Sixth Street chief warns of US$200 billion in stranded chip inventory! AI infrastructure faces credit divergence as "turning compute orders into cash flow" begins to dominate AI investment.

Sixth Street Partners CEO Alan Waxman said in a media interview on Wednesday that AI capital expenditure is at an unprecedented scale, while unit economics remain changing and difficult to predict. Rapid credit expansion combined with the herd mentality of chasing the AI investment boom has led some banks to seek large-scale sales of construction loans, and Sixth Street has therefore largely stayed on the sidelines, particularly cautious on debt investments. Waxman stressed that he sees increasing signs of turbulence in the AI boom, which is why the firm has largely maintained a wait-and-see stance. Unprecedented incremental capital has been invested in AI compute infrastructure, but how much money can actually be made from each hour of compute sold remains difficult to predict reliablythis is what Alan Waxman means by "unit economics." For new cloud AI compute lessors such as CoreWeave and Nebius that focus on selling/leasing compute resources, this typically means treating one billable GPU hour as a unit and comparing the rent collected for that hour with its allocated share of equipment depreciation, data center rent, power and O&M costs. Why does this calculation keep changing? Rents change, and the actual billable hours also change. Assume a single NVIDIA Corporation AI GPU has a fixed cost of US$48 per day. If it can be billed for 16 hours per day, the fixed cost allocated to each paid hour is US$3. If it can only be billed for 12 hours, the fixed cost rises to US$4 per paid hour, with variable costs such as power billed separately. If this coincides with the launch of a new-generation chip, a reduction in rents for old equipment, or delays in AI data center construction/delivery, the project will simultaneously face lower revenue per hour and higher allocated cost per hour. If the launch of a new-generation chip is followed by lower rents for old equipment, while data center construction or delivery is delayed, reducing billable hours, the project may face dual pressure: revenue per billable hour falls, while fixed costs already incurred must be allocated across fewer billable hours. This is especially important for lenders. A delay in a data center coming online postpones the first compute payment collection, but equipment procurement may already have been paid for, and interest begins accruing after loan drawdown. What Waxman is asking is whether a project can still repay principal and interest on schedule under more conservative assumptions for rent, utilization and commissioning time. The larger the capital expenditure, the more likely a slight deviation in these assumptions can a substantive cash flow gap. The AI trade begins to focus on "who can turn compute orders into cash flow" The AI investment frenzy across global stock and bond markets appears to have entered what Goldman Sachs Group, Inc. calls the "trough of disillusionment," as the market begins to test whether signed compute orders can be delivered on time and efficiently converted into real cash flow, and whether collections can keep pace with debt service timing. Sixth Street CEO Waxman estimates that about US$200 billion of chips are sitting in warehouses because data centers are not yet ready. If data center delays compress billable hours, fixed costs already incurred must be allocated across fewer billable hours. If this coincides with the launch of a new-generation chip and lower rents for old equipment, the project will also face pressure from lower revenue per hour. Some banks are already seeking to resell construction loans, and if insurers' exposure to a single borrower approaches limits, developers may also face higher financing costs. This is precisely the screening period over the next 18 months that Social Capital founder Palihapitiya describes; whether end customers are willing to keep paying for business outcomes brought by AI will affect the renewal of compute contracts, rental prices and collections. Goldman Sachs Group, Inc. is relatively optimistic about the returns of large cloud platforms, while new cloud compute providers reliant on external financing are more sensitive to commissioning delays and rent declines. This round of trade divergence ultimately seems to require close tracking of the same compute investment: who can start billing on schedule, sustain renewal prices, and cover principal and interest with actual collections. From idle chips to tightening credit, Sixth Street CEO warns that the AI compute infrastructure race is rife with "speculative behavior" In a television interview broadcast on Wednesday, Waxman said the scale of AI capital expenditure globally is unprecedented, "and no one really understands its unit economics benefits, because it is still constantly changing." He stressed in the interview: "At least from a credit investment perspective, it is difficult for us to assess this type of investment. The situation for equity investment is slightly different." Waxman stressed in the interview that San Francisco-based Sixth Street is concerned about the rapid expansion of credit in this field, as well as the "speculative frenzy, animal spirits, or fear of missing out." The firm has begun to see signs of market turbulence, such as banks seeking to sell construction loans. As concerns about risks in the field deepen, lenders have scaled back financing for AI infrastructure projects. "The entire ecosystem is rife with speculative behavior," Waxman said. His firm manages more than US$140 billion in assets, with a focus on private credit. Waxman, 52, also said he is concerned that insurers are reaching lending limits to single borrowers. Insurers have been one of the main sources of lower-cost financing for AI projects. This will force developers to seek more expensive financing from other institutional investors such as pension funds and sovereign wealth funds. "Because data centers are not yet ready, about US$200 billion worth of chips are still sitting in warehouses," Waxman said. He added in the interview that chips can depreciate rapidly, which poses a risk. Waxman said turbulence in the AI sector could hit equity valuations and reduce the willingness of higher-rated funding providers to finance projects. Banks seek to resell construction loans, testing the capacity of low-cost funding to absorb them Waxman stressed that rapid credit expansion combined with frenzied bullish sentiment chasing the AI investment boom has pushed some banks to seek to sell construction loans, and Sixth Street has therefore largely stayed on the sidelines, particularly cautious on debt investments. Creditors' income mainly comes from agreed interest, but project delays compress the repayment buffer. Equity investors have the opportunity to share in upside growth, so they can have