Author: Junxi
On October 28, Alphabet, Google's parent company, released its third-quarter financial results: operating activities generated approximately $42.3 billion in cash, while capital expenditures on equipment and properties reached about $51.4 billion, resulting in a negative free cash flow of roughly $9.1 billion. Although search, advertising, and cloud businesses continued to grow, the cash generated from operations in the quarter was insufficient to cover capital investments.
For neocloud providers such as CoreWeave and Nebius, which specialize in purchasing GPUs and renting out computing power, funding directly determines their expansion speed. On October 22, Jensen Huang addressed this group, stating that demand for computing power is extremely robust, service capabilities are strong, but working capital is lacking; simultaneous investment expansions across multiple AI sectors have pushed capital needs beyond available funds.
Three days later, Elon Musk offered a more pessimistic outlook. He noted that over the past few weeks, debt markets have largely stopped providing financing to companies lacking substantial revenue outside of AI data centers, predicting that over two hundred neoclouds will undergo mass bankruptcies and consolidations, with only a few surviving. While Jensen Huang believes the funding issue can be resolved, Musk argues that many companies may not survive long enough to see that day.
With one side showing continuously growing demand for computing power and the other facing increasing scrutiny over investment burdens and debt repayment pressures, are the big chip buyers truly running out of money?
01 Orders Are Rising Fast, But Cash Isn't Keeping Up
In recent years, neoclouds like CoreWeave and Lambda have expanded their clusters through GPU-backed loans, with funders extending from private credit institutions to major banks. Operators use equipment and customer contracts as collateral to secure upfront procurement funds, repaying debts with computing power rental income. This structure has supported rapid expansion but makes future procurement increasingly dependent on how lenders assess equipment value and customer creditworthiness.
In early October, in NVIDIA-backed computing financing projects, some banks and asset management firms demanded stronger safeguards than originally proposed. Lenders are concerned not only with whether GPUs can be rented out today, but also with how much they will earn in a few years, whether customer payments will remain sustainable, and what residual value the equipment holds in case of default.
Divergent views on GPU valuation have emerged. Tony Trzcinka, Senior Portfolio Manager at Impax Asset Management, pointed out that banks typically evaluate GPUs based on a 3-to-4-year depreciation schedule, whereas NVIDIA believes high-end GPUs can generate revenue for up to ten years. The physical lifespan of equipment and its ability to sustain sufficient income generation are not the same thing. For neoclouds lacking other business revenues, more conservative collateral valuations and higher guarantee requirements could mean needing to contribute more equity or accept more expensive financing.
CoreWeave has also faced pressure in the bond market. A batch of bonds issued by the company in June had an initial yield of approximately 9.7%; by mid-October, as trading prices fell, market yields rose to about 15%. Investors are demanding higher returns to bear its credit risk, meaning subsequent issuances or refinancing may incur higher costs.
Its second-quarter accounts showed net cash inflow from operating activities of approximately $679 million, while cash outflows for equipment and property purchases reached $6.422 billion. In the same quarter, net interest expense was about $640 million, equivalent to a quarter of its revenue. The company must purchase equipment for customer-booked computing power while servicing existing debt; sustained expansion cannot rely solely on current quarterly operating income.
On October 20, responding to debt concerns, CoreWeave emphasized that most of its debt is backed by long-term customer contracts. The company secures purchase commitments from creditworthy customers first, then uses these to finance infrastructure construction; the contracted revenue from these agreements is sufficient to repay related principal and interest without relying on contract renewals. Its backlog of contracted revenue stood at $104 billion at the end of Q2, with an additional $25 billion in new commitments secured in early Q3.
Existing contracts provide repayment sources for current debts, but new procurement still requires ongoing fundraising. If lenders demand higher returns and more guarantees, operators must recalculate the economics of expanding clusters even when securing new orders. The pressure on neoclouds thus extends from sheer scale of investment to the cost and terms of each financing round.
The sustainability of these procurement plans has also entered TSMC's capacity assessment. C.C. Wei stated during the October 15 earnings call that the company considers factors such as cloud vendors' capital expenditures and financial health, communicating with end customers. While wafer foundries serve chip designers directly, advanced process nodes and packaging capacity require lead times, and demands from designers ultimately depend on spending by end buyers. Scrutinizing these buyers' financial status helps TSMC gauge how long demand can be sustained.
02 Chipmakers Are Also Participating in Fundraising
NVIDIA has already provided support in several projects. Transaction details disclosed on October 28 revealed that Indonesian computing platform Zankore is bringing in more participating banks for a $3.1 billion GPU loan. Its six-year agreement with NVIDIA includes revenue sharing and credit support, covering unsold computing capacity in the first phase of a 100MW facility. Related financing was announced in September, with recent disclosures clarifying the supplier's involvement.
According to NVIDIA's previously described general model, the company provides minimum revenue guarantees for certain computing capacities, giving financial institutions a basis to support project construction. Neoclouds can still sell computing power to third parties; once conditions are met, NVIDIA participates in revenue sharing. Operators gain support beyond their own creditworthiness, while NVIDIA earns potential rental revenue shares beyond hardware sales, assuming corresponding support responsibilities.
