Act 01 — A Saturday in Q1 2026What you notice first
Saturday, 9 a.m. Your setup. Half your agents are throwing errors. The one skill your hub depends on: timeout. A message to Claude won't go through. The OpenAI API has been dead for three hours.
9:15. Stock widget. Oracle down 38 percent. Nvidia down 24. Microsoft down 12. Amazon down 9. Alphabet down 11. Over half a trillion in market value gone overnight.
9:30. Email. Anthropic's price list doubled overnight, token limits cut in half. "In response to the global compute shortage." OpenAI follows on Monday.
8 p.m. The evening news. Between the crash footage, a segment from Oberkochen: Zeiss, whose optics sit in every modern chip fab, puts it in context. Order book still full for years out, no short-time work. The company can't crash on the exchange anyway, it isn't even listed. But out on the Swabian Jura the first people start asking how safe their jobs still are if the chip industry hits the brakes.
" was right. Again."
— the headline everywhere. The working hypothesis I started this research with. The next three acts test it[46][05].
Act 02 — The mechanicsHow the money circles between the three main players
Before we get back to the Saturday-morning scenario, look at what actually happened in the first quarter of 2026. Three main players. Three steps. The same pot of money.
- →Microsoft → OpenAI: $13B in equity plus Azure cloud backlog.
- →Oracle → OpenAI: $300B Stargate contract, no payments until 2027.
- →Amazon & Alphabet → Anthropic: over $60B invested or committed between them.[24][31]
- →Anthropic → Google: reportedly around $200B for cloud and TPUs from 2027 (per The Information; Reuters corroborated the broad strokes, not the figure) — a good 40% of Google's $462B backlog.[26][07]
- →Anthropic → SpaceX: $1.25B a month for the Colossus datacenters, up to $45B.[28]
- →Alphabet → SpaceX: 6.11% direct stake, roughly $90B on paper at the IPO.[29]
Goldman Sachs puts the circular share at under 15% of Nvidia's revenue through 2027[13]. The skeptics do the math differently.
Act 03 — On the booksWhat is really on the AI giants' balance sheets
The $67B in book profits from the loop is not a projection. Here it is, sitting in the Q1 2026 SEC filings — six companies, six balance sheets. On the left, what is booked; on the right, what actually turned into cash. All of it gains on AI stakes and contractually committed but not-yet-delivered services (). US-GAAP standard, all legal.
Take Alphabet. The company booked $28.7B in profit on its Anthropic stake in the first quarter — without a cent moving, and before its own money was even wired. The stake dates to 2023; the $380B valuation that marked it up was set in February by outside investors, and Alphabet wasn't even on the round's investor list.[23] Its own $10B investment didn't come until April, after the quarter closed.[24] Amazon plays the same game: $8B put in, worth a multiple of that on paper, while free cash flow collapsed 95 percent — the real money flows into datacenters, not into the till.
Act 04 — Historical comparisonsEnron or Cisco?
Look at $67B in book profits with no cash behind them, and the reflex is Enron. has been drawing a different comparison in public since late 2025, "there is a Cisco at the center of it all … Its name is Nvidia"[46], and deepened the argument in May 2026 in a (paywalled) Substack post[05]. After Nvidia countered with a seven-page sell-side memo, he sharpened it further[03]. First what he rules out, then what he actually means.
"I am not claiming Nvidia is Enron. It is clearly Cisco."
— Michael Burry, May 2026[03]
What he means: no accounting fraud, no hidden losses. Instead, the growth-on-its-own-credit pattern that Cisco ran 25 years earlier. Three steps, the same loop shape as today.
Cisco then vs. Nvidia now — bigger and harder to see
A reflex when you make the comparison: "Cisco was much smaller than Nvidia is today, so it's not a real comparison." True in absolute terms — but it misses the point. In March 2000, Cisco was the most valuable company in the world, roughly four percent of US GDP, about four percent of the S&P 500. Nvidia today is larger in absolute terms but also weighted more heavily in the index: around six percent of the S&P 500, about twelve percent of US GDP.[15] If something breaks this time, more market cap goes at once.
