Technology • Finance • Semiconductor supply chain
What the post claims
A November 18, 2026 article by RespawnFirst[1] argues that Jensen Huang, in his years as Nvidia CEO, ran the company as a de facto central bank for the AI ecosystem, backing AI startups with financial guarantees and direct investments so they could absorb the company’s vast prepaid chip supply from TSMC. With Huang now gone, the article argues, TSMC has responded by tightening credit: asking AI chip customers for cash upfront on orders that exceed their credit ceilings. The headline frames this as cause and effect.
The article cites Taiwan’s Commercial Times and Chinese outlet Kuai Keji as sources for the TSMC policy shift, while the circular financing characterization draws on broader reporting, primarily from the Wall Street Journal.
Claim 1: Nvidia under Huang practiced circular financing, backing AI startups to absorb excess GPU supply
This characterization is substantially supported by Nvidia’s own SEC filings. The question of intent is a separate, contested layer the filings do not resolve: specifically, whether the purpose was to engineer artificial scarcity or to build a legitimate customer ecosystem.
What the SEC filings confirm
Nvidia’s 10-Q for the quarter ending July 26, 2026[8] discloses a web of financial commitments totaling $530.5 billion when supply obligations, guarantees, equity investments, and capacity agreements are summed. The structure is detailed and verifiable:
The word “circular financing“ describes a structure where Nvidia simultaneously acts as investor, loan guarantor, hardware supplier, and backup customer within the same set of deals. Money leaves Nvidia as investments and guarantees; it returns as GPU and wafer sales revenue from those same parties. That loop is documented. Nvidia’s accounts receivable rose from $40.7 billion to $63.1 billion in a single quarter, with extended payment terms of up to one year extended to some customers, consistent with a supplier subsidizing its own buyers.[7]
The MediaTek example confirms the mechanism directly
The article specifically cites Nvidia’s August 31, 2026 subscription to $3.5 billion of MediaTek’s $3.9 billion overseas convertible bond offering. Bloomberg confirmed the deal;[10] the bond was described as the largest ever by a Taiwanese company at the time. Nvidia’s purchase funded MediaTek’s expanded ASIC production capacity and wafer purchases from TSMC. MediaTek, in turn, adopted Nvidia’s NVLink Fusion standard, linking its custom chips into Nvidia’s rack architecture. The circular structure is explicit: Nvidia’s $3.5 billion financed a component maker that then built products compatible with Nvidia’s ecosystem and bought manufacturing capacity from the same foundry Nvidia uses.
What is alleged, not confirmed: the scarcity intent
The Wall Street Journal’s reporting, relayed by multiple secondary outlets,[4][9] goes further, alleging that Nvidia used this strategy specifically to manufacture the appearance of GPU scarcity and keep prices elevated, rather than primarily to build an ecosystem. That specific allegation of intentional deception has not been established by any regulatory finding or company admission. Huang himself disputed it at a Goldman Sachs conference before his resignation, saying: “We put in $1 and get $100 back — if that’s circular financing, we’ll do more of it.”[12] He characterized the strategy as standard investment due diligence, with reviewed contracts totaling $100 billion.
The structural description is confirmed: Nvidia invested in customers to backstop GPU purchases. The characterization of that as a deliberate fraud or scarcity scheme is alleged and contested.
Claim 2: Since Huang’s departure, TSMC is demanding cash upfront for orders above credit limits
This claim is sourced to Taiwan’s Commercial Times (November 16, 2026 print edition)[5] and Chinese outlet Kuai Keji, both citing unnamed “industry sources.” Multiple English-language outlets subsequently reported the same story. TSMC has not confirmed the policy.[6]
What TSMC has disclosed, and what is not new, is that it held NT$234.23 billion (over $7 billion USD) in customer prepayments as of June 30, 2026. The article cites this figure as evidence of the new policy. It doesn’t support that inference: advance payment has been standard in semiconductor foundry contracting for years, and that June figure predates the alleged policy change.
TSMC’s chairman C.C. Wei did discuss evaluating customer financials on prior earnings calls: balance sheets, payment history, and business outlook. That suggests some form of credit assessment was already standard practice before this month. The commercial press reports describe a formalization of that practice into explicit per-customer credit ceilings, but TSMC has neither issued a press release nor responded to inquiries about the policy, as of this publication date.
