned megawatts or future data-center campuses. Capacity tells shareholders what Nebius hopes to sell. Pricing tells them what customers are willing to pay right now.
Related articles
Nebius Already Proved Customers Will Pay More
The latest price hike becomes much more interesting when placed alongside Nebius’s second-quarter results. Group revenue reached $582.3 million during Q2, up 454% from $105.1 million a year earlier, while revenue from the core Nebius AI cloud business surged 514% to $574.9 million. In its regulatory filing, the company explicitly attributed that growth to capacity expansion supported by strong pricing and utilization. That is exactly the combination shareholders want to see from such a capital-intensive business: more infrastructure coming online, customers using it heavily, and pricing remaining firm rather than collapsing as capacity expands.
The profitability progression has also been dramatic. Nebius reported $236.2 million of adjusted EBITDA at the group level in Q2 compared with a $21 million adjusted EBITDA loss in the year-earlier period, while the AI cloud operation reached an adjusted EBITDA margin of about 50%. Management’s shareholder materials also indicated that annual contract value per megawatt has been stepping higher as new deals are signed, suggesting that every new block of infrastructure may be generating more contracted revenue than earlier deployments.
This is the financial logic behind Thursday’s stock reaction. A 20% price hike does not automatically increase total Nebius revenue by 20%, because large customers often sign long-term contracts with predetermined economics and may not be affected by pay-as-you-go changes. Barron’s specifically cautioned that investors should not expect an immediate one-for-one revenue boost because much of Nebius’s business is contracted. But the pricing decision still reveals what incremental capacity is worth in the current market, and that information becomes increasingly important as Nebius builds more of it.
If future contracts are being signed at higher prices while utilization remains strong, the economics of the entire expansion plan can improve. That is a much more important development than the near-term revenue contribution from customers whose October bills simply become more expensive.
Microsoft and Meta Have Already Given Nebius Something
Nebius is not building its infrastructure entirely on speculation. The company has secured several enormous contracts that give it considerably more visibility than a typical early-stage cloud provider. In September 2025, Nebius announced a five-year AI infrastructure agreement with Microsoft worth $17.4 billion, with the potential to reach approximately $19.4 billion if Microsoft exercises additional capacity options. Nebius shares jumped nearly 44% when the agreement was disclosed because the contract validated the company’s strategy of building specialized GPU infrastructure outside the largest traditional hyperscalers.
Meta has since become another major customer. In March 2026, Nebius announced a five-year agreement under which it will provide Meta with $12 billion of dedicated infrastructure capacity across several locations using Nvidia’s Vera Rubin platform beginning in early 2027. Meta also committed to purchase additional available compute capacity across specified Nebius clusters, potentially bringing the total commitment to as much as $15 billion over five years. Nebius intends to sell that additional capacity to third parties whenever possible, with Meta effectively acting as a buyer for remaining availability.
That arrangement is strategically powerful because it reduces one of the biggest risks associated with aggressive data-center construction: building expensive infrastructure before knowing whether anyone will use it. Nebius still carries execution and financing risk, but large multiyear commitments from Microsoft and Meta provide a baseline of demand against which new capacity can be planned. The latest pricing increases add another layer to that thesis. If contracted hyperscaler demand provides utilization protection while spot and shorter-term customers pay increasingly high rates for whatever capacity remains, Nebius could have a particularly attractive mix of visibility and pricing power.
The challenge is that delivering those contracts requires enormous amounts of money first.
The Price Hike Matters Because Nebius Is Spending Billions Before It Gets Paid
The AI cloud business is fundamentally different from conventional software. Software companies can often add customers at relatively low incremental cost because distributing another copy of an application requires little physical investment. Nebius needs data centers, power connections, networking equipment and thousands of extremely expensive Nvidia GPUs before revenue can be recognized. That makes the company’s growth spectacular but intensely capital hungry.
Nebius ended Q2 with around $8 billion of cash and said it had secured approximately $775 million of asset-backed financing in July. Its shareholder presentation showed contracted power capacity exceeding 4 gigawatts, with additional infrastructure scheduled to come online through 2027. Management also said four major deals signed in Q2 averaged more than $1 billion each, while customer prepayments and related structures can finance 50% to 60% of capital expenditures on certain projects.
The company subsequently went back to capital markets. In August, Nebius completed approximately $5.75 billion of convertible senior note financing, adding substantial resources for its global buildout but also reminding investors that rapid expansion has a financing cost.
This is precisely why pricing power is so important. Building a data center with expensive borrowed capital is far easier to justify when customers are accepting rising rental rates. If compute pricing eventually falls while interest expense, depreciation and power costs remain high, returns could deteriorate quickly. Thursday’s announcement does not eliminate that danger, but it shifts the near-term evidence in Nebius’s favor: capacity remains scarce enough that the company believes it can charge materially more even after expanding aggressively.
