Server Price Increases in 2026: Forecasts, Causes, and Recommendations

Every time a new server market forecast is released, the comments section follows the same pattern. Some claim the shortage is real; others dismiss it as manufacturer marketing. We decided to investigate the primary sources.

The beginning of 2026 feels different from previous waves of price hikes. Dell announced server price increases as early as December 2025, with Lenovo following suit in January. Samsung and SK Hynix raised prices on server dynamic random-access memory (DRAM). The shortage of servers based on graphics processing units (GPUs) has evolved from a quarterly issue into a chronic one.

For those planning infrastructure, the question has become uncomfortably practical. Buy equipment now or wait? Rent a dedicated server or move to the cloud? In which segments will prices truly skyrocket, and where will growth remain moderate?

Surge in Demand for Compute Power

Analyst estimates vary significantly here, and it is honest to acknowledge this upfront. According to DataIntelo, the GPU-based server market reached $8.7 billion in 2024 and could grow to $44.9 billion by 2033. IDC takes a broader view: the entire server market showed $95.2 billion in just the first quarter of 2025, a 134% increase year-over-year. This explosive growth is largely explained by a low comparison base and the rapid scaling of artificial intelligence (AI) infrastructure. IDC estimates the year-end total at $366 billion, representing a 44.6% growth over 2024.

The primary driver pulling this market upward is the AI race. Training large language models (LLMs) requires tens of thousands of GPUs, and production cannot keep up with demand. Research Nester estimates the AI server market will grow from $169.8 billion in 2025 to $3.47 trillion by 2035. The figure sounds fantastical, but this is precisely the capital major players are currently investing in AI infrastructure.

Next, we will break down what specifically drives prices in each segment, which forecasts appear justified, and how to handle this in practice.

Current Server Market Situation (Early 2026)

Hardware Prices

By early 2026, the server market is experiencing one of the sharpest component shortages in recent years. Dell announced server price increases of 15–20% as early as December 2025, with Lenovo following in January 2026. Samsung and SK Hynix raised prices on server DRAM by 60–70% compared to the fourth quarter of 2025. The primary targets of these increases are major clients like Google and Microsoft.

According to TrendForce, DRAM inventory levels shrank critically by the end of 2025. Contract prices for server DRAM rose by 50% in 2025, with another ~20% increase projected for early 2026. In February 2026, NAND flash memory prices jumped 25% in a single month.

Cloud Service Prices

Cloud provider OVHcloud is currently the only major supplier to publicly announce upcoming rate hikes. CEO Octave Klaba confirmed a 5–10% price increase between April and September 2026. AWS, Microsoft Azure, and Google Cloud remain officially silent, although they purchase equipment from the same manufacturers as OVH. A similar rate increase of 5–10% appears to be the baseline scenario for all major players, simply with a delay of a few months.

Traditionally, there is a three-to-six-month lag between rising procurement costs and rate changes for clients. The peak of cloud rate hikes will likely occur in the second half of 2026.

GPU Shortage

Morgan Stanley forecasts that demand for NVIDIA AI server racks will grow from approximately 28,000 units in 2025 to at least 60,000 units in 2026, more than doubling. According to Reuters, Chinese companies have already ordered over two million H200 chips for 2026, while NVIDIA has only about 700,000 units in stock. High Bandwidth Memory (HBM), essential for AI accelerators, is fully contracted through the end of 2026. Micron confirmed this as early as December 2025.

Factors Influencing Prices

Surge in AI Demand

ChatGPT, Claude, and Other Large Language Models

The race between OpenAI, Anthropic, Google DeepMind, Meta AI, and hundreds of startups has become the primary driver of demand for server hardware. Training a single large language model (LLM) requires thousands of GPUs and many months of operation. According to NVIDIA CEO Jensen Huang, the compute power required for AI has already grown 100-fold in recent years, and this growth continues.

Model Training Requires GPUs

The Thermal Design Power (TDP) of flagship NVIDIA GPUs has risen from 700W for the H100 to 1000W for the B200 and 1200W for the GB200. The Vera Rubin (VR200) platform, scheduled for shipment in the second half of 2026, demonstrates a TDP of up to 2300W per GPU. The GB200 NVL72 rack consumes approximately 120kW in total. Liquid cooling is no longer optional at these parameters; it is becoming the standard for new data centers (DCs).