| Metric | Cisco 1999/2000 | Nvidia 2026 (reported) | Nvidia 2026 (ex loop entries) |
|---|---|---|---|
| 201[47] | 30 | ~45–53[△] | |
| Market cap (absolute) | $555B[48] | $3.4T | $3.4T |
| Share of S&P 500 | ~4% | ~6% | — |
| Share of US GDP | ~4% | ~12% | — |
| Loop volume | $2.4B vendor loans (open on balance sheet) | $40B+ equity stakes (in other income) | hidden in the loop |
| Q1 book profit from loop | — | $15.9B (ASC 321) | stripped from income statement |
The Cisco risk was obvious in 1999: 201, anyone could see that was too expensive. Nvidia's reported P/E of 30 looks cheap by comparison in 2026. Strip the ASC 321 book profits out of its own income statement, and the real ratio sits at about 45. And that's conservative: on the buyer side sit OpenAI, Anthropic and CoreWeave, with cash that comes in large part from equity stakes and funding rounds at the hyperscalers themselves. Strip out that prefinanced purchasing power too (Goldman Sachs estimates under 15 percent of Nvidia's revenue as circular through 2027[13]), and the price-to-earnings ratio lands closer to 53[△]. This time the risk is not only bigger, it is also harder to see.
"Money that isn't really there finances buyers who couldn't really pay, from a seller who books the profit off it."
— Max Fraunhofer
Interlude — And Germany?Not a purely American story
So far this has sounded like an American story. US companies, US balance sheets, US exchanges. It isn't. Germany is in it too. Just not where you'd look first. No German company is a contracting party in any of these loop deals. Yet Germany sits at both ends of the same chain: where the chips come from, and where the savings sit.
The two companies without which the loop would have no chips
The whole fight is about Nvidia chips. They come from TSMC, and TSMC can only build them because one single machine can do it: ASML's . ASML, in turn, builds none of them without two German suppliers. The optics come from Carl Zeiss SMT in Oberkochen, the only maker in the world. The laser that generates the EUV light comes from TRUMPF in Ditzingen. No second source for that either. Two family-owned firms, neither listed, both at a point in the chain that knows no substitute.
And they ride the cycle harder than their size suggests. At Zeiss, the SMT semiconductor segment accounts for €5.1B of €11.9B in revenue[32], almost 43 percent of the group, and the most profitable part on top. Those optics go to ASML alone; ASML builds in Zeiss alone. At TRUMPF it's roughly 17 percent[33]. If the AI investment boom cools, the fabs order fewer EUV systems, and that shows up, with a lag, in Oberkochen and Ditzingen.
Behind them sits a second row that runs the opposite direction from Zeiss. Siltronic, out of Munich, makes the wafers: the bare 300mm silicon disks on which modern chips get built in the first place. Roughly 70 percent of revenue rides on exactly these large disks for leading-edge chips[43], so directly on the same cycle that drives the loop. Only Siltronic has no buffer like Zeiss: wafers are consumed continuously, not ordered years ahead like EUV systems. When the chip market turns, the orders are gone in the same quarter.
That four other wafer makers exist doesn't help; if anything it makes things worse. It only means Siltronic can't set the price. How real that is shows up in the numbers already: in 2025 the company slid into an operating loss and dampened its 2026 outlook[43]. At the Burghausen plant an older production line has already been shut down, roughly 400 jobs, though without forced layoffs[44]. For the people there the downturn is no longer a thought experiment.
Merck is the counterexample: it does supply the specialty chemicals for chip fabrication, but the bulk of its business is pharma and life science. A chip dip dents the balance sheet, it doesn't threaten it.
How much is really riding on it?
| Company | Character | Exposure to the chip/EUV cycle |
|---|---|---|
| ZEISS (SMT) | EUV optics, single-source worldwide, only to ASML | ~42.5% of group revenue |
| TRUMPF | EUV laser with no second source | ~17% (EUV unit) |
| Merck KGaA | broadly hedged across pharma & life science | ~12–13% (semiconductor chemicals) |
| Siltronic | fully in the cycle, no order-book buffer like Zeiss | 100% semiconductors, ~70% 300mm |
| Wacker | solar and silicones carry the group | only ~3–5% (semiconductor poly) |
The other exposure is more direct. Amazon, Microsoft, Google and Oracle, four of the companies from the loop, have announced over €25B in datacenters in Germany. AWS in Brandenburg[36], Microsoft in Frankfurt and the Rhineland[37], Google in Hesse[38], Oracle in the Rhine-Main area[39]. These are not annual figures but commitments over years, some out to 2040. If the boom cools, the first thing cut is what's been announced but not yet built. Then the sites, jobs and trade-tax revenue that were already on the books go missing.
