I'm Eugenia, an investor in 100+ startups and a board director, formerly at Expedia and Meta. This newsletter is my personal view, it is not sponsored by anyone (even though I wouldn’t mind maybe 😉) and does not provide financial advice. Enjoy and let me know what you think!
EDITION Nº 09 · GROW · 11 MIN READ
⏩ Not in the mood for analysis? Skip straight to the to-do part: Path A and Path B: how to earn from Anthropic’s IPO
I use Claude every working day. Full disclosure - some parts of this edition were even drafted with it✌️.
And - despite every news outlet exploding this week on the topic of Anthropic’s IPO - I still don't know whether I'll buy a single share when it lists. Because - what’s in it for us?
On Monday, Anthropic made its S-1 public.
What is S-1? That's the 300-plus-page document a company must publish before its shares can trade. Ticker ANT, on the Nasdaq Stock Exchange (Anthropic's announcement).
When? No date is set yet. Reports point to the second half of November: first trading on the 19–20 November or in Thanksgiving week (23–27 November). So, very soon ⏰.
At what price? Officially, nobody knows: the price range in the filing is still blank. Unofficially, the number is drifting down - from the ~$2 trillion investors were whispering about in the summer to around $1.5 trillion now. Three reasons, and none of them is "Claude got worse".
Growth is slowing - a bit. Revenue still grew 45% in the third quarter, after more than doubling the quarter before (Yahoo Finance). When you are priced on the hope of doubling forever, "only" +45% reads like a warning.
The whole AI trade has the jitters. The Nasdaq-100 is about 6% below its early-October peak. Investors are asking out loud whether all this AI spending will ever pay back, and some of the debt-funded data-centre names are wobbling (OnInvest, The Daily Upside).
Everyone wants the same money at the same time. On Monday, Meta announced a $95 billion share sale to pay for its own AI build-out - in the very weeks Anthropic hopes to raise up to $100 billion from the same big funds. More shares for sale, the same buyers: of course, prices soften.
So what? A lower IPO price is not bad news for us. It's the difference between paying for a great company and overpaying for a great story.
I've sat on the private side of more than a hundred rounds. So the uncomfortable part first: the cheapest tickets to Anthropic were sold years ago, to people who never had to open a broker app. Amazon AMZN started putting money in in 2023. By March this year, its own books valued that position at about $74 billion.
Nobody sent us that WhatsApp 📲. I checked twice.
So in this edition of Your Money, Your AI:
How do regular European women with a salary, a broker app and an ETF or two - can actually earn from this IPO?
The belief to drop: the IPO is the opportunity
First and foremost, the IPO is the most expensive seat at this table.
Look at the private ladder. In March 2025, Anthropic raised at $61.5 billion. In May 2026, at $965 billion. That is roughly 15x in fourteen months, and none of it was open to general public. At the reported $1.4-2 trillion, a day-one buyer pays 1.5 to 2 times the last private price.
So what? By the time a company like this reaches the stock exchange, the people who took the early risk have already made most of the money. On day one you are not getting in early. You are buying from them.
The 15x climb happened before the IPO Anthropic's valuation by round. The private rounds (plum) were not open to us; the IPO range (gold) is reported, not set. | |||||||||||||||||||||||||
| |||||||||||||||||||||||||
| SOURCE: ANTHROPIC ROUND ANNOUNCEMENTS · IPO RANGE: BLOOMBERG, WSJ, 26–27 OCT 2026 · US |
And the price isn't even the scariest number in the filing. The bills 🧾 are.
For most of its short life, Anthropic spent more on computing power than it earned. In 2025, every $1 of revenue cost about $1.60 in compute alone. This spring, that flipped. In Q2 2026 compute fell to under half of revenue, and Anthropic made its first operating profit - almost $1 billion in a single quarter (Fortune has the breakdown). The third quarter then brought $16.8 billion of revenue, up another 45%.