Demand Exceeds Supply

Analysts estimate that NVIDIA's share of the discrete GPU market reached 92% in the first half of 2025. Bloomberg Intelligence estimates the company's share of the AI accelerator market at 70–75%, a figure expected to hold through the end of the decade. The company controls not only the GPUs but also the software stack—the CUDA platform—without which most AI frameworks cannot function. This creates sustained demand that is relatively insensitive to price changes.

Chip Shortage

Supply Chain Issues

Production of advanced chips is concentrated primarily at TSMC, whose 3-nanometer lines are operating at maximum capacity. By the end of the fourth quarter of 2025, 77% of the company's revenue came from 7nm and more advanced technologies. Even NVIDIA is forced to request production ramp-ups, yet physical lithography limitations prevent this from happening instantly.

Geopolitical Situation

US export restrictions on advanced chips for China have created a backfire effect. Chinese companies are rushing to purchase export-allowed equipment (H20 and MI308), further depleting already scarce inventories. According to Reuters, AMD has received orders from Chinese buyers for MI300 series chips, including the MI308, which is positioned as a more affordable alternative to the H20.

Additional market pressure stems from the escalating military-political situation in the Middle East. According to Bloomberg, insurance for vessels passing through the Strait of Hormuz has risen to 5% of the vessel's value, and Hapag-Lloyd has introduced a war risk surcharge of up to $3,500 per container. Rising oil prices trigger a chain reaction throughout the supply chain, increasing fuel costs for maritime container ships, air freight, and ground logistics. However, transportation is only the first link.

Morningstar analysts note that dependence on oil means significantly higher costs for AI data centers, which consume three to five times more energy than standard facilities, potentially substantially increasing the total cost of ownership for major cloud providers. Electricity accounts for about half of a data center's operating expenses; if memory prices rise due to supply instability and energy costs simultaneously, operators may reduce capital expenditures and demand for semiconductors. According to Sean Kim, head of Morgan Stanley's Asian technology division, as cited by Data Centre Magazine, disruption of the Strait of Hormuz could impact electricity costs, material supplies, and the economics of AI infrastructure construction.

The conflict has also impacted infrastructure itself. According to Fierce Network, in the first days of the conflict, four AWS data centers were damaged by drone strikes in the UAE and Bahrain. Furthermore, Qatar produces more than one-third of the world's helium, which is critical for semiconductor production, with no viable alternative for this gas. While a mass collapse of server equipment supply is not yet observed, experts note that the immediate risk to the tech sector lies not in direct halts of chip production but in broader inflationary impacts via energy and transport costs.

TSMC and Production

New Intel factories (Fab 52 in Arizona) and TSMC capacity expansions in the US will not yield results until 2027–2028. Until then, production bottlenecks will remain a structural constraint on supply.

Cryptocurrencies

The impact of cryptocurrency mining on the GPU market in 2026 is secondary compared to the AI frenzy, yet it has not disappeared entirely. Periods of rising Bitcoin and Ethereum prices traditionally create additional demand for video cards, competing with gamers and small-scale AI researchers. Unlike the 2021–2022 peak, server GPUs (H100, A100) are virtually unused in mining; however, consumer RTX series cards face dual pressure: from AI tasks on edge devices and from miners.

Energy Crisis

Global data centers have become the largest consumers of electricity. GB200 racks consume over 100kW, and the announced Vera Rubin platforms will reach the megawatt level per rack. This implies a multiple increase in costs for electricity, cooling, and specialized infrastructure construction. In many regions, grid capacity has already become a limiting factor for opening new data centers, directly affecting hosting and server rental costs.

Inflation and Exchange Rates

General inflation of production costs, rising logistics expenses, and the strengthening of the US dollar against several currencies create additional pressure on prices in local markets. For Russian companies working with imported equipment, currency risks compound the general price increase, and the final cost escalation may be significantly higher than in dollar terms.

Expert Forecasts for 2026

Morgan Stanley Analysis

Morgan Stanley characterized 2026 AI servers as "incredibly expensive." Platforms like GB300, Vera Rubin, and NVIDIA's Kyber architecture, along with AMD's Helios project (based on the MI400 series), will bring significant growth in compute density per rack. At the same time, demand for NVIDIA racks will more than double—from 28,000 to over 60,000 units annually.

