Computing Power in Crisis! SemiAnalysis In-Depth Review: From GPUs to Memory to Optical Fiber, the AI Supply Chain Is Fully Stretched and Prices Are Rising Across the Board

Computing Power in Crisis! SemiAnalysis In-Depth Review: From GPUs to Memory to Optical Fiber, the AI Supply Chain Is Fully Stretched and Prices Are Rising Across the Board

The explosive growth in AI computing power demand is pushing the entire supply chain to its limits. From GPU rental to DRAM memory, from fiber optic cables to data center hosting, prices have soared across the board in just a few months, while supply is almost nowhere to be found.

According to the latest report by research firm SemiAnalysis, the one-year lease contract price for H100 GPUs has surged from a low of $1.70 per GPU per hour in October 2025 to $2.35 in March 2026, an increase of nearly 40%. Meanwhile, on-demand computing power has sold out across all GPU models.

The core drive behind this round of computing power shortage comes from a structural leap in demand. Anthropic’s annual recurring revenue (ARR) soared from $9 billion to over $25 billion in a single quarter, as multi-agent workloads like Claude Code are pushing computing power consumption into parabolic growth.

SemiAnalysis believes that GPU rental prices are highly likely to keep rising in the short term. Neocloud has begun to hold the bargaining initiative, yet the stock prices of related companies like CoreWeave, Nebius, and IREN have not reflected this change.

Demand Side: Multiple Surges, Parabolic Growth in Computing Consumption

The demand-side momentum for this round of computing power shortage comes from several stacked directions.

The explosion of Claude Code is the most remarkable turning point. Anthropic’s ARR skyrocketed from $9 billion to over $25 billion in a single quarter, tripling previous figures. The rise of open-source models like GLM and Kimi K2.5 further expanded the scale of inference workloads.

Meanwhile, large-scale fundraising by AI labs like Anthropic and OpenAI directly created GPU demand.

AI agents are another crucial driver. These workloads execute multi-step workflows with high concurrency and continuous iterations, causing computing power consumption to rise parabolically.

Native media generation platforms Seedance and Nano Banana are fueling massive computing power demand for image and video generation scenarios.

From an economics perspective, this demand is relatively price-inelastic. SemiAnalysis points out that if the investment return rate of AI tools reaches 5 to 10 times, GPU rental prices still have considerable room to rise before demand is suppressed.

"The demand curve shifts upward to the right, providing a strong and relatively inelastic force that pushes GPU rental prices higher."

Supply Side: Soaring Memory Prices, Server Procurement in Chaos

Explosive demand is only one side of the computing shortage; surging supply-side prices are the other.

January 2026 became another critical turning point. At that time, DRAM and NAND memory prices accelerated their rise. According to SemiAnalysis’s memory model, the contract prices of LPDDR5 and DDR5 in Q1 2026 are estimated to rise by about 4x and 5x year-on-year, respectively.

The surge in memory prices quickly passed on to entire servers. Original Equipment Manufacturers (OEMs) re-priced AI servers, but their price increases significantly exceeded the actual rise in component costs.

Higher server procurement costs squeezed expected project returns, forcing some operators to slow down or even shelve new deployment plans. In other words, supply that was originally about to enter the market is delayed, further tightening the rental market.

The supply situation of Blackwell’s new generation GPUs is equally unoptimistic. According to SemiAnalysis, the deployment lead time for new Blackwell clusters has now extended to June-July 2026, mainly due to strong demand from open-source weights models and continued inference compute shortages.

Market Structure: Neocloud Holds Pricing Power, Contract Terms Tighten Across the Board

The power structure in the GPU rental market has undergone a fundamental transformation in six months.

Before the second half of 2025, market competition was fierce, with multiple Neoclouds racing to cut prices to secure asset utilization.

Now, Neocloud and hyperscale cloud providers have total initiative—they can demand higher prepayment proportions, better pricing, longer contract terms, and flexibly arrange contract start and end times based on their inventory.

The GPU rental market can be divided into three main tiers:

Short-term rental (on-demand, spot, and contracts under 3 months): Usually for remaining capacity, this has now sold out across the board, and holders are unwilling to return capacity to the market even in the face of sharp price increases.

Mid-term contracts (3 months to over 3 years): The most active segment in the market. One-year contracts capture marginal demand from non-AI lab clients and are the most sensitive indicator of market tightness.

Long-term bulk agreements (4-5 years): Mainly dominated by large AI labs, single transactions typically reach 50-100 megawatts or more, equivalent to about 24,000 to 48,000 GB300 NVL72 GPUs.

These transactions are highly attractive to Neocloud—by arranging advantageous debt financing with long-term contracts, locking in double-digit project IRR, and avoiding GPU price risks. Sometimes hyperscale cloud providers act as guarantors in these deals, further reducing financing costs.

Price Outlook: Self-reinforcing Spiral, Yet Neocloud Valuations Remain Undervalued

SemiAnalysis believes the likelihood of GPU rental prices continuing to rise in the short term is much greater than falling.

The current dynamics have obvious self-reinforcing characteristics: tightened supply pushes up prices, rising prices prompt Neocloud to accelerate hardware lock-in, which further tightens supply, and prices rise again.

SemiAnalysis compares this to the GPU shortage cycle of 2023-2024, but believes the server market is now mature enough that OEMs’ room for excess profit may be limited.

The rise in rental prices affects Neocloud’s finances in two ways: on one hand, profit margins on deployed capital expand, improving return on invested capital (ROIC); on the other, higher rental prices extend the economic life of existing GPUs, enabling assets to generate cash flow for longer before reinvestment is needed.

In the current environment, SemiAnalysis believes the most obvious beneficiaries among Neocloud are those with shorter contract terms (allowing faster repricing), large H100 inventories, and recently added capacity coming online.

However, there is a significant divergence between this fundamental improvement and public market sentiment. The stock prices of listed Neocloud companies like CoreWeave, Nebius, and IREN remain at low points in their 6-12 month trading ranges.

SemiAnalysis points out that the market is still dominated by the narrative of "eventual supply glut and commoditization," and has not yet fully reflected the continuous scarcity and pricing power clearly visible on the ground.

SemiAnalysis lists three core observational indicators for judging whether GPU rental prices can remain high.

First, the pace of GB300 cluster rollout. Whether new compute capacity can ease the current shortage, or if the growth in compute consumption will continue to outpace new supply—this will determine to what extent AI labs participate in the sub-4 year contract market and further influence pricing in that segment.

Second, the severity of wafer shortages. SemiAnalysis’s previous reports have noted tight supply for TSMC N3 logic wafers as well as HBM, DRAM, and NAND memory. The complex manufacturing process means execution risk is always present.

Third, the growth trajectory of AI lab ARR. The speed of user adoption and sustained token consumption are the fundamental variables determining the slope of the overall demand curve.

In summary, SemiAnalysis's conclusion is clear: Until there are obvious reversals in these three indicators, the direction of compute prices is only one—up.

Risk Warning and DisclaimerThe market has risks, and investments should be made cautiously. This article does not constitute personal investment advice and does not take into account individual users’ special investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing accordingly is at your own risk.