The first profit: Q2 2026 Operating income per quarter, $ billions. Losses below the line. | ||||||||||||||||||||||||
| ||||||||||||||||||||||||
| SOURCE: ANTHROPIC S-1 · UNAUDITED QUARTERS |
4 in 5 dollars are pay-per-use Share of revenue billed by usage, not by seat. | ||||||||||||||||||||
■ pay-per-use ■ subscriptions | ||||||||||||||||||||
| SOURCE: ANTHROPIC S-1, NOTE 2 · H1 = JAN–JUN |
How to read the two charts above. On the left, each bar is three months of Anthropic's business. Below the line, it lost money. Above the line, it made money. Five quarters of losses, then the first profit in Q2 2026 - not because it cut costs, but because sales grew faster than the bill for computing power.
On the right is how customers pay. Plum is pay-per-use, like an electricity meter: the more work Claude does, the bigger the bill. Beige is a flat subscription, like Netflix. Four out of every five dollars come through the meter. And customers keep turning it up: the S-1 says existing customers now spend roughly nine times what they did a year ago.
So what? The meter is why profit showed up so fast: when customers use more, revenue rises the same month. It works in reverse too. If companies tighten their budgets, they can turn the tap down tomorrow - and that is part of what you'd be buying with ANT.
What Microsoft's news this week has to do with Anthropic
On Monday, Microsoft offered voluntary exit packages to about 60,000 people - roughly 25% of its workforce - while it keeps pouring $80 billion into AI infrastructure. Satya Nadella has tied this year's cuts to restructuring the company around AI. Tellingly, data-centre staff were not offered the package (Tech Times). Two charts explain why.
What each employee brings in Revenue per employee, last 12 months. | |||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||
| SOURCE: ANTHROPIC S-1, YAHOO FINANCE, VIA MORNING BREW · LATEST FULL-TIME HEADCOUNT |
What investors pay per employee Market value divided by staff. | ||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||
| ANTHROPIC AT A $2T VALUATION, ~5,300 STAFF (AT $1.5T: ~$280M) · OTHERS AS OF 23 OCT 2026 · VIA MORNING BREW |
Left chart: what each employee brings in. Every person at Anthropic brings in about $6.6 million of revenue a year. At Microsoft it's $1.5 million. Anthropic does it with roughly 5,300 people; Microsoft has more than 220,000.
Right chart: what investors pay per employee. At the $2 trillion valuation that was being whispered about, investors would be paying $377 million for every person at Anthropic. For each Microsoft employee, they pay $18 million. That gap is the market saying out loud: it believes the future belongs to companies where a few people and a lot of computing power do the work.
It's the same shift, seen from both sides: company budgets are moving from salaries to compute. Microsoft is cutting people and buying chips - some of which it rents to Anthropic. Investors reward exactly that move. And Anthropic already lives at the far end of both charts ☝️.
So what for us? As investors: $377 million per employee means the price already assumes Anthropic keeps this up for many years. A great company can still be an expensive share - which is why the IPO price matters more than the hype. As professionals: it's worth asking which part of your job is a "seat" that could be billed per use - and which part is judgement nobody can meter.
Now the bill. To keep up with demand, Anthropic has signed for about $268 billion of computing power it cannot cancel - more than 13 times what it earned in the last 12 months. Add every lease and option at its ceiling and you get the "$800 billion" in the headlines. Against that sits $131.7 billion of cash.
Picture a restaurant with a queue around the block that signs 15-year leases on ten new dining rooms. If the queue keeps growing, it's genius. If it doesn't, the rent is still due. So what? Anthropic has to keep growing fast just to stand still - and its landlords get paid either way.
Google, Amazon and Microsoft get paid "regardless of actual usage."
That is the most expensive phrase in any contract. Also my favourite one - depending on which side of the table I'm sitting 😉
In a gold rush, the people who reliably got rich sold shovels. Anthropic is the miner. The shovel sellers get paid first, whether the gold shows up or not.
Most of them are listed - meaning you can buy their shares. Some of them are already in your portfolio.
The S-1 in six lines
I read the prospectus so you don't have to. Here are my most important six things:
The growth is real. Revenue went $41 million (2023) → $386 million (2024) → $4.6 billion (2025) → $16.3 billion in the first half of 2026 alone. More than 11,900 companies pay it $100,000+ a year. So what: this isn't a demo with a hype cycle. Anthropic went full on on B2B, and thousands of businesses already pay real money for it every month.