TrendForce и IDC

TrendForce forecasts that contract prices for server DRAM in Q1 2026 will rise by 55–60% compared to the previous quarter, while NAND flash memory will increase by 25% in February alone. Micron estimates the High Bandwidth Memory (HBM) market at $35 billion in 2025, growing to $100 billion by 2028.

Memory manufacturers are already abandoning long-term contracts in favor of quarterly pricing. Judging by statements from Samsung and SK Hynix, this practice will continue through 2027.

Manufacturer Opinions

NVIDIA confirmed that all six Rubin series chips have returned from production partners and are preparing for release in 2026. The company is consciously maintaining high prices—demand exceeds supply at any price level.

AMD recorded record sales of EPYC Turin server processors in Q3 2025. The MI308 accelerator is positioned as a more affordable alternative to the H20 for the Chinese market.

According to KeyBanc analysts (January 2026), both Intel and AMD have sold out their server CPU inventories for 2026, driven by hyperscaler demand. Both companies are reportedly considering price increases of up to 15% to manage the supply-demand balance.

Market Segments: Detailed Forecasts

GPU-Based Servers

The most acute situation has developed in the AI accelerator segment. NVIDIA H100 and A100 remained in short supply throughout 2025, and the Blackwell platform (B200/GB200) suffers from the same supply issues. The expected price increase for GPU-based servers will be 30–50% compared to early 2025 levels.

The shortage of GPU-based servers will persist through the end of 2026. TSMC production capacity is fully contracted; new capacity at US factories will not take effect until 2027.

CPU Servers

The situation in the server processor segment is significantly calmer. Intel and AMD are experiencing demand growth, but data center (DC) processor production is less centralized and less prone to shortages. The expected price increase is 10–15% compared to early 2025 levels. AMD EPYC Turin servers demonstrate a good price-to-performance ratio and are seeing growing demand from major ODM suppliers.

Cloud Virtual Dedicated Servers

The baseline forecast for major cloud operators (hyperscalers) (AWS, Azure, GCP) is a 5–10% rate increase in the second half of 2026. OVH has already confirmed a 5–10% price hike between April and September 2026. Services with high memory consumption, such as Redis, ElastiCache, and in-memory databases, are most susceptible to price increases. Compute-optimized instances may see a more modest rise of 3–7%.

Spot instances remain a key cost-saving tool. AWS offers discounts of up to 90% off on-demand prices, albeit with the risk of interruption. For fault-tolerant tasks like batch data processing or machine learning (ML) training with checkpoints, this is the most economical cloud option.

Dedicated Servers

The segment of physical dedicated server rentals expects price increases in the 10–20% range. Hosting providers updating their equipment fleets in 2026 are forced to factor in increased procurement costs into their rates. Servers purchased before the price hike wave (late 2024 – early 2025) may still be offered at old prices for several months, creating a "window of opportunity" for long-term rentals.

Actionable Advice for Business and Individuals

The "wait and see" strategy in current conditions guarantees a 10–50% price increase depending on the segment.

1. Lock in current prices as early as possible

   Long-term server rental contracts signed today will protect against rate hikes in 2026–2027. Most telecom providers offer 15–30% discounts for annual or two-year contracts.

2. Optimize the use of existing resources

   Auditing current infrastructure often reveals 30–40% unused resources. Server consolidation, migration to containerization (Kubernetes), and database query optimization can reduce hardware needs without sacrificing performance.

3. Consider alternative hosting providers

   Major cloud providers have the highest margins and can hold older prices longer, but their rates will rise. Regional and niche hosting providers (including Hostkey) often offer more competitive terms for dedicated servers.

4. Consider ARM-based processor architectures

   Servers based on ARM architecture (Ampere Altra, AWS Graviton 3/4) consume less energy and are cheaper for comparable performance in most web workloads and microservices. Savings compared to x86 can reach 20–40%.

5. Use spot compute resources for the cloud

   For machine learning, rendering, analytics, and any interruptible computations, spot instances remain the optimal choice. AWS provides a 2-minute warning before interruption, while Google Cloud automatically applies discounts for continuous usage.