Model companies also need to raise funds for equipment. On the custom chip front, Broadcom has joined Anthropic's financing arrangements. On October 26, Broadcom and financial partners announced a $63 billion financing deal where a newly formed project company purchases its custom AI accelerators and network equipment, leasing them to Anthropic with deliveries expected to start in 2027. Banks and Blackstone provide upfront procurement funds; the model company acquires equipment via leasing, and Broadcom organizes supply accordingly.
To make this arrangement viable, Broadcom provided equipment residual value support. Regulatory filings submitted on October 27 show that after all racks are deployed, its maximum potential liability is approximately $43 billion, calculated on an undiscounted basis. This liability increases with rack deliveries and decreases as Anthropic pays rent; if the client defaults, Broadcom may bear the difference between agreed outstanding amounts and equipment disposal value.
The $43 billion represents a potential liability cap, not cash already paid by Broadcom. However, the implications are specific: suppliers must monitor clients' ability to pay subsequent rents even after selling equipment. Financial institutions provide capital, clients amortize equipment costs, and Broadcom accepts partial risk to facilitate procurement, tightening the link between chip supply and financing structures.
Broadcom supports equipment value, while NVIDIA supports partial computing revenue in the aforementioned model. Both approaches provide lenders with an extra layer of security and enable clients struggling to independently shoulder massive investments to expand procurement. For suppliers, winning such orders involves not just product performance, delivery, and price, but also the extent of support offered to clients.
Lenders, however, maintain their own judgments. In early October, some banks and asset managers involved in NVIDIA-related financing projects demanded stronger safeguards than initially proposed. Supplier participation hasn't eliminated risk but shifted who bears it. Whether transactions can further expand depends on whether financial institutions accept these terms and how much residual responsibility chipmakers are willing to retain for new sales.
03 Capital Is Still Flowing In, But Terms Vary
The day after Musk expressed skepticism about neocloud financing, SpaceX announced obtaining approximately $40 billion in GPU lease financing commitments. Entities managed by Apollo will purchase GPUs and related equipment, leasing them to SpaceX, with leases commencing in batches alongside equipment deliveries in the first half of 2027. Financing remains subject to closing conditions, but the committed scale indicates large GPU procurements still attract financial institution participation.
When explaining SpaceX's financing advantage, Musk highlighted its investment-grade credit and revenue from Starlink and launch services. Unlike neoclouds primarily relying on GPU rentals for debt repayment, SpaceX presents different funding sources to lenders. Being pessimistic about some operators' financing prospects while raising funds for its own procurement expansion is not contradictory; the difference lies in who has the capacity to sustain investments.
Cash from mature businesses gives hyperscalers more flexibility. Microsoft disclosed on October 28 that net operating cash inflow for the quarter ended in September was $58 billion, with capital expenditures on equipment and properties totaling $39.4 billion, leaving $18.6 billion in free cash flow. The company projects approximately $170 billion in capital expenditure for calendar year 2026, expects positive free cash flow in fiscal year 2027, and stated it would use additional financing when appropriate.
Alphabet, mentioned earlier, raised approximately $41.1 billion net through common stock and senior unsecured notes in the same quarter. Operating cash flows failing to cover current capital investments did not prevent it from continuing to access capital markets. For such buyers, fundraising is part of arranging massive investments; procurement capability cannot be judged declining solely based on financing actions.
Neoclouds cannot all leverage such business revenues, but some have secured support through major customer contracts. On October 22, Nebius signed a $3.5 billion credit facility with Citigroup for data center construction and GPU procurement. The company stated that its long-term contracts exceeding $40 billion with investment-grade clients, including Microsoft and Meta, can support subsequent asset financing.
This illustrates another relationship between hyperscalers and neoclouds. Microsoft and Meta invest in their own infrastructure but also purchase computing power from external operators. Neoclouds handle hardware procurement and operations for clients, using multi-year payment commitments from these customers to secure financing. Lenders focus not only on the neocloud's scale but also on who is on the other side of the lease and whether contract revenue is reliable.
Nebius's recent financing has a 364-day term, to be used alongside customer prepayments and operating cash flows to support near-term expansion. Long-term funding for 2027 and beyond still needs to be arranged. Major customer contracts haven't solved all funding needs but have enabled phased procurement, distinguishing its financing capability from operators lacking stable tenants.
On October 27, BUZZ Alpha Cloud under HIVE also announced a three-year loan of approximately $129 million at a fixed annual interest rate of 9% for purchasing GB200 clusters, planned to be repaid with revenue from a three-year contract with Cohere. The loan covers about 80% of acquisition costs, with remaining equity coming from previous convertible bond proceeds; equipment is expected to be delivered by year-end.
With clients and matching-duration contracts, operators contributing partial equity, this procurement has concrete funding arrangements. The 9% fixed rate reminds the market that securing funds and cheap funds are two different things. Neoclouds must compare not just GPU rental prices and equipment costs, but also borrowing expenses and how these affect the scale of future procurement they can afford.
04 Conclusion
Big chip buyers have not collectively run dry, but sustained procurement expansion is testing each party's fundraising capability. Mature business revenues, long-term customer contracts, and supplier support determine the conditions under which different buyers can continue investing. For the chip industry, beyond the magnitude of demand, it is crucial to see who can pay according to plan, complete procurement, and sustain this level of investment over time.