The $80.7 billion loss is mostly paper. $72.4 billion of it is an accounting charge: loans that will turn into shares got revalued as the company got more valuable. On its own adjusted measure, the first half made $1.6 billion of operating profit. So what: don't let the "$80 billion loss" headline scare you. It is NOT cash going out of the door.
The revenue is pay-as-you-go. 80% of it is billed by usage, and many of the largest customers have no long-term contract. In June, US export controls forced Anthropic to suspend its two newest models for almost three weeks - and the compute bills kept running. So what: revenue can grow very fast and shrink just as fast. Customers can turn the tap down tomorrow, the compute contracts can't be.
Curcular relationships: the suppliers are also owners, customers and rivals. Amazon and Google own stakes, resell Claude and build competing models. SpaceX
SPCX, which now owns xAI, has a compute deal worth up to $120.5 billion that either side can walk away from on 90 days' notice from early 2027. So what: relationships are excellent today, but Anthropic's landlords also want its tenants.You buy the economics, not the steering wheel. One Class F share, held by the seven co-founders, carries 50.1% of the vote, and a Long-Term Benefit Trust elects most of the board. In the filing's own words, ordinary shareholders "generally will not be able to affect" outcomes. The FT looked at the founders' grip today.
My view: it's a good move. Read the 92 pages of risks - misuse, models behaving in ways nobody planned, regulators switching models off. Keeping control is the founders' hedge against exactly those risks: no activist fund can push them to ship faster than they think is safe just to hit a quarterly number. The flip side: anyone can't vote them out either. As a buyer, you are betting on these seven people's judgement. You better trust them.
Dilution is pre-agreed. 5% of the company goes to a Public Benefit Fund, and the board may add up to another 2% each time the share price rises 50% above the last milestone. So what: if the shares do very well, a small slice of each big rise goes to the mission, not to shareholders. It's written down in advance, which I prefer to surprises.
The risk factors run to 92 pages, about three times SpaceX's. They include a warning that advanced AI could "pose catastrophic or existential risks to humanity". The business is real. The price is the open question, and so is how little say a public shareholder gets.
Who gets paid before Anthropic does
Every euro a company pays for Claude passes through six groups of suppliers before any of it becomes Anthropic's profit. The further from the centre you go, the more each company depends on a handful of tenants and on borrowed money.
Who gets paid before Anthropic does Six groups that earn from every Claude dollar. Company names with tickers in plum. | ||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||
| * NSCALE HAS FILED TO LIST AS NSCL. SOURCE: ANTHROPIC S-1, COMPANY NEWS · NOT A SHOPPING LIST | ||||||||||||||||||||||||||||
How the money reaches each group, and what to watch:
Clouds and strategic investors — Amazon
AMZN, AlphabetGOOGL, MicrosoftMSFT. They rent out compute, resell Claude through their marketplaces and hold equity. Watch: they are also Anthropic's competitors, and already inside most world and US index ETFs.Chips — Nvidia
NVDA, BroadcomAVGO(builds Google's TPU chips), AMDAMD. They sell or lease the accelerators Claude runs on. Watch: AMD also agreed to buy up to $5B of Anthropic stock. The money goes round.Neoclouds — CoreWeave
CRWV, Nscale (filed to list asNSCL), Lambda and Fluidstack (private). They rent out GPU capacity. Watch: debt-funded, with few big tenants. Two European names: Nscale is British, and Fluidstack was founded in Oxford.Ex-bitcoin-miner data centres — TeraWulf
WULF, Hut 8HUT, Cipher DigitalCIFR. They sign 15–20-year leases: ~$19B for TeraWulf in Kentucky, $7B (up to $17.7B) for Hut 8 in Louisiana. Watch: Google backstops part of the rent. One tenant can be most of the story.Network and extra capacity — Akamai
AKAM, SpaceXSPCX. Capacity deals; SpaceX gets ~$3.1B a month once all promised capacity is delivered. Watch: the SpaceX deal can be exited on 90 days' notice from early 2027.Energy and financing — power developers next to the campuses, lending banks and private credit such as Apollo
APO, whose data-centre developer is reportedly in talks on a 1 GW lease. Watch: Google guarantees part of Anthropic's lease and power payments.