Alternative Solutions

Chinese Telecom Providers

Cloud platforms like Alibaba Cloud, Tencent Cloud, and Huawei Cloud offer competitive prices for AI accelerators, including US chips formally banned for export to China but accessible via Hong Kong jurisdictions. For international businesses, this can be a viable alternative, though jurisdictional risks should be considered separately.

Local Servers vs. Cloud

The golden rule of 2026 remains simple. With server utilization above 60–70% in a 24/7 mode, purchasing or renting dedicated equipment will be cheaper than the cloud. For variable workloads, the cloud remains the optimal option.

When Will Prices Stabilize: Forecasts for 2027–2028

New Chip Production

Intel Fab 52 in Arizona and TSMC expansions in the US will begin commercial production in 2027–2028. This will gradually reduce production concentration and lower geopolitical supply risks. However, the full effect on prices is not expected until 2028.

Cooling of AI Hype

Historically, technology supercycles (personal computers in the 1990s, smartphones in the 2010s) ended with periods of market oversaturation. The current AI cycle will likely reach peak demand by 2027, after which a correction in both prices and procurement volumes is possible. Developments in efficient models, particularly DeepSeek, may accelerate this process by reducing compute requirements for model inference. IDC also noted this, highlighting that the release of DeepSeek R1 raised questions about whether such massive infrastructure investments are justified.

Yearly Forecast

Historical Analysis: Price Changes 2020–2025

The current crisis is far from the first and certainly not the last. The history of the server hardware market over the past five years offers several important lessons:

The key takeaway from history is that every supply crisis in the tech sector resolved within 18–36 months of onset. However, those who failed to lock in prices at the start of the rise bore the losses.

How Large Companies Are Responding

OpenAI, Google, Microsoft

Microsoft invested over $10 billion in OpenAI, ensuring priority access to Azure server capacity. Google is developing its own Tensor Processing Units (TPUs) v5 and v6 series, reducing dependence on NVIDIA. OpenAI announced the construction of its own data centers under the Stargate project with Microsoft and SoftBank, with a stated budget of $500 billion over five years.

Mid-Sized Business

Mid-sized companies are increasingly adopting hybrid strategies. They place baseline loads on dedicated servers via long-term contracts and handle peak loads in the cloud using spot instances. This minimizes costs while maintaining flexibility.

Frequently Asked Questions

Will virtual servers (VPS) become more expensive in 2026?

Yes. OVH has already announced a 5–10% increase between April and September 2026. AWS, Azure, and GCP will likely announce similar hikes in the second half of the year. Services with high memory consumption are most vulnerable; standard compute instances will see moderate price increases.

Should I buy a GPU-based server now?

If you have a specific need for GPU computing, do not delay. Prices for H100 and B200 are near peak levels, but no decrease is expected in 2026 as inventories are bought out for a year ahead. Consider renting from specialized hosting providers as an alternative to purchasing.

When will prices drop?

Significant normalization in the GPU-based server segment is not expected before mid-2027. The CPU server and VPS markets will stabilize faster—possibly by late 2026 or early 2027. Full shortage relief depends on new production capacity coming online (TSMC USA, Intel Fab 52) and how quickly the AI frenzy cools down.

What is cheaper: cloud or dedicated server?

With constant utilization above 60–70%, a dedicated server becomes cheaper than the cloud after 12–18 months of operation. Given rising cloud rates, this payback period is shortening. There is also good news: the price increase for dedicated servers (10–20%) remains lower than for cloud GPU resources.

Conclusion and Recommendations

The server hardware market in 2026 is experiencing a structural, not cyclical, crisis. The convergence of several factors (AI supercycle, physical chip production limits, geopolitics, energy constraints) has created conditions for simultaneous price increases across all segments.

In short, the situation looks like this: GPU-based servers will rise 30–50%, with shortages persisting through the end of 2026. Server DRAM has increased by 60–70% in a quarter, which must be factored into budget planning. Cloud services will add 5–10% in the second half of 2026. CPU-based and dedicated servers will show moderate growth of 10–20%.

Normalization is not expected before 2027–2028.

Three actions worth taking now: Lock in rental terms before rate hikes in April–June 2026. Conduct an infrastructure audit, as optimization often reduces the need for new equipment by 30–40%. Explore ARM alternatives and spot instances with potential savings of 20–40% compared to current on-demand prices.