GPU Cloud Services Surge in Price: Nebius Raises Prices Again by 20%, Computing Power Supply Side Gains Bargaining Power

Nebius has announced a comprehensive price increase for its GPU cloud services effective October 1, with an average uptick of about 20%. Coupled with NVIDIA’s strong quarterly earnings, this drove Nebius stock nearly 7% higher in after-hours trading, further confirming market sentiment that AI computing power supply remains tight relative to demand.

This round of price adjustments covers multiple chip models including H100, H200, B200, and B300, and marks Nebius' second price hike since May this year. The market interprets this as a strong signal that AI computing power demand remains overheated while supply stays tight. As a result, Coreweave shares rose nearly 4% after hours, the optical communication sector generally strengthened, with AAOI and Credo Technology each up nearly 3%, Marvell Technology and Astera Labs up more than 2%, and Coherent up nearly 2%.

The timing of the price hike coincides closely with NVIDIA's impressive financial guidance, further bolstering market confidence in the ongoing boom in AI data center construction and sparking a broad rebound in sentiment across the cloud computing and infrastructure sectors. Meanwhile, data center operators are gaining greater bargaining power, with contract terms shifting in their favor, quietly altering the industry's landscape.

According to pricing screenshots circulating on X, this round of price hikes is significant. The H100 now costs $4.50 per GPU hour, up from $3.85—a gain of about 16.9%; H200 rises from $4.50 to $5.40 (20% increase); B200 increases from $7.15 to $8.50 (approximately 18.9%); and B300 jumps from $7.85 to $9.50 (about 21% increase).

This marks Nebius's second price hike within a few months. In May of this year, the company announced an average 29% increase in on-demand capacity prices and a 51% increase in preemptible capacity. For instance, B300’s hourly price has now climbed about 56% from the pre-May level of approximately $6.10.

In a May email to customers, Nebius stated that the adjustment "reflects the sustained, strong demand for high-end GPU computing power. Even after the updated pricing, Nebius remains one of the most competitively priced GPU infrastructure providers on the market." According to Stocktwits, as of press time, Nebius has yet to publicly respond to the latest price hike rumors.

Market analysts point out that this price hike reflects deeper supply and demand dynamics. According to X user @MelvinInvests, Nebius previously attempted manual price adjustments, but management indicated that strong demand still made the adjustments insufficient to balance supply and demand. Subsequently, the company conducted auction trials for scarce Blackwell computing power, with customers paying 15%–20% above the prior highest prices to secure access.

The analyst also noted that Nebius management revealed customers have already pre-ordered computing capacity for Q1 and Q2 of 2028, with some orders involving tens of thousands of GPUs. The company’s demand visibility now exceeds 24 months, up from about 18 months previously.

It is worth noting that this round of price hikes covers both the older Hopper architecture GPUs and the newer Blackwell series, indicating market demand is not limited to any single chip generation, but is broadly distributed across various types of AI computing resources.

"Nebius is benefitting from two major trends—expanding computing capacity and simultaneously charging more for existing capacity," wrote @MelvinInvests. This means that as utilization remains high, the revenue generated per available GPU hour will increase significantly, improving margins and accelerating payback on infrastructure investments.

Recent Nebius fundamentals have also supported market sentiment. According to Stocktwits, since securing a major Microsoft contract last September, the company has continued to expand rapidly, successively acquiring Staryps, Eigen AI, and Tavily. In March this year, Nebius announced a $27 billion contract with Meta Platforms and received $2 billion in strategic investment from NVIDIA.

NVIDIA’s quarterly earnings report on Wednesday further attested to the continued strength of upstream demand, reinforcing the logic that large tech companies’ AI data center buildout booms are driving sustained demand along the computing supply chain.

X user Jonah Lupton commented that Nebius's average 20% price increase indicates a significant bottleneck in AI computing supply. As a result, he estimates Nebius’s current short-term customer contract prices could exceed $60 million per MW, with future Vera Rubin architecture products potentially breaking $80 million per MW in contract pricing.

Nebius stock is up about 150% year to date, making it one of the standout performers among AI concept stocks. Stocktwits data shows discussion of NBIS has surged over 120% in the past 30 days, with the number of followers up by 7.3%. As of Thursday morning, retail investors maintained a “bullish” rating, and message volume remained “elevated.”

The supply-demand imbalance reflected by Nebius’ price increases is reshaping the dynamics of contracting throughout the data center industry.

Previously, large cloud providers held the upper hand in negotiations, often demanding extremely stringent requirements: each server rack had to run at nearly 100% uptime, with very strict standards for temperatures and humidity. One data center executive reported having seen contract provisions where even a single rack outage from a power cut, overheating, or switch failure would allow the cloud provider to cancel six months’ rent; violations of service-level agreements (SLAs) that accumulated could even allow the cloud provider to directly terminate the lease.

This executive explained that negotiating SLAs is essentially about balancing “optimal price” versus “contract durability”—the stricter the terms, the higher the price, but also the greater the risk. “If you can negotiate for lighter SLA penalties, it’s worth a slightly lower price.”

However, as operators gain bargaining power, these extreme provisions are gradually being softened. There are also signs of change in payment terms. According to a credit executive cited in reports, he saw an example where a client leasing only a small part of a huge data center was nonetheless contractually required to cover the entire facility’s rent for a period if they missed a payment. On this requirement, the data center owner was candid: “He said, ‘Look, we know this is crazy … but we can do it.’”

As Microsoft and other cloud providers urgently seek to deploy NVIDIA server racks, the bargaining leverage of data center operators like CoreWeave is rising, and the supply side of computing power is gaining increasing initiative in contract negotiations.