The answer to the two test cases: demand is still there, but the bills have arrived too

The answer to the two test cases: demand is still there, but the bills have arrived too

AdenXie

Automatically translated. Read the Chinese original.

Introduction: This is the sixth instalment in the AI Capital Expenditure series. The first instalment‘AI: The Mainstay of Capital Expenditure’A discussion on how computing infrastructure drives macroeconomic growth in the US; Part Two‘In the same week, macro and micro perspectives delivered conflicting verdicts on AI capital expenditure’The dichotomy between describing GDP as ‘very strong’ and financial results as ‘too expensive’; Part Three‘The Political Calculus of AI Capital Expenditure in the Wake of the Mid-Term Elections’Assessing the impact of political variables at state level; Part Four‘Pricing Three Times a Week’Treat the 11–17 financial results and the 11–18 briefing as ‘the first and second halves of the same event’; Part Five‘Jensen Huang Resigns’It is said that what NVIDIA has lost is its political patron.In that post, we set out two specific conditions for a pullback this week. They have now all been met.


0. This time, we set the exam papers ourselves.

At the end of Part V, we set out two clear test conditions:

  1. 17 November: NVIDIA Q3: Data centre results meet expectations and gross margin holds steady at 74% → ‘Sustained’ correction to 40 per cent
  2. 11–19 Anthropic IPO: If the share price falls below the issue price on the first day (dropping below $70) → ‘Breakdown’ followed by a further upward revision

When they were written, they were conditional sentences. Now they have become school reports.

This series has undergone numerous revisions: the ‘S-1’ piece overturned its own valuation model; the ‘Split Verdict’ piece retracted the claim that ‘AI accounts for half of growth’; and the ‘Jensen Huang’ piece admitted to having ‘predicted the turning point but not the actual scenario’. However, this is the first time the threshold has been set in stone in advance and the author has assessed their own performance upon expiry. Having set the criteria themselves, they must accept the marks they have given themselves.

Results:I passed both questions, but I didn’t feel entirely comfortable with either of them.

Decision cards for two checkpoints and scenario iteration
Decision cards for two checkpoints and scenario iteration

Figure 1Decision cards for two test points. On the left is Anthropic (priced at $83, closed at $97.17 on the first day, did not breach $70; evidence of demand is confirmed), on the right is NVIDIA (data centre revenue of 107.6 billion, non-GAAP gross margin of 74.9 per cent; both criteria met, but Q4 gross margin guidance has been lowered to 71–72 per cent); at the bottom are the scenario iterations: 35/45/20 → 42/40/18.


I. Key Point 1: Anthropic has not traded below its IPO price, and has in fact risen amidst a falling market

Let’s look at the figures first

Project Results
Offering price $83 per share(Above the range $70–78)
Number of shares issued 980 million shares(Originally set at 825 million,Issue of 155 million additional shares)
Total issue size Approx. $81.3 billion($81.3B; of which approximately 23,648,600 existing shares were sold, raising approximately $1.96 billion)
Net funds raised by the company Approx. $79.4 billion($79.4B)
Over-allotment option 147 million shares, 30 days,Not yet exercised
Opening on the first day $111.50
Highest on the first day $115.10(up 38.6 per cent on the issue price)
Closing price on the first day $97.17(+17.07%)
The following day (20 November) Approx. $87.65(-9.8 per cent, intraday snapshot; not confirmed at close), still approximately 5.6 per cent above the issue price

Ruling: There was no breach; the claim is substantiated. Our prediction of ‘a break above $70 → a sharp upward revision’ did not materialise.

However, if one were to write only this sentence, it would oversimplify the matter.