And then there's the spot that hits almost everyone with a savings plan. Nvidia, at around six percent, is the largest single position in the MSCI World[40], the index most German ETF savings plans run on. The German ETF market sits at half a trillion euros, nearly half of it US equities. Through this one channel alone, roughly €15B of German nest eggs hang on Nvidia's price. BaFin warns in January 2026 explicitly about the concentration of a few US tech names — Nvidia, Apple and Microsoft together make up a good 20% of the S&P 500[41]; the Bundesbank flanks that with warnings about high valuations and sudden corrections[42].
The other side: a runway, not a cliff
There is a buffer, though. ASML's order book stands at €38.8B at the end of 2025, a good two-thirds of it EUV, and capacity is sold out through 2027[35]. That gives Zeiss and TRUMPF a year and a half to two years of visibility on the books. And leading-edge chips don't go to AI alone: smartphones, memory and high-performance computing pull in parallel. An AI dip slows the cycle; it won't kill it. For Zeiss and TRUMPF that means repricing rather than crash — one floor below the big bang.
That's the uncomfortable part for a German reader. It hits the portfolio, it hits the tools, and it hits the industry two states over that rides the same bet. Except no one there shouts "Burry," because hardly anyone sees the connection.
Act 05 — ClosingThree possible outcomes
"Burry was right."
"Everything goes fine."
AI demand is real, and the larger part of the numbers stands on real ground. Goldman Sachs estimates the circular share of Nvidia's revenue at under 15 percent[13]; over 85 percent comes from paying customers outside the loop. Oracle's order book grew 325 percent in the quarter[09] — contractually bound demand, not a book profit. And even at the loop companies the circle is only one part: of Alphabet's $62.6B quarterly profit, $33.9B is earned operationally, not circularly. If OpenAI closes its $122B funding round and the hyperscaler investments pay off over the contract terms, then the circular book profits are a prepayment on real growth. The loop resolves cleanly, nobody loses. Zeiss and TRUMPF barely feel it anyway: order book full, demand real.
"Burry was right."
Just one link in the chain wobbles. OpenAI misses its funding target, Anthropic's valuation gets halved, Oracle's RPO is partly revised. The $67B in book profits have to be written off, stakes get remarked, the backlog shrinks. A balance-sheet cascade like Cisco 2001. Stocks fall, token prices rise, tools go down. How thin the cover can be shows in the SpaceX prospectus (SEC, May 2026): xAI ran a $6.36B operating loss on $3.2B in revenue in 2025, and $2.47B in the first quarter of 2026 alone.[30] If Anthropic's compute purchasing falls away, that hits SpaceX and xAI directly. has been wrong more often than right since 2008[12]. But the pattern he describes has happened before: Cisco 1999. And in Germany? If orders fall away over years, the dip reaches Oberkochen and Ditzingen too, with a lag. Like 2009, when Zeiss SMT already went on short-time work once[45].
Outcome C · the middle path (the most likely). A mix of both. The really big crash stays away, because the hyperscalers generate operating cash flow and can throttle CapEx if needed. But two or three players in the chain take a hit: an AI lab has to raise funding on worse terms, an RPO backlog gets trimmed, a hyperscaler stock loses 30 to 40 percent. Token prices stabilize or rise slightly, a handful of tools disappear. Nobody talks about a "crash," everyone about a "repricing." In Germany, Siltronic gets hit first, with no order-book buffer and already in the red[43]; Zeiss and TRUMPF only once the books get thinner. More likely than mass layoffs: dampened forecasts, short-time work if it comes to that.
The Saturday morning with the errors and the headline is hypothetical. But it could happen. Cisco investors lived it on March 27, 2001. Back then all they lost was money. In 2026 it would also take away the tools they'd otherwise use to rethink their strategy — Claude, ChatGPT, the AI agents that sit in the workflow. The double hit: your portfolio down, and the tool you'd reach for to rethink it gone with it.*
Disclaimer · Not investment advice · Disclosure
* This site is operated by Max Fraunhofer as a private journalist and holds no BaFin authorization under § 32 KWG. The article is journalistic analysis, not investment advice within the meaning of § 1 Abs. 1a S. 2 Nr. 1a KWG / § 2 Abs. 2 Nr. 4 WpIG, and not an investment recommendation with intent to sell within the meaning of § 85 WpHG. All figures come from public SEC filings, earnings-call transcripts, analyst reports and media coverage (see the source list). Past performance is no indicator of future results. Anyone who makes investment decisions based on this article does so at their own risk and should consult a licensed financial or tax advisor.
