The Financial Pressure on Big Chip Buyers: From Alphabet's Negative Free Cash Flow to Rising Financing Costs for New Clouds

The big buyers of chips are running out of money?

Original source: Semiconductor Industry Observer Published: 2026-10-31 15:10 Author: Junxi

Computing power

Google-generated poster

Computing power orders are still increasing, but the money to buy chips is becoming harder and harder to arrange.

On October 28, Google's parent company Alphabet released its third-quarter financial report: operating activities generated about $42.3 billion in cash, but purchases of equipment and property incurred about $51.4 billion in expenditures, resulting in negative free cash flow of about $9.1 billion. Search, advertising, and cloud businesses are still growing, but the cash generated by operations in the quarter is no longer enough to cover capital investment.

For new cloud service providers such as CoreWeave and Nebius, which specifically purchase GPUs and rent out computing power, funding more directly determines the pace of expansion. On October 22, when Jensen Huang spoke about this group, he said that demand for computing power is extremely strong and their service capabilities are also very strong; what they lack is working capital. Multiple areas of AI are expanding investment at the same time, and funding needs have exceeded available capital.

Three days later, Musk offered a more pessimistic judgment. He said that in the past few weeks, the debt market had almost stopped providing funds to companies lacking large-scale revenue outside of AI data centers, and predicted that more than two hundred new clouds would undergo large-scale bankruptcy and consolidation, with only a few able to survive. Jensen Huang believes the funding problem can be solved, but Musk believes many companies may not last until that day.

On one side is continuously growing demand for computing power; on the other are growing concerns about investment and debt repayment pressure. Are the big buyers of chips really running out of money?

01 Orders are rising fast, but cash has not kept up

In recent years, new clouds such as CoreWeave and Lambda have expanded their clusters through GPU-backed loans, and funding providers have expanded from private credit institutions to large banks. Operators use equipment and customer contracts as the basis for financing, obtain procurement funds in advance, and then repay debt with computing power rental income. This arrangement has supported rapid expansion, but it also makes the next round of procurement increasingly dependent on how lenders assess equipment value and customer credit.

In early October, in a computing power financing project promoted by Nvidia, some banks and asset management institutions demanded stronger safeguards than the original plan. Lenders care not only about whether GPUs can be rented out today, but also about how much money they can earn several years from now, whether customer payments can continue, and how much the equipment can be sold for after default.

The two sides already disagree on the value of GPUs. Tony Trzcinka, senior portfolio manager at Impax Asset Management, pointed out that banks usually evaluate GPUs based on a three- to four-year depreciation schedule, while Nvidia believes high-end GPUs can generate revenue for as long as ten years. How much longer the equipment can be used is not the same thing as whether it can continue to generate sufficient revenue. For new clouds lacking other business revenue, more conservative collateral valuations and higher guarantee requirements may both mean they need to put up more of their own funds or accept more expensive loans.

CoreWeave has also encountered pressure in the bond market. A batch of bonds issued by the company in June had a yield of about 9.7% at issuance; by mid-October, as trading prices fell, the market yield had risen to about 15%. Investors are demanding higher returns to bear its credit risk, and the company may also need to accept higher financing costs for subsequent bond issuance or refinancing.

Its second-quarter accounts show that net cash inflow from operating activities was about $679 million, while cash expenditure on purchases of equipment and property reached $6.422 billion. In the same quarter, net interest expense was about $640 million, equivalent to a quarter of revenue. The company must both buy equipment for computing power reserved by customers and bear interest on existing borrowings; sustained expansion cannot rely only on cash generated from operations in the quarter.

On October 20, when CoreWeave responded to debt concerns, it emphasized that most of its debt is supported by long-term customer contracts. The company first obtains purchase commitments from customers with stronger credit, then uses them to finance and build infrastructure; the contracted revenue from these contracts is sufficient to repay the relevant principal and interest without relying on customer renewals. At the end of its second quarter, revenue corresponding to remaining performance obligations reached $104 billion, and in early the third quarter it obtained another $25 billion in new commitments.

Existing contracts provide a repayment source for the relevant debt, but new procurement still requires continued fundraising. If lenders demand higher returns and more guarantees, operators will have to recalculate whether expanding clusters is worthwhile even if they obtain new orders. The pressure facing new clouds thus extends from the large scale of investment to the cost and terms of each round of financing.

Whether these procurement plans can continue has also entered TSMC's capacity assessment. C.C. Wei said at the October 15 earnings call that the company would consider factors such as cloud vendors' capital expenditure and financial condition, and would communicate with end customers. The direct customers of wafer foundries are chip designers, but advanced process and packaging capacity must be arranged in advance, and the demand proposed by designers must ultimately be supported by the investment of end buyers. Verifying these buyers' financial condition helps TSMC judge how long demand can last.

02 Chipmakers are also starting to participate in raising money

Nvidia has already provided support in some projects. Transaction information disclosed on October 28 shows that Indonesian computing power 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 100MW facility. The related financing was announced in September, and the recent disclosure makes the way suppliers participate more concrete.

Under the general model Nvidia previously described, the company provides minimum revenue commitments for part of the computing capacity, giving financial institutions a basis to support project construction. The neocloud can still sell computing power to third parties, and once conditions are met, Nvidia participates in revenue sharing. The operator gains support beyond its own credit, while Nvidia, beyond hardware sales, has an opportunity to share in subsequent leasing income while bearing corresponding support obligations.

Model companies also need to raise equipment funding. On the custom chip side, Broadcom has already become involved in Anthropic's financing arrangements. On October 26, Broadcom and financial partners announced a $63 billion financing transaction, under which a newly established project company will purchase its custom AI accelerators and networking equipment and then lease them to Anthropic, with deliveries expected to begin in 2027. Banks and institutions such as Blackstone provide upfront procurement funding, the model company obtains the equipment through leasing, and Broadcom organizes supply as planned.