different judgments on the same project. Recently, media reports said ING Groep NV Sponsored ADR (ING) plans to sell about US$500 million of data center loans, including a loan share related to the New Mexico "Stargate" project, which can be described as the "Manhattan Project of the new era." Waxman is also concerned that some insurers' lending quotas to single borrowers are approaching limits, and developers may be forced to raise funds from institutions demanding higher returns. For example, he said bluntly in the interview that he personally predicts about US$200 billion of chips are stranded in warehouses because data centers are not yet ready. For lenders, delays have two effects: the time when a project starts billing is pushed back, and chip iteration may also equipment resale value. Once project return downgrades transmit to equity valuations, highly rated fund providers may also tighten financing. For investors, judging whether a project can navigate this adjustment can focus on three specific key indicators: when a large AI data center project is powered on, when it starts billing, and when collections can cover principal and interest. AI investment enters the "trough of disillusionment," and the next 18 months will test who can turn compute resources into real cash flow Wall Street financial giant Goldman Sachs Group, Inc., in its heavyweight research report "Top of Mind" released on November 10, described the current stage as the "trough of disillusionment" for AI investment. Stock and bond market volatility, increasingly complex circular financing structures, and regulatory uncertainty after the US midterm elections have led investors to re-examine how much return massive capital expenditure can bring. Social Capital founder Chamath Palihapitiya sees the next 18 months as a screening period. As more AI companies come under public market scrutiny, investors will more easily distinguish which compute commitments can be delivered on time, generate paying demand and recover capital, and which still rely on the next round of financing to sustain expansion. Whether model service providers can make customers pay for quantifiable business outcomes is also a clue for testing end demand. For new cloud computing service providers such as CoreWeave and Nebius (Neoclouds), pressure is basically concentrated in the timing mismatch among equipment investment, billing revenue and debt service. GPU procurement, site and power commitments often precede revenue. If grid connection or data center delivery is delayed, compute billing will be postponed, while the technology refresh cycle for equipment continues. Palihapitiya predicts that the break-even utilization rate for such companies is about 70%75%, and the actual threshold also depends on rental prices, electricity costs, depreciation, contract terms and financing costs. CoreWeave disclosed in the third quarter about US$132.5 billion of revenue to be recognized, part of which depends on whether delivery can be completed and services provided. Operating profit in the same quarter was US$207 million, and net interest expense was US$846 million. These figures show that while order visibility has improved, the financing burden has already entered current profit and loss. Before AI neocloud leader CoreWeave released its latest earnings report, the credit market had already sounded the alarm that AI investment returns may fall far short of expectations, or even that debt pressure could cause CoreWeave to go bankrupt before receiving strong cash flow. From September 25 to October 26, the five-year credit default swap (CDS) spread of CoreWeave (CRWV.US)the stock at the very center of the "AI compute selloff storm" and "AI bubble burst" narrativerose from 835 to 1,148 basis points. Under a 40% recovery assumption, the model estimates a corresponding five-year cumulative implied default probability of about 62%this indicator also mainly includes credit risk premium and cannot be directly equated with the company's actual default probability over the next five years; in addition, Meta's CDS spread for the same maturity rose from 98 to 121. The CoreWeave CDS spread then narrowed to 1,041.4 basis points on the 27th, indicating that pressure began to ease. On November 5, its five-year CDS further rose above 1,289 basis points, then retreated somewhat. These data also continuously highlight that the 10-year US Treasury yield at historically high levels raises the overall financing benchmark, and credit spreads are then layered according to cash flow, leverage and reliance on refinancing. Joseph Lupton of JPMorgan's global economic research team noted in a November 10 study that recent labor productivity improvement has come more from capital deepening and increased intensity of factor utilization, while the broad efficiency dividends brought by AI through process reengineering, data integration and organizational adjustment still need time to materialize. To a large extent, this also means that orders for chips, servers and power equipment can grow first, while sustained investment returns for compute operators and end users may not arrive simultaneously; in an environment of rising interest rates and widening credit spreads, this waiting period itself will increase financing burdens and the present value of forward returns. Production and revenue increments during the AI construction phase first transmit to chip, server and related equipment manufacturers, which is why the share prices of Asian AI infrastructure-related manufacturing leaders have been exceptionally strong in recent years. If rising stock correlations further trigger unwinding of crowded dispersion trades, the adjustment could be amplified. What will determine whether AI assets can regain a premium will be the speed at which orders convert into cash, whether financing capacity can support delivery, and whether cost reduction and efficiency gains on the application side can be gradually realized. Large cloud vendors have more mature customer channels and stronger balance sheets, and Goldman Sachs Group, Inc. is relatively optimistic about their returns. Calculations by Wall Street financial giant Goldman Sachs Group, Inc. show that for six major US hyperscale cloud service providers to achieve a 15% return on invested capital (ROIC) on related investment in 20262027, they would need to generate about US$1.42 trillion in cumulative revenue in 20282030. At the same time, five hyperscale cloud vendors had issued about US$296 billion of debt year-to-date as of early November, while AI-related high-yield financing has seen almost no new deals completed in the past month. Viewed together, the two sets of figures show that financing divergence has already emerged. If end demand, funding supply or power delivery continue to fall short of expectations, some projects may be delayed, and assets already in operation may also shift to buyers with stronger financing capacity.