Disclosure (conflict of interest): At the time of publication, the author holds direct equity positions in, among others, Alphabet, Amazon, Microsoft, Meta, ASML and TSMC, plus broadly diversified index funds (including the MSCI World) that carry indirect Nvidia exposure. There are no direct positions in Nvidia, Oracle, Siltronic or CoreWeave; there are no short positions in any of the names mentioned.
SourcesEvidence, filings, track record
Convention in the article: hard numbers, quotes and external model outputs are marked in the body text with a superscript source number. A click jumps straight to the entry. Three confidence levels, visually distinguished:
- 01CNBC · OpenAI closes $122B round at $852B valuation (Mar 31, 2026) · Nvidia with a $30B equity stake.
- 02CNBC · Nvidia tops $40B in AI equity investments (May 9, 2026) · Aggregate of all Nvidia stakes 2026.
- 03Stocktwits · "Michael Burry vs. Nvidia heats up: Big Short investor stands by his analysis" (May 2026) · Burry's response to Nvidia's 7-page sell-side memo, incl. "I am not claiming Nvidia is Enron. It is clearly Cisco."
- 05Michael Burry · "The Heretic's Guide to AI's Stars Part III: Tracepalooza & the Bezzle" (May 22, 2026) · Cisco reference and bullwhip argument. [Paywall — only the intro is freely readable; the verbatim Cisco quote is freely accessible in source 46.]
- 06Microsoft 10-Q Q3 FY2026 (SEC EDGAR, April 2026) · RPO $627B, +99% YoY. OpenAI share derived from earnings-call commentary.
- 07Alphabet 10-Q Q1 2026 (SEC EDGAR, April 2026) · Google Cloud backlog $462.3B.
- 08Amazon 10-Q Q1 2026 (SEC EDGAR, April 2026) · AWS performance obligations $364B, weighted remaining term 5.5 years.
- 09Oracle Q3 FY2026 IR press release (Mar 10, 2026) · RPO $553B, +325% YoY. $300B OpenAI contract.
- 10CoreWeave 8-K SEC (Sep 9, 2025) · Backstop agreement with Nvidia, $6.3B through 2032. Primary source for the CoreWeave mechanics.
- 11CoreWeave 10-Q Q1 2026 (SEC EDGAR, May 2026) · Backlog $99.4B, OpenAI share $22.4B.
- 12Elite Currensea · Burry track-record analysis (2024) · 2023 sell tweet, SPY/QQQ puts with a 40 to 50 percent loss.
- 13Proactiveinvestors · Goldman Sachs on Nvidia's circular revenue (Oct 7, 2025) · Goldman analyst James Schneider: "circular deals represent less than 15% of revenue in 2027" — under 15 percent of Nvidia's revenue through 2027.
- 15TechnoStatecraft · Cisco vs. Nvidia — dot-com comparison (2025) · Cisco P/E 201, $555B market cap, 88% fall, ATH March 2000.
- 21American Affairs Journal · "Who Lost Lucent?" (August 2020) · Vendor-financing volume 1999–2001, SEC restatement detail.
- 22Howard Marks / Oaktree Capital · "Is It a Bubble?" memo (December 2025) · Top-4 hyperscalers $451B operating cash flow 2024, "Some AI revenue is currently circular in nature."
- 23Anthropic · Official Series G press release (Feb 12, 2026) · $30B in fresh capital, $380B post-money. Co-leads GIC, Coatue; full investor list without Alphabet/Google. Confirmed by TechCrunch Feb 12, 2026.
- 24CNBC · Google announces a standalone $10B investment in Anthropic (Apr 24, 2026) · Separate deal in Q2, up to $40B against performance targets. Confirmed by Bloomberg Apr 24, 2026.
- 26The Information / US News · Anthropic commits to $200B in Google cloud spending (May 5, 2026) · 5-year commitment from 2027, TPUs + cloud services. Reuters corroborated the broad strokes but could not independently verify the figure; no official confirmation from Anthropic or Google.
- 27Anthropic · Official Series H press release (May 28, 2026) · $65B in fresh capital, $965B post-money. Leads Altimeter, Dragoneer, Greenoaks, Sequoia. $15B "previously committed hyperscalers," $5B of it Amazon. Google only as a compute partner (5 GW TPU, Broadcom). Confirmed by Bloomberg, CNBC, TechCrunch May 28/29, 2026.
- 28SpaceX · Form S-1 (SEC EDGAR, May 20, 2026) · Cloud Services Agreements with Anthropic PBC: $1.25B/month for Colossus, nominally through May 2029 (90-day notice period), up to $45B. SpaceX took over xAI and thus Colossus on Feb 2, 2026. Context via TechCrunch May 20, 2026.