Three points that must be made clear

Anthropic currently has three ‘market capitalisations’ in existence, depending on the denominator:

Calibre Number of shares Priced at $83 Corresponding to a closing price of $97.17
A · Shares in circulation 13.087 billion shares(Class A 7.428 billion + N 5.659 billion) Approx. $1.09 trillion Approx. $1.27 trillion
B · Fully diluted Approx. 18 billion shares(including potential ordinary shares such as options and RSUs) Approx. $1.5 trillion Approx. $1.75–1.8 trillion
Source of the shortfall Approx. 4.9 billion shares — —

The text below states that ‘market capitalisation’ refers to Method A by default. There is a 65 per cent difference between 1.09 trillion and 1.8 trillion; as news headlines tend to emphasise the larger figure, it is worth asking what the denominator is when reading them. Of the 4.9 billion-share discrepancy, 3.86 billion shares were set aside for the 2021 share scheme alone.

Anthropic’s three market capitalisation metrics and first-day price trajectories
Anthropic’s three market capitalisation metrics and first-day price trajectories

Figure 2Anthropic’s three market capitalisation metrics. Above is the price trajectory on the first day ($83 → $111.50 → $115.10 → $97.17 → next day’s snapshot $87.65); below is a comparison of the three methodologies: 13.087 billion shares in circulation correspond to $1.09 trillion / $1.27 trillion; fully diluted at approximately 18 billion shares corresponds to $1.5 trillion / $1.75–1.8 trillion. The difference between the two metrics is approximately 4.9 billion shares, representing potential ordinary shares.

The downside: it rose during a bear market

19 November, the day of the listing,The Nasdaq fell by 2.3 per cent, whilst the Dow Jones fell by 1,101 points。

In other words,This share rose by 17 per cent against the trend in a falling market. There are two sides to this detail:

  • The positive side: The demand is genuine. Whilst the broader market was falling, the share received more than six times the number of applications than the number of shares on offer (Reuters, citing Bloomberg), and rose by 17 per cent on its first day of trading – this was not driven by the broader market.
  • The downside: It fell by nearly 10 per cent the very next day. The hype lasted for just one trading day.

In the previous instalment, we used the phrase ‘making up for price with volume’; this week, we need to upgrade that to ‘making up for price with volume, and then some’: it accepted a valuation even lower than the rumoured $2 trillion (approximately $1.5 trillion on a fully diluted basis), and alsoSold 19 per cent more shares(825 million → 980 million shares). Buyers are willing to purchase, but will only buy more if offered a larger discount.

Who is selling at $111, and who is buying at $111?

This is my personal view, and I’d like to put it quite bluntly.

Let’s start by looking at a few figures:

Project Numbers
Offering price $83
Opening on the first day $111.50(It opened 34 per cent above the issue price)
Highest on the first day $115.10
Trading volume on the first day Approximately 846 million shares(A total of just 980 million shares were issued on this occasion)
Actual allocation ratio for retail investors Approximately 15 per cent(CNBC, citing sources familiar with the matter)
Co-founder sells shares in the IPO per person 3.38 million shares, totalling approximately $1.96 billion

On the first day, 846 million shares changed hands, which is almost equivalent to the entire issue being traded once.

Who’s willing to sell at $111? Institutions that were allocated shares at the $83 issue price. According to a CNBC report, half of the institutions that placed orders came away empty-handed, as Anthropic gave priority to ‘partner institutions’. Those institutions that were allocated shares were able to sell them within the first minute, without any restrictions.

Who is not allowed to sell? The 15 per cent of retail investors who received the $83 allocation. Fidelity stipulates that selling within 15 days of the listing constitutes ‘speculation on new listings’, whilst Robinhood and SoFi allow 30 days; those who breach the rules will have their eligibility to subscribe to new share issues suspended for several months or even permanently. CNBC made this very clear:These rules apply only to retail investors.

Who’s buying at $111? Those retail investors who failed to secure an allocation and bought in on the exchange after the market opened. This offering has been described as ‘the IPO with the highest demand from retail investors in history’.

As the details of the transactions are not made public, I cannot name the specific institution that sold the shares. However, the structure is clear:The rules leave the door open for institutions, whilst locking it shut for retail investors. It fell from $115 to $97, then dropped further to $88 in pre-market trading the following day; it is highly likely that the retail investors who bought in at the opening were the ones who absorbed this decline.