The H100 Price Increase May Be the Most Important Number in the Entire Story
The roughly 17% increase for Nvidia H100-based computing deserves special attention because the H100 is no longer Nvidia’s newest AI accelerator. Nvidia has subsequently introduced Blackwell systems and is moving toward its next-generation Vera Rubin architecture. In a conventional technology cycle, investors might expect the rental value of older hardware to decline sharply as newer processors become available.
Instead, H100 pricing at Nebius is rising.
That does not prove H100 economics will remain strong forever, nor does it guarantee that every older GPU will maintain pricing power indefinitely. But it suggests AI compute demand is broad enough that older but still highly capable chips remain economically useful. Customers do not always need the newest accelerator available; price, availability, software compatibility and workload type all matter. Inference, fine-tuning and less computationally intensive training workloads can continue using previous-generation equipment even as frontier model developers migrate toward newer Nvidia systems.
That potentially extends the economic life of Nebius’s installed hardware and could improve returns on capital. Depreciation assumptions have become one of the most controversial issues surrounding the AI infrastructure boom because billions of dollars of profitability depend on how long expensive GPUs remain commercially valuable. Rising H100 rental rates do not settle that accounting debate, but they provide real-world evidence that previous-generation hardware can remain in high demand even while superior chips enter the market.
Nebius Stock Still Carries the Same Risk That Destroyed Earlier Infrastructure Booms
The bullish evidence is powerful, but investors should not mistake today’s scarcity for a permanent law of economics. Reuters Breakingviews recently compared the extraordinary expansion of AI “neoclouds” with the alternative-network infrastructure boom of the late 1990s, when companies borrowed aggressively to build fiber networks in anticipation of exploding internet traffic. Demand ultimately grew enormously, but capacity expanded even faster for a period, destroying pricing and bankrupting many infrastructure providers before the internet itself fulfilled the long-term growth thesis.
The analogy is imperfect, but the warning is relevant. Nebius, CoreWeave, hyperscalers and a growing group of specialized infrastructure companies are simultaneously adding huge amounts of AI capacity. Nvidia is shipping increasingly powerful GPUs, data-center construction is accelerating across multiple continents, and customers such as Microsoft, Meta, Amazon and Google are investing directly in their own infrastructure as well. Today’s 20% pricing increase tells investors that demand currently exceeds supply. It does not tell them what that relationship looks like in 2028.
That uncertainty explains why Nebius stock can rally sharply on evidence of pricing power while still remaining highly volatile. Shares have risen more than 150% in 2026, yet Barron’s noted that the stock remained roughly 27% below its recent three-month peak before Thursday’s move. Investors are simultaneously trying to price extraordinary revenue growth and the possibility that an infrastructure glut eventually compresses returns.
Nebius Stock’s Next Test Is Whether Higher Prices Survive the Capacity Boom
Thursday’s announcement gives the Nebius bull case something especially valuable: evidence that the company is not merely growing because it keeps buying more GPUs. It appears to be growing while gaining pricing power at the same time. Q2 revenue rose 454%, the AI cloud business generated a roughly 50% adjusted EBITDA margin, utilization remained strong, and Nebius is now raising pay-as-you-go pricing for a second time in three months. Those pieces fit together into a much stronger story than raw megawatt expansion alone.
But the next phase will be harder. Microsoft and Meta contracts require Nebius to bring enormous amounts of infrastructure online through 2027, and the company has raised billions of dollars to fund that expansion. Investors should therefore watch whether annual contract value per megawatt continues increasing, whether utilization remains high as new capacity arrives, and whether older Nvidia hardware such as the H100 retains attractive rental economics after Vera Rubin and other next-generation systems become widely available.
For now, the latest evidence points to a compute market that remains remarkably tight. Customers are competing for capacity, rivals are also signing contracts at higher prices, and Nebius feels confident enough to impose another double-digit increase only months after its previous one.
That is why Nebius stock reacted so strongly.
The company has spent much of the AI boom proving it can build data centers and secure giant customers. Thursday delivered evidence of something potentially even more valuable: Nebius may have enough demand to dictate the price.
If that pricing power survives the massive wave of GPU capacity arriving through 2027, the economics of its infrastructure buildout could look dramatically stronger than investors expected. If it disappears once supply catches up, today’s rally will look more like a snapshot of peak scarcity.
For Nebius stock, that may now be the single most important question.
Disclaimer
This article is for informational purposes only and does not constitute financial or investment advice. Readers should conduct their own research and, where appropriate, consult a qualified financial advisor before making investment decisions. This article was researched and drafted with the support of AI, then reviewed, fact-checked and edited by the editorial team before publication.