The claim may well prove accurate: the sourcing is from Taiwan’s business press and plausible given the broader financing environment. But it is not yet confirmed by TSMC, and an unconfirmed industry report is the appropriate level of confidence at this stage.
The causal frame: a better question than the headline asks
The respawnfirst headline frames TSMC’s move as a consequence of Huang’s departure: “With Jensen Huang’s Circular Financing Engine Gone, TSMC Reportedly Demands Cash Upfront.” That framing rests on a 4-day gap between Huang’s resignation (November 12) and the Commercial Times report (November 16). Timing alone does not establish causation.
An alternative reading of the same evidence: TSMC had already been monitoring customer credit quality as AI valuations stretched, and the Commercial Times report surfaced an ongoing internal practice at a moment when Huang’s departure made the story maximally newsworthy. TSMC’s customer credit assessment wasn’t invented in response to Huang leaving; it was described as standard foundry practice before this week.
The better framing for this story isn’t “did TSMC react to Huang leaving?” It’s: was the AI chip supply chain financing ecosystem structurally fragile, and what does the apparent shift in TSMC’s posture signal about its resilience? On that question the evidence is richer and more substantive, and suggests real stress regardless of causation.
Precedent and historical comparison
Compared against: Supplier-funded demand inflation in prior technology cycles. The closest precedents are Lucent Technologies (1999–2001) and Cisco Systems (2000–2001), both of which extended vendor financing to telecom customers to absorb overproduction, then recognized those loans as revenue. When customers couldn’t repay, both companies collapsed into major write-downs: Lucent wrote off $7.8 billion in 2001; Cisco destroyed $2.25 billion in inventory. Nvidia’s total exposure ($530.5 billion) is an order of magnitude larger than either. The MediaTek convertible bond at $3.9 billion was described by Bloomberg as the largest overseas convertible bond ever by a Taiwanese company, a superlative consistent with the scale of the moment, though it also reflects Taiwan’s market size.
The critical structural difference between 2001 and 2026: Nvidia’s guarantees are equity-linked and partially secured rather than pure vendor loans, and the AI demand thesis has documented enterprise adoption behind it, unlike dot-com revenue projections. But the mechanism is the same pattern that ended badly in both Lucent and Cisco cycles: a supplier financing its own customers to clear inventory.
The strongest case for the circular financing model
The case against treating this as a crisis rests on several real points. Nvidia’s investments followed $100 billion in documented forward contracts, not speculative bets. The AI infrastructure buildout has measurable commercial demand from enterprises, cloud hyperscalers, and national AI programs. Circular financing, as Huang noted, is analytically indistinguishable from normal venture investment in a supplier’s ecosystem; Apple invests in display manufacturers that then supply Apple. TSMC itself approved up to $500 million each in Anthropic and OpenAI equity (the same “circular” structure critics assign to Nvidia).
Under Colette Kress’s management, Nvidia is reportedly scaling back the most aggressive commitments (the OpenAI backstop was reportedly halved from $250 billion to $125 billion before Huang’s departure[4]). That suggests a correction in scale, not a collapse of the model. The difference between a sustainable ecosystem-investment program and a Lucent-style implosion depends heavily on whether the AI demand materializes at scale, a question the market hasn’t answered yet.
What the post implies
Implied: The collapse of Nvidia’s circular financing model signals that the broader AI investment bubble is now bursting.
The article’s closing line (“Hyperscaling is in trouble”) and its reference to data center projects “shutting down or slowing down... especially since the blue wave in the United States mid-terms” go further than the evidence supports as a verdict. That Nvidia is winding down some commitments, that TSMC may be tightening credit, and that NAND flash spot prices have dropped 20% since October are real data points. Whether that constitutes a bubble bursting or a supply-demand normalization after three years of AI-driven oversupply is contested among analysts. The evidence is consistent with either reading.
Implied: GPU prices will fall materially for consumers as the excess supply clears.