This is exactly what I mean by not treating retail investors as human beings:Retail investors’ role in this offering is not that of investors, but of liquidity providers. The issuer wants a buzz, institutional investors want to cash out, and brokerages want order; the rules cater to every party—except for the person who bought in at $111. It isn’t illegal, but it’s not exactly honourable. A public-benefit company that has enshrined ‘the long-term interests of humanity’ in its corporate mission has, on its very first day of listing, placed those who trusted it most at the very top.

The political manoeuvring began even before the listing

A point in time that is easily overlooked:Senator Warren wrote to Dario Amodei on 27 October… more than three weeks ahead of the IPO. The letter pointed out that the company itself had acknowledged that AI could pose ‘catastrophic or existential risks’, and warned that ‘entering the public market on an unprecedented scale could cause market-wide volatility on the very first day of trading’.

She also singled out one particular issue:Index providers have abolished the waiting period for new shares, enabling newly listed companies to be included in pension accounts more quickly.

This tip paid off on 19 November: Rapid inclusion in the Nasdaq 100 (assessed at the close of the seventh trading day; a company need only rank amongst the top 40 by market capitalisation to qualify, with inclusion taking place approximately 15 trading days later; SpaceX set a precedent in July this year). On the same day, four senators from both parties introduced an AI disclosure bill requiring listed companies to disclose more AI-related information.

Regulatory intervention had already begun before pricing took place.


II. Key Point 2: NVIDIA outperformed across all four metrics, but has already factored in next quarter’s results

Let’s look at the figures first

Project Results Comparison
Revenue $115.844 billion($115.844B),+103% YoY、+20% QoQ Consensus: 109.1 billion
Data Centre $107.6 billion($107.6B),+110% YoY、+21% QoQ Consensus: 101.87 billion
GAAP gross profit margin 74.8% —
Non-GAAP gross profit margin 74.9% Approximately 74 per cent,Beyond the Guidelines
EPS GAAP $2.70 / Non-GAAP $2.68 —
Q4 Revenue Guidance $130 billion ±3 per cent($130B) Consensus: 125.1 billion
Q4 Gross Profit Margin Guidance 71–72% A fall of around 3 percentage points month-on-month
China GuidanceDoes not contain anyRevenue from computing power in Chinese data centres Consistent with the recent definition
FY2028 Outlook from approximately 70 per cent to approximately 75% YoY —
Vera Rubin First-quarter shipments totalled nearly $20 billion, the fastest climb in history —
Buyback New licences granted in September $150 billion; $28.1 billion returned to shareholders in Q3; $227.3 billion remaining —

Conclusion: Both statements are true. Data centre revenue of $107.6 billion significantly exceeded the consensus estimate of $101.87 billion; non-GAAP gross margin of 74.9 per cent was above the 74 per cent forecast. The pullback conditions we had set have been triggered.

A cause for concern: Q4 gross profit margins will take a hit

74.9 per cent is this season’s result. 71–72 per cent is the forecast for next season.

Management described the 71–72 per cent figure as ‘the bottom’ and stated that it would rebound to 72–73 per cent in FY2028. In other words, this is not a permanent decline, but ratherVera Rubin: The Phased Costs Arising from Product Structure and Costs During the Ramp-up Phase。

But we must do the maths properly:Revenue guidance has jumped from $115.8 billion to $130.0 billion (+12 per cent), whilst the gross margin has fallen from 74.9 per cent to 71–72 per cent (approximately -3 percentage points). This is a classic case of ‘sacrificing profit margins for growth’.