To make this arrangement work, Broadcom provided equipment residual-value support. A regulatory filing submitted on October 27 shows that after all racks are deployed, its maximum potential liability is about $43 billion, calculated on an undiscounted basis. This liability increases as racks are delivered and decreases as Anthropic pays rent; if the customer defaults, Broadcom may bear the difference between the agreed outstanding amounts and the equipment's disposal value.

The $43 billion is a potential liability ceiling, not cash Broadcom has already paid. Yet the change it represents is very concrete: after a supplier completes equipment sales, it still needs to watch whether the customer can pay subsequent rent. Financial institutions provide funding, the customer bears equipment costs in installments, and Broadcom exchanges partial risk-taking for procurement progress, tightening the link between chip supply and financing arrangements.

Broadcom supports equipment value, while Nvidia in the above model supports part of computing-power revenue. Both approaches give lenders an extra layer of protection and give some customers that would struggle to bear huge investments on their own a chance to expand procurement. For suppliers, winning such orders involves not only product performance, delivery and price, but also how much support they are willing to provide to customers.

But lenders still make their own judgments. In early October, some banks and asset management institutions participating in Nvidia-related financing projects demanded stronger protections than the initial proposals. Supplier participation did not make risk disappear; it changed who bears the risk. Whether transactions can continue to expand depends on whether financial institutions accept these conditions and how much follow-on liability chipmakers are willing to retain for new sales.

03 Money is still coming in, but on very different terms

The day after Musk expressed pessimism about neocloud financing, SpaceX announced about $40 billion in committed GPU financing leases. An entity managed by Apollo will purchase GPUs and related equipment and then lease them to SpaceX, with leasing expected to begin in batches alongside equipment deliveries in the first half of 2027. The financing still needs to meet agreed closing conditions, but the committed scale shows that large GPU purchases can still attract financial institutions.

When explaining SpaceX's financing advantages, Musk emphasized its investment-grade credit and revenue from Starlink and the launch business. The sources of funds it can show lenders differ from those of neoclouds that mainly rely on GPU rental income to service debt. Being pessimistic about some operators' financing prospects while raising funds for its own expanded procurement is not contradictory; the difference lies in who is able to bear sustained investment.

Cash provided by mature businesses also gives large cloud vendors more room. Microsoft disclosed on October 28 that net cash from operations in the quarter ended September was $58 billion, cash spending on equipment and property was $39.4 billion, and free cash flow remained $18.6 billion. The company expects calendar 2026 capital expenditure of about $170 billion and expects free cash flow to remain positive in fiscal 2027, while saying it will use additional financing when conditions are appropriate.

Alphabet, mentioned earlier, raised about $41.1 billion net in the same quarter through common stock and senior unsecured notes. Operating cash flow did not cover capital investment in the period, but that did not stop it from continuing to obtain funds from capital markets. For such buyers, raising money is part of arranging huge investments; financing actions alone cannot be used to judge that their purchasing power is declining.

Not all neoclouds can draw on such business revenue, but some of them have already obtained support through major customer contracts. On October 22, Nebius signed a $3.5 billion financing facility with Citi for data center construction, GPU procurement and other purposes. The company said its long-term contracts with investment-grade customers worth more than $40 billion, including Microsoft and Meta, can support subsequent asset financing.

This also reveals another layer of the relationship between large cloud vendors and neoclouds. Microsoft and Meta both invest in infrastructure themselves and buy computing power from external operators. Neoclouds bear hardware procurement and operations for customers, then use those customers' multi-year payment commitments to obtain financing. What lenders value is not only the neocloud's own scale, but also who is on the other end of the lease and whether the contract revenue is reliable.

Nebius's facility has a term of 364 days and will be used alongside customer prepayments and operating cash flow to support near-term expansion. Long-term funding for 2027 and beyond still needs to be arranged. Major customer contracts do not solve all funding needs at once, but they provide conditions for phased procurement and distinguish its fundraising ability from operators lacking stable tenants.

On October 27, HIVE's BUZZ Alpha Cloud also announced a three-year loan of about $129 million at a fixed annual interest rate of 9%, to be used to purchase GB200 clusters and planned to be repaid with revenue from a three-year Cohere contract. The loan covers about 80% of the purchase cost, with the remainder coming from proceeds of a previous convertible bond financing, and the equipment is expected to be delivered before the end of the year.

With a customer and a contract with matching terms, and with the operator also bearing part of the funding, this purchase thus has a concrete financing arrangement. The 9% fixed rate also reminds the market that being able to raise money and money being cheap are two different things. What neoclouds need to compare is not only GPU rent and equipment prices, but also borrowing costs and how those costs affect the scale the next batch of purchases can bear.

04 Conclusion

The big chip buyers have not all fallen into a funding drought together, but continuously expanding procurement is testing each one's fundraising ability. Revenue from mature businesses, long-term customer contracts and support provided by suppliers determine on what terms different buyers can continue to invest. For the chip industry, beyond how large demand is, it also matters who can pay on schedule, complete purchases, and how long such investment can last.

This content is original to the platform, and the copyright belongs to the platform. Reproduction without permission is prohibited; to reprint, please contact the platform. Comments: No user comments yet. Latest news More Related recommendations

Net profit surged 241%, but non-recurring net profit fell by 20%; where did the money in AMEC's third-quarter report come from? 2026-10-31 15:07

Kioxia's single-quarter revenue exceeds the full previous fiscal year, while slower NAND price increases reveal performance concerns 2026-10-31 15:05

In 2026, everyone selling things is talking about chips 2026-10-31 15:03