The logic holds for certain customer segments: mid-tier ASIC designers and smaller AI startups losing access to vendor-backed financing cannot place speculative excess orders. That reduces artificial demand and should normalize pricing pressure in affected segments. Whether this reaches consumer GPU prices depends on whether Nvidia’s data center oversupply spills into gaming GPU allocation, a supply routing decision Nvidia controls. The article’s implication that this is straightforwardly good for gamers is plausible but not certain.
Implied: Jensen Huang’s departure was directly connected to the circular financing exposure becoming untenable, not primarily the affairs scandal.
The article does not make this claim explicitly but the headline and framing strongly suggest it. The official reason for Huang’s resignation was his admission of a romantic affair with a person described as an alleged Chinese national under intelligence scrutiny.[3] Whether the board used the scandal as an opportunity to remove a CEO whose financial commitments Nvidia’s own CFO had already questioned is not established. The two stories are concurrent; whether they are causally related is not settled by any public record.
Implications
If TSMC’s credit policy is confirmed as reported, it functions as a supply-chain sorting mechanism: companies with strong balance sheets (hyperscaler-backed ASIC houses, megacap cloud providers) retain capacity access; leveraged AI startups reliant on convertible bonds and vendor backstops lose it. This accelerates consolidation in the custom AI chip market toward players with direct cloud hyperscaler backing.
The unwinding of Nvidia’s circular financing commitments under Kress creates a different pressure: Nvidia’s $279 billion in supply commitments to its own foundry partners now needs to be absorbed without a financing backstop. If demand doesn’t materialize at the pace those commitments assumed, Nvidia faces either write-downs or renegotiated delivery schedules, both of which carry cost. The Q3 FY2027 earnings call on November 17 showed $390 billion in commitments (up from $530 billion previously tallied across categories, per different accounting), suggesting the exposure is still growing even as Kress rationalizes individual deals.
For thecrosscheck.org’s purposes: the specific alarm the article raises (that TSMC is now demanding cash upfront) deserves confirmation from TSMC before readers treat it as established fact. The underlying structural story about Nvidia’s cross-investment model is well-evidenced. The specific policy change attributed to TSMC is not yet confirmed.
What to watch
- TSMC Q4 2026 earnings call (expected January 2027): Will Chairman C.C. Wei address the credit limit reports? A direct question from an analyst is the most likely path to official confirmation or denial.
- Nvidia Q4 FY27 earnings (expected February 2027): First full quarter under Colette Kress. Watch for changes in accounts receivable, the size of outstanding guarantees, and any disclosure of commitment renegotiations with OpenAI or other large counterparties.
- MediaTek and Alchip filings: If TSMC credit tightening is real, Taiwanese ASIC designers will reference TSMC payment terms in their bond prospectuses and financial disclosures. The first such reference would be a primary-source confirmation.
- Q1 2027 memory contract prices: The article cites a projected 5–10% drop in NAND contract prices. If memory pricing falls further, it weakens the demand narrative that underpins Nvidia’s $279 billion in supply commitments.
- Nvidia’s OpenAI backstop: The $105 billion guarantee on OpenAI’s Ohio data center lease (documented in Nvidia’s 8-K, August 17, 2026) is the single largest contingent liability. Whether Kress renegotiates or unwinds it will be the clearest signal about how far the commitment rollback goes.
Parties affected
| Party | Effect | Why |
|---|---|---|
| Nvidia / Colette Kress | Mixed | Reducing overexposure lowers financial risk; unwinding may reveal write-down liabilities on existing commitments. |
| AI startups and ASIC designers† | Costs | Lose access to vendor-backed financing that bridged the gap between TSMC wafer payments and cloud customer revenue. Must tap equity or bond markets on their own terms. |
| TSMC† | Gains | Tightening credit (if confirmed) reduces TSMC’s exposure to leveraged customers who may default on wafer commitments. Near-term revenue may slow; credit quality improves. |
| Cloud hyperscalers (AWS, Google, Microsoft)† | Mixed | Well-capitalized enough to access TSMC directly; benefit competitively if leveraged rivals lose capacity. Face risk if the AI buildout slowdown reduces their own demand projections. |
| Consumer GPU buyers† | Gains (potential) | Reduction in speculative data-center GPU orders may free up supply and normalize prices for gaming and workstation hardware. Not guaranteed — depends on how Nvidia routes excess inventory. |
† These rows reflect analytical inference about likely outcomes, not documented statements of position from these actors.