NVIDIA: Comparison of Q3 Actual Results with Q4 Guidance and Breakdown of Financial Commitments
NVIDIA: Comparison of Q3 Actual Results with Q4 Guidance and Breakdown of Financial Commitments

Figure 3On the left is a comparison of NVIDIA’s Q3 actual results and Q4 guidance (revenue: $115.844 billion → guidance: $130 billion; non-GAAP gross margin: 74.9% → guidance: 71–72%); On the right is the breakdown of the approximately $522 billion in financial commitments (purchase commitments of $390 billion, data centre leasing and cloud service commitments of $100 billion, and a financial guarantee for Neocloud of $32 billion); At the bottom are key points from Kress’s conference call regarding the increase in GPU rental fees, along with a clarification that “$2.5 trillion represents the backlog in the cloud industry, not orders specific to NVIDIA itself”.

This means, ‘If we hold out this season, we’ll have to pay the toll next season.’

Summarise the nature of this financial report in one sentence:It demonstrated the demand, whilst foreshadowing the cost.

It is true that four figures exceeded expectations, and it is also true that the Q4 guidance was higher than the consensus. However, the downward revision to the gross margin guidance indicates that:This round of growth does not come without a cost; it requires sacrificing profit margins on the product mix. Management says this is the bottom and that things will pick up next year—This is a promise that needs to be verified; it is not a fact that has already occurred.

The strongest piece of evidence is not the target price, but the rent

During the earnings call, Kress made a statement that carried more weight than any analyst’s target price (translation):

“Under multi-year agreements, Hopper GPUs have been re-contracted at prices of up to twice their original cost; rates for the B300 have risen by more than 60 per cent since its launch.”

This is evidence of price, not evidence of expectations. This does not prove that ‘people still want to buy’, but rather that ‘existing contracts are not only being renewed, but are being renewed at higher prices’.

This also explains why the assertion that ‘demand is alive’ holds water:If demand is artificial, rents cannot possibly double.

That $2.5 trillion isn’t Nvidia’s.

Another striking figure emerged during the conference call:$2.5 trillion…three months ago, the figure stood at $2 trillion. This represents the entire ‘cloud industry backlog’, not Nvidia’s own order book. Omitting the phrase ‘cloud industry’ is tantamount to attributing the livelihood of an entire sector to a single company.

Circular Financing: From Qualitative to Quantitative

During the conference call, Kress disclosed three measures of the company’s financial commitments:

Project Amount
Procurement Commitment $390 billion($390B)
Data Centre Leasing + Cloud Services Commitment Over $100 billion($100B)
neocloud Financial Guarantee $32 billion($32B)
Total Approximately $522 billion($522B)

The key lies in the footnote on the last line: that $32 billion is explicitly ‘contingent on these companies being able to raise funds themselves’.

This statement transforms the label ‘revolving finance’ from a qualitative description into a quantifiable risk exposure.It and CoreWeave’s CDS spread are two ends of the same chain—The quality of some of Nvidia’s trade receivables depends on its customers’ ability to secure financing. Kress also stated that the company is “taking a very close look” at which companies to finance.

The board has put its foot down

The exact words of Harvey Jones, Chairman of the Board, during the conference call:

“NVIDIA’s board of directors unanimously accepted his resignation and has no intention of reinstating him as CEO.”

The title of Part Five states that ‘there is no substitute for a political guarantor’, a statement that has now been confirmed by the Board itself.

The company has also established a new“National Security and Export Control Oversight Committee”(comprising independent directors, external advisers and legal advisers), the executive search firm Heidrick & Struggles was engaged to find a successor. The system is being strengthened, but it addresses processes rather than the individual.

The market did not respond favourably

On the day following the release of its financial results (18 November), Nvidia’s share price fell 2.89%, with a cumulative fall of over 13%. On the same day AMD and Broadcom rise by more than 3 per cent. Extending the timeframe to the week in which Jensen Huang resigned,AMD and Broadcom both rose by more than 18 per cent, Micron +13 per cent, Qualcomm and Intel around +5 per cent.

NVIDIA’s crisis has turned into a boon for its rivals. The money hasn’t left the AI sector; it has simply shifted from one company to several others.