Over half of the bill is tied to Google's chips Anthropic's compute commitments by partner, $ billions. Google-linked: $426B of $807B (53%) | |||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||
| * TPUs FROM GOOGLE OR OTHERS. MAXIMUM USED WHERE A RANGE WAS GIVEN; INCLUDES ESTIMATES. SOURCE: ANTHROPIC S-1, AS COMPILED BY THE FINANCIAL TIMES · US |
This is who is waiting at the end of the pipe. More than half of the bill is tied to Google's TPU chips - through Broadcom, Google Cloud and extra chip orders. And look how many of the biggest names are also Anthropic's shareholders, distributors or rivals.
A European 🇪🇺 footnote I enjoyed: six European banks - BNP Paribas BNP, Crédit Agricole ACA, Société Générale GLE, Deutsche Bank DBK, Barclays BARC and UBS UBSG - sit in the underwriting syndicate on the S-1 cover. The shovel sellers include the people selling the IPO itself.
So, should you buy them? Is this the back door into the IPO?
Not really - and this matters. Buying a shovel seller is not a cheaper ticket to ANT. For most of these companies, Anthropic is one customer among many, so their share prices move with the whole AI build-out, not with Anthropic's first day of trading. And when an IPO is in the news, they often jump on the headline - TeraWulf rose 8% on filing day - so part of the "indirect" upside may be gone by the time we read about it.
So read the map as an X-ray of what you already own, not a shopping list:
The clouds and chip makers (groups 1 and 2) are most likely already in your world or US ETF. Check before adding more - that's your one tonight's step.
The closer a company sits to Anthropic - neoclouds, ex-miners - the more its share price is a bet on Anthropic alone: one big tenant, lots of borrowed money. That's the nearest thing to an "indirect ANT", and also the riskiest. Only money you can afford to lose.
If you simply want more AI in your portfolio, a broad ETF or a UCITS ETF on AI infrastructure or chips is the calm way. Path B below has the steps.
Beyond the six groups: who else can grow after the IPO
Let’s take an even deeper look.
The six groups above follow the compute bill. They aren't the whole story. An IPO moves money through four more channels: stakes get repriced, up to $100 billion of fresh cash flows to suppliers, index funds are forced to buy, and investors re-rate everything that looks similar.
Group | Examples | Why it could grow - and the catch |
|---|---|---|
Owners of Anthropic stakes | Amazon | A public price marks their stakes up, and after the lock-up those stakes become sellable. Amazon's notes could convert into ~10.9% of Anthropic; Alphabet holds up to 14.99% plus an option to buy $10B at the IPO (Investing.com). Zoom put $51M in at a $4.1B valuation in 2023. Catch: much is already in their share prices, and a weak debut marks them down too. |
The chip chain behind the chips | TSMC | Every accelerator is made in a fab, on lithography machines, with high-bandwidth memory the filing names as an input. Catch: cyclical. Anthropic is one customer of many - which is also the safety. |
Networking and optics | Arista | Bigger clusters need more switches and optical links. Catch: analysts' inference; not named in the filing. |
Power, cooling and grid | Schneider Electric | Data centres need turbines, transformers, cables and cooling. Catch: not named in the filing, and many already trade as AI winners. |
Distribution and services | Accenture | Named in the S-1 as partners that sell and implement Claude. Catch: Anthropic is moving into applications itself. |
Financial plumbing | Goldman Sachs | Underwriting fees, listing and index-licence income. Catch: mostly one-off. |
European data-centre builders | Developers in Frankfurt, London, Amsterdam, Paris, Dublin, the Nordics and Southern Europe | Anthropic is hiring in London to sign European leases; Bloomberg tied it to a reported $10B, six-year deal for a 133 MW site in Norway. Catch: not confirmed by Anthropic, and there are few listed pure plays. |
And also - who could lose. Seat-based software, as Claude does more of the work itself. Nvidia's pricing power, if Google's TPUs keep winning the biggest share of the bill. Any small supplier whose biggest tenant is Anthropic, if the IPO disappoints - and the valuation talk has already slid from about $2 trillion to about $1.5 trillion.
