Author｜Zhang Qiming

Editor｜Wang Yimo

// Advancing the Multi-Cloud Decentralized Procurement Approach

According to Bloomberg, Anthropic has inked a six-year, $10 billion compute capacity deal with Volta Infra, a cloud infrastructure startup supported by Nvidia.

Behind this large-scale forward compute procurement, leading large model companies are restructuring their infrastructure supply chains. Yet, amidst these substantial investments, concerns about commercialization and cyclical fluctuations persist and cannot be overlooked.

Major Agreement Ensures Stable Supply

The agreement centers around a 133-megawatt data center cluster in Norway, which will utilize Nvidia’s next-generation AI chips to meet the training and inference requirements of the entire Claude model series.

As a top global large model provider, Anthropic’s Claude series excels in long-form text and logical reasoning. Its enterprise API and service revenues are on the rise, while model updates and intelligent agent business expansions are fueling exponential growth in compute consumption.

Prior to this, Anthropic had forged compute partnerships with several entities, including Google, Amazon, and SpaceX. This latest collaboration with a compute startup underscores its commitment to a multi-cloud decentralized procurement strategy.

Traditional public cloud giants are facing constrained compute order queues, while emerging compute operators like Volta Infra can offer customized large-scale GPU clusters.

For Anthropic, the primary benefit of this long-term agreement is to secure compute supply and cost stability for the coming years, mitigate supply chain limitations imposed by single cloud providers, and ensure that the development pace of next-generation frontier models remains unhindered by hardware constraints.

This deal also redefines the collaborative dynamic between supply and demand, transcending mere cloud service transactions.

Leading model companies leverage large forward contracts to support data center construction and hardware delivery timelines, while compute startups depend on long-term orders to secure financing and build confidence. Both parties are deeply intertwined, sharing infrastructure cycle risks.

Industry Transformation, Shifting Dynamics

The emergence of this $10 billion compute order underscores a shift in the AI industry’s competitive landscape.

The AI race is no longer solely about algorithms and model parameters; chips, power, data centers, and cluster operations now form the industrial bedrock that defines the upper limits of large model companies.

Historically, large model companies primarily sourced compute from traditional public cloud providers like Microsoft, Amazon, and Google.

Today, a new wave of compute startup service providers is emerging, armed with chip channels and resources for overseas large-scale green data centers, steadily diverting significant orders from traditional cloud providers.

TrendForce data indicates that the year-on-year growth rate of global AI server shipments for 2026 has been revised upwards to nearly 31%, with substantial increases in capital expenditures from the world’s nine core cloud service providers and a growing number of participants in the compute supply sector.

Multi-cloud and multi-supplier strategies have become the norm among leading AI companies.

Both OpenAI and Anthropic engage multiple compute providers simultaneously to hedge against uncertainties caused by capacity constraints and price fluctuations.

Over-reliance on a single cloud provider can directly disrupt model iteration schedules if supply contracts are not met. Decentralized procurement enhances bargaining power and mitigates supply chain risks.

Compute leasing and intelligent computing operations are currently enjoying periodic dividends, with the “water seller” narrative being repeatedly emphasized by the market.

However, it’s crucial to note that this sector has formidable barriers, requiring access to high-end chips, large-scale power infrastructure, and substantial capital investments. Not all compute startups can manage billion-dollar long-term orders.

Hidden Risks in the High-Stakes Venture

Beneath the allure of the $10 billion long-term agreement, the entire supply chain must grapple with practical limitations.

For Anthropic, the $10 billion over six years represents significant forward expenditures, with contractual obligations unaffected by market conditions. If subsequent commercialization growth falls short of expectations and revenues cannot cover escalating compute costs, the company’s cash flow will face immense strain.

AI chips evolve rapidly, and compute assets secured years ago risk hardware technology depreciation, introducing uncertainties in return on investment.

Compute startups also shoulder heavy burdens. Securing a long-term agreement does not ensure smooth operations; subsequent chip procurement, data center construction, and cluster deployment demand massive upfront capital expenditures. Any delays in chip supply or power infrastructure shortfalls will expose the company to delivery default risks. If global compute capacity becomes concentrated in the future and compute rental prices decline, profit margins on long-term agreements will also be compressed.

Across the industry, substantial capital is pouring into the compute sector, with a continuous rollout of large-scale intelligent computing clusters.

If new supply is released concentrically in the future, the market will swiftly transition from undersupply to oversupply, driving down compute rental prices and subjecting participants across the supply chain to cyclical downturns.

Massive compute investments will also amplify the Matthew effect in the industry, with leading companies using capital to secure sufficient hardware resources for accelerated iteration, further widening the compute gap with smaller model teams. Compute can be purchased with money, but product launch, commercialization validation, and genuine customer demand cannot be directly traded for capital.

Anthropic’s $10 billion compute agreement signifies a pivotal moment in the AI industrialization era, with multi-cloud decentralized procurement expected to become the standard for leading companies.

Therefore, the capital market must look beyond the order’s allure and rationally assess the inherent tensions between long-term expenditures, technological iteration, and supply-demand cycles. The compute race is merely a means; achieving a commercial closed loop is the ultimate determinant of success.