Why I’m bearish on Nvidia: it’s banking on someone who’s already left

This section also reflects my personal views. Although all four key financial metrics exceeded expectations, I remain bearish on Nvidia; the reason lies not in the financial results themselves, but beyond them:Over the past few years, many of NVIDIA’s relationships with suppliers and customers have been built on the personal credibility of Jensen Huang and his business philosophy.

First, some background.

And TSMC: in thirty years, not a single contract has been signed. In March this year, Jensen Huang said on Lex Fridman’s podcast that NVIDIA and TSMC had been working together for around thirty years, with transactions amounting to tens of billions and hundreds of billions of US dollars, yet “we’ve never signed a single contract”. He simply tells the other party what he wants, and they promise to deliver—it’s as simple as that. In 2013, Morris Chang even invited him to become TSMC’s CEO, but he politely declined. This relationship was not forged through contracts between the two companies, but through decades of personal rapport between the two men.

And the Taiwanese supply chain: a series of ‘trillion-yuan dinners’. When Huang Renxun returns to Taiwan, he often treats the heads of suppliers such as TSMC, Wistron, Quanta and Foxconn to dinner; the combined market capitalisation of the companies present amounts to over one trillion US dollars, leading the Taiwanese media to simply refer to these gatherings as the ‘trillion-dollar dinners’. In May this year, he stated that NVIDIA would invest approximately US$150 billion. These dinners transformed the partnership from a matter of contractual terms into one based on personal goodwill and mutual understanding.

His business approach: place the order first, then identify the demand. Jensen Huang’s approach has always been to “secure production capacity in advance, lock in long-term contracts and dare to make forward-looking investments”. The reason suppliers are willing to follow him in expanding production is that they trust his judgement and believe he will cover any shortfalls. Following reports of his potential departure, an executive from a Taiwanese partner company told the media: “We will continue to support the production capacity required by NVIDIA, but the new CEO may not be as aggressive as Jensen Huang.” TSMC, Foxconn and Wistron have already begun internal assessments of the potential impact on their production schedules.

To customers: NVIDIA is personally guaranteeing the purchase for buyers. In recent years, Nvidia has not only been selling chips, but has also been providing a safety net for Neocloud (an emerging computing power cloud) – a buyer of its chips: from July onwards, it has pledged to lease back their idle GPU computing power at a fixed price, thereby providing banks with a floor price for lending; it has joined a $500 billion financing platform and may underwrite up to $125 billion of it; During the conference call, Kress himself disclosed that the guarantees for Neocloud alone amounted to US$32 billion. The reason these arrangements have won the banks’ approval lies in the market’s trust in NVIDIA and in Jensen Huang’s judgement.

On the day Huang Renxun resigned, Joe Kaiser, founder of Switchyard Partners, made a rather stinging remark to *Fortune*:“If investors now feel that NVIDIA’s guarantee is no longer worth what it was a week ago, these Neoclouds are in trouble.”On that day, CoreWeave fell by around 9 per cent, whilst Nebius fell by 16 per cent.

When you put these points together, the bearish logic becomes clear:

  1. Trust cannot be handed over. A contract can be signed by a successor, but thirty years of tacit understanding and promises made over dinner cannot. Kress is an outstanding CFO, but the primary duty of a CFO is to manage risk, whereas Huang Renxun’s value lies precisely in his willingness to take risks. If the new leadership adopts even a slightly more conservative approach, suppliers will follow suit and slow down their expansion plans.
  2. The cost of providing a safety net will rise. These guarantees are conditional; they only come into effect once the other party has secured funding and the computing power has been commissioned and accepted. Previously, banks were willing to extend loans because they trusted the individual behind NVIDIA. Now that this trust has been eroded, Neocloud’s financing costs will rise, ultimately putting pressure on NVIDIA’s own orders.
  3. The gross profit margin is already sending a signal. Next quarter, the gross profit margin is set to fall from 74.9 per cent to 71–72 per cent; management attributes this to the temporary costs associated with Rubin’s ramp-up phase. That may well be the case, but at a stage when the company needs to regain the trust of suppliers and investors, there will be more, not fewer, areas where it will have to concede on margins.
  4. Our competitors have been waiting for this opportunity for a long time. AMD and Broadcom have risen by 18 per cent in a week, indicating that the market has begun to bet on a ‘post-NVIDIA’ landscape.