What this article is about
A website called RespawnFirst published an article on November 18, 2026, making two big claims. First: that Jensen Huang, while he was CEO of Nvidia, ran the company like a bank — putting billions of dollars behind AI companies so they could buy Nvidia’s chips. Second: that since Huang left, the chip manufacturer TSMC has started asking AI companies to pay cash upfront for large orders. This fact-check asks: are those claims true?
Who is Jensen Huang and what happened to him?
Jensen Huang co-founded Nvidia and ran it as CEO for 33 years. He resigned on November 12, 2026, after news broke of an extramarital affair involving a person reportedly under U.S. intelligence scrutiny. Nvidia’s board accepted his resignation immediately. The company’s chief financial officer, Colette Kress, became interim CEO.
Claim 1: Nvidia under Huang practiced circular financing — Verdict: Mostly True
The basic description holds up. Nvidia’s own SEC filings show $530.5 billion in total commitments and guarantees as of July 2026. Some of the most striking examples:
- Nvidia guaranteed up to $105 billion of OpenAI’s data center lease in Ohio. If OpenAI could not pay its rent, Nvidia would cover it.
- Nvidia committed $25 billion more in future investments in AI companies.
- Nvidia agreed to buy $36 billion worth of unused cloud computing capacity from companies it had invested in.
- In August 2026, Nvidia spent $3.5 billion buying bonds from MediaTek, a Taiwanese chip company. That money helped MediaTek buy manufacturing capacity. MediaTek then agreed to build chips compatible with Nvidia’s systems.
The pattern is clear: Nvidia invested in its own customers so they could keep buying Nvidia’s products. Critics call this circular financing because the money goes in a loop — out of Nvidia, into a customer, and back to Nvidia as chip sales. Nvidia’s CEO disputed this label before he resigned, saying the investments were based on solid contracts worth $100 billion.
What is not proven: whether Huang did this deliberately to create an artificial chip shortage. That is a more serious allegation reported by the Wall Street Journal. No regulator has confirmed it.
Claim 2: TSMC is demanding cash upfront since Huang left — Verdict: Unproven
Taiwan’s Commercial Times reported on November 16, 2026 that TSMC has started requiring cash payments from AI chip companies whose orders go above a set credit limit. Many English-language websites picked up the story.
But TSMC has not confirmed this policy. There is no TSMC press release and no TSMC official has been quoted confirming it. TSMC already held over $7 billion in customer prepayments as of June 2026, but that is a standard industry practice, not proof of a new policy.
The claim may turn out to be accurate. But until TSMC confirms it, readers should treat this as a reported rumor, not a confirmed fact.
Did TSMC’s move happen because Huang left?
The article’s headline implies a direct cause: Huang left, so TSMC started demanding cash. That link is not proven. The Taiwan newspaper report came four days after Huang resigned. Timing alone is not proof of cause and effect. TSMC evaluating customer credit has been standard practice before this week.
Historical comparison
This is not the first time a technology supplier has funded its own customers to clear excess inventory. Lucent Technologies and Cisco did the same thing during the 2000-2001 dot-com crash. When their customers could not repay, Lucent wrote off $7.8 billion and Cisco destroyed $2.25 billion in inventory. Nvidia’s total exposure ($530.5 billion) is far larger. The key difference is that AI computing has real commercial demand behind it, unlike many dot-com business plans in 2001.
What the article implies
The article implies that the AI investment bubble is now bursting. That is contested. Some signals point toward stress: Nvidia is winding down some commitments, memory prices have dropped 20% since October 2026, and some data center projects are slowing. Whether this is a bubble bursting or a normal slowdown in a maturing market is an open question.
What to watch
- TSMC’s next earnings call (expected January 2027): Will its chairman address the credit limit reports?
- Nvidia’s next earnings report (expected February 2027): First full quarter under Colette Kress.
- Taiwanese chip company filings: If TSMC is tightening credit, companies like Alchip and MediaTek will reference TSMC payment terms in their financial documents.
- Memory prices: A further drop in Q1 2027 would add evidence for the slowdown story.