U.S. AI infrastructure stocks rallied on Oct. 27 as traders focused on the scale of long-term computing commitments and a new financing structure designed to fund data-center equipment without requiring the end customer to purchase every system outright. The Philadelphia Semiconductor Index rose 3.6%, with all 30 of its constituents finishing higher, while the Nasdaq Composite gained 1.18% and the S&P 500 added 0.60%.
Broadcom rose 4.00%, Oracle gained 3.55%, Cerebras advanced 10.10%, and CoreWeave climbed 7.94%. Nvidia added 0.90%, while NXP Semiconductors finished the regular session up 3.12% before falling about 7% in after-hours trading following its earnings release.
The session extended well beyond the largest chip designers. Nebius jumped 11.22%, IREN gained 7.90%, and electronics manufacturer Celestica rose 11.93%, reflecting demand for companies supplying computing capacity, servers, networking and data-center infrastructure.
The market’s reaction centered on a growing distinction within the AI trade: announcements about future model demand are being supplemented by contracts, lease structures and financing arrangements that could determine who bears the cost of supplying the hardware.
Anthropic-related commitments put focus on future capacity
Anthropic’s IPO filing, combined with media calculations based on disclosed agreements, brought fresh attention to the size of its potential computing and infrastructure requirements. Media estimates placed the upper ceiling of Anthropic-linked long-term arrangements at about $807 billion, including cloud services, equipment leases, procurement agreements and future capacity plans involving supply channels connected to Broadcom, Google, Amazon, Microsoft and Nvidia.
That figure does not represent recognized Anthropic revenue or a single aggregate contractual obligation disclosed in its S-1 registration statement. The filing did not provide one total commitment number, and reported estimates differed depending on whether analysts included conditional arrangements, unpriced capacity and optional components.
Compiled reporting put signed payment obligations at approximately $417 billion. Agreements involving capacity scheduled for delivery between 2028 and 2031, where final pricing had not been established, could add roughly $125 billion to $150 billion in obligations under the same estimates.
The range of estimates, from $542 billion to $807 billion, illustrates the difficulty of treating long-term AI infrastructure announcements as conventional order books. Some commitments may be contingent on deployment schedules, equipment availability, pricing negotiations or actual computing consumption. Yet the figures also point to a pipeline large enough to shape procurement decisions across chipmakers, cloud providers, networking suppliers and power-intensive data-center operators for years.
Broadcom financing shifts equipment costs into an SPV
A separate announcement from Broadcom on Oct. 26 added a financing mechanism to the capacity story. Broadcom said Bank of America, Blackstone, Citigroup and Morgan Stanley would participate in a $63 billion arrangement involving AI accelerators and networking equipment intended for Anthropic deployments.
Under the proposed structure, a special purpose vehicle, or SPV, would purchase Broadcom XPU accelerators and networking hardware, then lease those systems to Anthropic. Deliveries are expected to begin in 2027, according to Broadcom.
The arrangement gives Anthropic access to equipment through a leasing model while placing the assets within a separately financed vehicle. Broadcom said it would provide residual value support associated with the first-lien portion of the financing, leaving the company exposed to how much the equipment is worth after use as well as the customer’s lease payments.
That residual-value element places more attention on a question that has become central to AI infrastructure financing: how quickly accelerators depreciate when new chip generations arrive. The economics of an SPV depend on lease income, financing costs, customer performance and the value that equipment retains in secondary markets. A large commitment can support demand visibility for suppliers while also creating credit and asset-value risks that do not appear in a standard product sale.
Broadcom’s latest reported quarterly figures show why it has attracted attention as a supplier capable of supporting large-scale deployment models. The company reported $29.591 billion in revenue, $13.088 billion in net profit and $14.197 billion in operating cash flow for the quarter. Capital expenditures totaled $532 million, leaving free cash flow of $13.665 billion, according to its financial results.
For its latest full fiscal year, Broadcom reported $63.887 billion in revenue, a 23.9% increase. Its residual-value support for the planned SPV will be watched alongside equipment delivery schedules, customer concentration and margins from AI-related products.
Oracle expands its hardware options
Oracle added another layer to the AI capacity buildout on Oct. 27, announcing plans to expand deployment of Nvidia’s Vera Rubin systems in cloud infrastructure that could reach multi-gigawatt scale. Oracle also said it would offer Google TPU capacity on its cloud platform for the first time and add Cerebras inference services.
Inference refers to the process of running a trained AI model to generate an answer or complete a task. It has become a major infrastructure market alongside training, which requires the computing power used to build and refine models.
Oracle’s approach would place Nvidia systems, Google TPUs, Cerebras offerings and Oracle’s own Acceleron networking chips within the same cloud platform. The company did not disclose transaction values for the hardware plans, so the announcement cannot be directly converted into confirmed revenue or unit shipments. It does indicate that cloud operators are seeking multiple architectures for different workloads rather than relying on one supplier for every AI task.
Oracle’s financial profile also shows how capital-intensive the strategy has become. The company reported quarterly revenue of $19.345 billion and operating cash flow of $23.103 billion, but capital expenditures reached $28.499 billion. That produced negative free cash flow of $5.396 billion for the quarter. Oracle reported fiscal-year revenue of $67.358 billion, up 17.3%.
CoreWeave reported a similar gap between fast revenue growth and infrastructure spending. It posted quarterly revenue of $2.575 billion, capital expenditures of $6.422 billion and negative free cash flow of $5.743 billion. Its fiscal 2025 revenue rose 168% to $5.131 billion from $1.915 billion a year earlier.
Cerebras reported $180 million in quarterly revenue, down from $193 million in the preceding quarter, and a net loss of $451 million. Its fiscal 2025 revenue reached $510 million, up 75.7%.
The next phase of the AI infrastructure rally will likely depend less on broad demand projections and more on measurable execution: cloud revenue growth, utilization rates, equipment delivery schedules, borrowing costs and the resale value of rapidly evolving accelerators. Those figures will determine whether today’s long-term capacity commitments translate into durable cash generation for the companies building the physical foundation of AI.
Explore how tokenized equities can mirror AI infrastructure exposure through on-chain trading and innovative financing structures.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