Demand is not dead, nor will Nvidia collapse. However, I believe the premium associated with its past ability to ‘single-handedly influence an entire industry chain’ is fading, and the share price has not yet fully reflected this. This is my assessment; it is not investment advice.


III. Considering the two criteria together

Checkpoints The conditions we have set In fact determine
Anthropic Broke through $70 on the first day → Breakout and upward revision $83 offer price, $97.17 closing price on the first day It has not fallen below the issue price; demand is confirmed.
NVIDIA Data centres meet targets + gross profit margin held steady at 74 per cent $107.6 billion (exceeding consensus), 74.9 per cent (exceeding guidance) Both conditions are met

I’ve passed both questions. But if you put the two ‘discomforts’ side by side, you’ll see the same thing:

  • What is unsettling about Anthropic isValuation: Demand is strong, but this comes at the cost of lower pricing and a larger print run.
  • NVIDIA’s unease stems fromGross profit: Growth is strong, but it comes at the cost of a 3-point drop in the gross profit margin next quarter.

In other words: the evidence on the demand side has grown stronger, and the bills on the cost side have started to arrive. This is not a contradiction; it is two sides of the same coin—AI capital expenditure is shifting from ‘expansion at any cost’ to ‘cost-conscious expansion’。


IV. Scenario Iteration: 35/45/20 → 42/40/18

Scenario Part One This article On the basis of
Continuation: CAPEX continues; valuation recovery 35% 42% NVIDIA beats expectations on four fronts; Q4 revenue guidance exceeds consensus; Anthropic raises its pricing, expands its offering and gains 17 per cent on its first day of trading; GPU rental renewals double
slow down: Growth has slowed but has not collapsed 45% 40% Q4 gross margin guidance down by 3 percentage points; Nvidia falls by over 13 per cent in a week; CEO vacancy remains unresolved; export controls shift from a policy issue to an enforcement issue
Rupture: a precipitous fall in capital expenditure 20% 18% Evidence on the demand side has strengthened, so the forecast has been revised downwards; however, three unresolved issues are preventing it from returning to 15 per cent.

Why has the default rate only fallen to 18 per cent, rather than returning to the earlier level of 15 per cent?

BecauseStrong demand does not necessarily mean low risk, there are three matters still pending:

  1. There is still no list of successors. The board says it is considering candidates from both inside and outside the organisation, but to date there have been no reliable reports naming any specific candidates. The longer the vacancy remains, the more pressing the question becomes: ‘Who is responsible for providing guidance?’
  2. The scope of the DOJ is unknown. The Ministry of Justice and the Ministry of Commerce confirmed this week that Megaspeed is under investigation on suspicion of smuggling chips into China.Criminal investigation. The investigation has only just begun, and it remains unclear which transactions are involved or whether it will lead to the discovery of internal NVIDIA documents.
  3. Q4: The decline in gross profit margin is structural. Management says this is the trough of Rubin’s ramp-up phase and that it will rebound in FY2028—this is a promise, not a fact. If the figure actually falls below 71 per cent, the situation will change.

Compared with 35/45/20, the net changes in this iteration are ‘continuation +7, slowdown -5, rupture -2’.The demand side has been given full marks, but the risk side has not been deducted a single point.

Trigger conditions (if any one of these is met, the process must be triggered again)

Direction Trigger conditions
“Continuation”increase December hearing concludes peacefully + ANT holds steady above $90 + No rate rise at December FOMC meeting
“Slowdown” or “Breakdown”increase DOJ investigation uncovers internal Nvidia documents; Congress pushes for a freeze on export licences; Q4 actual gross margin falls below 71 per cent

V. Background: The political track is advancing in parallel

This section is included merely as background, as it merits a separate article.

