GPU cloud services trigger price hikes: Nebius raises prices by another 20%, leading to a reversal of bargaining power for computing power supply.

Nebius announced that it will raise the prices of its GPU cloud services across the board starting October 1, with an average increase of about 20%. This, coupled with Nvidia's strong quarterly earnings report, drove the stock up nearly 7% in after-hours trading, further confirming the market's assessment that the supply and demand of AI computing power will remain tight.

This round of price increases covers multiple chip models, including H100, H200, B200, and B300, marking Nebius' second price hike since May of this year. The market interprets this as a strong signal of continued overheating demand for AI computing power and persistently tight supply. Boosted by this, Coreweave rose nearly 4% in the afternoon session, and the optical communication sector generally strengthened, with AAOI and Credo Technology rising nearly 3%, Maiwei Technology and Astera Labs rising over 2%, and Coherent rising nearly 2%.

The price increase news, highly consistent with Nvidia's strong earnings guidance, has further strengthened market confidence in the continued construction cycle of AI data centers, driving a comprehensive recovery in sentiment across the entire cloud computing and computing infrastructure sector. At the same time, data center operators' bargaining power is increasing, the balance of contract terms is beginning to tilt towards the supply side, and the industry landscape is quietly changing.

Price Increase Details: Prices of multiple GPU models have been significantly raised.

According to pricing screenshots circulating among X platform users, this round of price adjustments is significant. The H100's price per GPU hour has increased from $3.85 to $4.50, an increase of approximately 16.9%; the H200 from $4.50 to $5.40, an increase of 20%; the B200 from $7.15 to $8.50, an increase of approximately 18.9%; and the B300 from $7.85 to $9.50, an increase of approximately 21%.

This marks Nebius' second price increase in several months. In May, the company announced an average price hike of 29% for on-demand capacity and 51% for preemptible capacity. Taking the B300 as an example, its price has increased by approximately 56% since May, from around $6.10 per hour.

In an email to customers in May, Nebius stated that the price adjustment "reflects the continued strong demand for high-end GPU computing power, and even after the pricing update, Nebius still offers one of the most competitive GPU infrastructure prices on the market." According to Stocktwits, as of press time, Nebius has not publicly responded to the latest rumors of price increases.

Supply and demand logic: Demand visibility exceeds 24 months, capacity remains critically low.

Market analysts point out that this round of price increases reflects a deeper supply and demand logic. According to X platform user @MelvinInvests, Nebius had previously attempted to manually adjust prices, but management stated that strong demand meant price adjustments were insufficient to balance supply and demand. Subsequently, the company conducted auction tests on scarce Blackwell computing power, with customers actually paying 15% to 20% higher than the previous maximum price to secure access.

The analyst also pointed out that Nebius management revealed that customers have already booked computing power for the first and second quarters of 2028, with some orders involving tens of thousands of GPUs. The company's current demand visibility has exceeded 24 months, significantly longer than the previous 18 months.

It is worth noting that this price increase covers both older Hopper architecture GPUs and newer Blackwell series, indicating that market demand is not limited to a single generation of chips, but is widely distributed across various types of AI computing power.

"Nebius is benefiting from two forces simultaneously—the continued expansion of its computing capacity and the increase in fees for existing capacity, " wrote @MelvinInvests. This means that, with utilization remaining high, the revenue generated per available GPU hour will increase significantly, helping to improve profit margins and accelerate the return on infrastructure investment.

Fundamental Support: Large Orders in Hand, Expansion Continues to Accelerate

Nebius' recent fundamentals have also supported market sentiment. According to Stocktwits, the company has been accelerating its expansion since securing a major contract with Microsoft last September, acquiring Staryps, Eigen AI, and Tavily. In March of this year, Nebius announced a $27 billion contract with Meta Platforms and received a $2 billion strategic investment from Nvidia.

Nvidia's quarterly earnings report released on Wednesday further confirms the assessment that upstream demand remains strong, reinforcing the market logic that the boom in AI data center construction by large technology companies will continue to drive demand in the computing power supply chain.

X platform user Jonah Lupton commented that Nebius' average price increase of about 20% means that there is still a significant bottleneck in the supply of computing power in the AI ecosystem. Based on this, he estimated that Nebius' current short-term customer contract price may have reached more than $60 million per megawatt, and the contract price of Vera Rubin architecture products may exceed $80 million per megawatt in the future.

Nebius's stock price has surged by approximately 150% year-to-date, making it one of the best-performing AI concept stocks. Stocktwits data shows that discussions related to NBIS have surged by over 120% in the past 30 days, with the number of followers increasing by 7.3%. As of Thursday morning, retail investor sentiment remained "bullish," and news volume was at a "high" level.

Bargaining power shifts: Data center operator contract terms begin to reverse.

The supply-demand imbalance reflected by Nebius's price increase is reshaping the contract negotiation landscape of the entire data center industry.

Previously, large cloud providers held a dominant position in negotiations, often making extremely demanding requirements: each server rack had to be up to 100% capacity, and the data center's temperature and humidity were subject to very strict standards. A data center executive revealed that he had seen contract clauses such as: if a rack went down due to a power outage, overheating, or switch failure, the cloud provider could cancel six months' rent; if service level agreement (SLA) breaches accumulated to a certain extent, the cloud provider could even terminate the lease directly.

The executive pointed out that negotiating an SLA is essentially a trade-off between "optimal price" and "contract durability"—the more stringent the terms, the higher the price, but the greater the risk. "If we can secure an SLA with lighter penalties, it's worth it even if the price is lower."

However, as operators gain more bargaining power, these extreme terms are gradually softening. There are also signs of a shift in payment terms. A report cites a credit executive who has seen cases where clients rent only a small portion of a large data center, but the contract stipulates that if the client fails to pay on time, they must bear the full rent for the entire facility for a certain period. The data center owner himself is outspoken about this demand: "He said, 'Look, we know this is outrageous… but we can do it.'"

As cloud providers like Microsoft are eager for Nvidia server racks to be operational as soon as possible, data center operators like CoreWeave are gaining more bargaining power, and the supply side of computing power is increasingly taking the initiative in contract negotiations.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.