- 29Bloomberg · Alphabet's SpaceX stake, ~$100B on paper (Apr 15, 2026) · Primary source SpaceX Alaska Biennial Report (Apr 11, 2026): Google LLC 6.11%. Diluted to ~5% after the xAI merger. IPO target at least $1.8T (Bloomberg May 29, 2026, disputed by Musk).
- 30SpaceX · Form S-1 (SEC EDGAR, May 20, 2026) — xAI financials · xAI FY2025: operating loss of $6.36B on $3.2B revenue; Q1 2026 loss $2.47B; 1.9M paying SuperGrok subscriptions out of 117M Grok MAU. Context via TechCrunch May 20, 2026.
- 31Anthropic · Expanded Amazon partnership (Apr 20, 2026) · Amazon up to $25B in additional equity investment (on top of the $8B stake) + a compute commitment of over $100B (AWS Trainium) through 2036. Confirmed by Amazon post Apr 20, 2026.
- 32Carl Zeiss AG · Annual report fiscal year 2024/25 · Group revenue €11.9B, SMT segment €5.06B (~42.5%), the strongest and most profitable segment. EUV optics exclusively to ASML.
- 33TRUMPF · Fiscal-year figures 2024/25 · Group revenue €4.33B, EUV unit €724M (~16.7%). Sole supplier of the high-power CO₂ lasers for ASML's EUV light source.
- 34Merck KGaA · Annual report 2025, Electronics business · Electronics 17% of group revenue (€3.59B), of which Semiconductor Solutions is ~12–13% of group revenue.
- 35ASML · FY2025 results (SEC Form 6-K) · Order book €38.8B, ~65% of it EUV. EUV capacity sold out through 2027.
- 36AWS · €7.8B European Sovereign Cloud Brandenburg (through 2040) · plus €8.8B for the existing Frankfurt region (2024–2026).
- 37Microsoft · €3.2B AI infrastructure Germany (2024–2026) · Frankfurt region plus a new datacenter in North Rhine-Westphalia.
- 38Google · €5.5B Germany (through ~2029) · Datacenters in Hanau and Dietzenbach, Berlin-Brandenburg cloud region.
- 39Oracle · ~$2B cloud/AI infrastructure Frankfurt/Rhine-Main (5 years, July 2025).
- 40finanzen.net · Concentration risk MSCI World 2026 · Nvidia ~6% largest single position, the "Magnificent Seven" together over 21% of the index.
- 41BaFin · Risks in Focus 2026 (Jan 28, 2026) · "Nvidia, Apple and Microsoft alone, for example, make up a good 20 percent of the market capitalization of the S&P 500" — a warning about concentration risk from a few tech names.
- 42Deutsche Bundesbank · Financial Stability Review 2025 (November 2025) · A warning about high valuations and sudden corrections on disappointing AI announcements.
- 43Siltronic · Fiscal year 2025 · ~70% of revenue from 300mm wafers (the substrate for modern logic/memory chips). First operating loss in 2025 (EBIT −€26M, net loss €69M), no dividend, dampened 2026 outlook. One of five large wafer makers worldwide (Shin-Etsu, SUMCO, GlobalWafers, Siltronic, SK Siltron), hence no pricing power in a downturn.
- 44Siltronic · End of small-diameter wafer production, Burghausen plant (2025) · roughly 400 jobs affected, capacity adjustment in the downturn, without forced layoffs (partial retirement/demographics).
- 45photoscala · "Carl Zeiss AG: a cautious start to the new year" (Mar 11, 2009) · Short-time work for nearly 800 employees in semiconductor technology (SMT) and metrology in Oberkochen during the 2009 semiconductor downturn — a historical precedent.
- 46Fortune · "Big Short investor Michael Burry on an AI bubble" (Nov 24, 2025) · freely readable evidence for Burry's Cisco comparison: "And once again there is a Cisco at the center of it all … Its name is Nvidia."
- 47Liberty Through Wealth · "The Cautionary Tale of Cisco Systems" (Mar 22, 2022) · Cisco's peak P/E in March 2000 at 201 ("its price-to-earnings ratio stood at an astonishing 201"). Secondary source; Macrotrends only lists P/E from 2001 on.
- 48GuruFocus · "Cisco (CSCO) Reaches New Highs, Surpasses Dot-Com Bubble Peak" (December 2025) · Cisco's peak market cap in March 2000 around $555B; reclaimed the all-time high on Dec 10, 2025.