Items Status
Closed-door hearings by three House committees 12-01, Huang Renxun has agreed to attend
Senate Banking Committee hearing Scott said that both Huang and Lutnick had agreed to attend, and hoped this would take place ‘before Christmas’ (originally scheduled for 3 December for Huang and 10 December for Lutnick).
Megaspeed: Criminal Investigation The Ministry of Justice and the Ministry of Commerce confirmed this week that they are conducting an investigation(Alleged smuggling of chips into China)
Two investigative reports in the NYT 18 November: Chips flowing to Alibaba and the Chinese military are within easy reach; 19 November: Nvidia has been asked to sell chips to companies with links to the military

Jensen Huang’s status has changed from that of a ‘policy adviser’ to that of a witness required to give evidence under oath; and NVIDIA’s business in China has also shifted fromBusiness issuesbecameLaw enforcement issues. Should the DOJ’s criminal investigation focus on specific transactions, Nvidia would be required to update its compliance risk disclosures in its 10-Q filing; **this represents a tail risk that was not priced in at all in the 11–17 financial results. **It is the company Megaspeed that is under investigation; as for Alice Huang herself, the Department of Justice has yet to reach any conclusion.**

On the same day, Trump, Putin and Xi Jinping met in Shenzhen and issued a ‘Declaration of Peace’, stating that nuclear war ‘must not and will never be waged’, that oil production and supply should be increased, that UN reform should be promoted, and that the ICC ‘must not abuse its jurisdiction’. Among the points listed in the declaration, there was no mention of semiconductors.


VI. Preview of the Next Part

The hearings on 1 December and 3 December, together with the Lutnick line, will form the main subject of the next article.

The evidence is now complete: the Megaspeed investigation and why it was terminated; the Teter-Kessler phone call (presenting both the allegations and the defence side of the story); reports that Lutnick linked the easing of chip export restrictions to China with the ‘China deal’; and the divisions within the administration.

The hearing provided fresh ammunition—for the first time, statements were made under oath. This is also the only thread in this series that links ‘policy’ and ‘law enforcement’.


VII. Several Points of Interpretation

  • Anthropic’s $81.3 billion represents the total offering size (including existing shares), with the company raising approximately $79.4 billion net; the over-allotment option of 147 million shares has not yet been exercised, so the final figure may change.
  • The figure of $87.65 for ANT the following day is a snapshot from the market page, not the closing price.
  • NVIDIA’s financial commitments of approximately $522 billion were disclosed verbally by Kress during the conference call; the company has not published any details.
  • The fact that the offering was oversubscribed by more than six times, that retail investors accounted for approximately 15 per cent of the allocation, and that half of the institutional investors walked away empty-handed were all reported by the media, citing sources familiar with the matter.

Main Sources of Information and Verification Index


The probability of the scenarios described in this article, as well as my views on ANT’s first day of trading and on NVIDIA, are my personal judgements based on publicly available information as at 20 November 2026 and do not constitute investment advice. With regard to Alice Huang and Megaspeed, I have only referred to publicly available reports and official documents.

Cover image:Close-up Photo of Survey Spreadsheet — Photo by goumbik on Pexels (licensed for free commercial use)

  • Title: The answer to the two test cases: demand is still there, but the bills have arrived too
  • Author: AdenXie
  • Created at: 2026-11-21 00:30:00
  • Updated at: 2026-11-21 22:00:00
  • Link: https://blog.adenxie.com.cn/2026/11/21/2026-11-21-two-checkpoints-demand-alive-bill-arrives/
  • License: Unless otherwise noted, original text and images by AdenXie: CC BY-NC-SA 4.0; original code: GPL-3.0. Third-party material retains its original rights and licenses. Scope & exceptions.
Comments