Again - the names are examples for you to research, not a shopping list.
And finally - how to earn from it: two paths
You don't necessarily need to buy ANT to earn from this. But - you need to decide which path you are on before the noise starts.
Which path is yours depends on your investor type. If you took the quiz from Edition 7, you already know it. A Preserver or a Balancer will probably feel at home on Path B. A Grower or an Adventurer may want a small slice of Path A on top. Haven't taken it yet? Three minutes, no login: find your type here, then come back and pick your path 😉
Path A: you actually want to own ANT
Can I even buy it? Yes. From the first day of trading,
ANTis a normal Nasdaq stock: any broker app that lets you buy Apple or Nvidia will let you buy it. What's hard is getting it at the IPO price, before trading opens. The S-1 says the offer includes some EEA countries, the UK and Switzerland, but not which ones or through which brokers - so ask your broker this week. Even if they offer it, expect a small fill: in SpaceX's IPO, retail got about 20% of shares and some people got just one. And selling within a few weeks can bar you from future allocations.Decide how much before the excitement starts. A hot IPO can move by a third or more in a few days: SpaceX went from $135 to over $200 within four. That's thrilling on the way up and very hard to sit through on the way down. So choose the amount while you're calm: money you could watch fall by half without selling. For most of us that's a small slice next to our ETFs, not instead of them.
Pick your moment.
On day one you pay the opening price, not the IPO price. SpaceX priced at $135 and opened at $150. Use a limit order with your maximum price written in, never a market order.
Later is usually calmer: after the stock joins the index, after its first earnings report, or around the end of the lock-up (usually about 180 days, when early investors may sell; the "Shares Eligible for Future Sale" section of the prospectus says who can sell earlier).
Know what the price implies. At $1.4–2 trillion, you'd pay 71–101 times last year's revenue. In plain words: you're paying today for many years of very fast growth that hasn't happened yet. The lower the IPO price, the more room you have to be wrong.
Remember the dollar.
ANTtrades in USD, so your result in euros includes a currency move you didn't choose.Write your exit rule today. What would make you sell? A sharp slowdown in growth, a big customer leaving, a model switched off by regulators again. Decide while you're calm.
Path B: you don't want ANT, but you still want to benefit
Check what you already own. Open the factsheet of your biggest ETF and read the top ten. If it tracks the MSCI World, the S&P 500 or the Nasdaq-100, Amazon, Alphabet, Microsoft, Nvidia and Broadcom are probably already there.
Nasdaq-100 trackers get a bonus. Since 1 May, a new listing that ranks among the index's top 40 by market value can join within 15 trading days. SpaceX listed on 12 June and was in by 7 July. At any of the reported valuations,
ANTwould very likely follow, and your ETF buys it for you.
Decide how much bumpiness you can live with, and pick the layer that matches. The big diversified companies (clouds, leading chip makers) earn from thousands of customers and move less. The small suppliers (neoclouds, ex-bitcoin miners) can double on one contract and halve on losing it. Most of our money belongs in the first group. Only money we can afford to lose belongs in the second.
Look for owners whose Anthropic stake is big relative to their own size. The same stake moves Amazon
AMZNa little and ZoomZMa lot. The latest quarterly report shows the stake's carrying value. Then ask how much of it the share price already reflects.If you want European shovels, look one step up the chain. Lithography, power equipment, cables and grid companies listed in Amsterdam, Paris, Frankfurt, Zurich and Milan earn from every AI build-out, Anthropic's included. A UCITS ETF on AI infrastructure or semiconductors is the one-click version; check its top ten so you don't double what your world ETF already holds. (US-domiciled ETFs are generally not sold to EU retail investors.)
Before buying any shovel, ask: who pays the rent if the tenant can't? For the ex-miners building for Anthropic, Google guarantees part of the leases. Your risk is Anthropic's ability to keep paying, plus the guarantor's credit.
What happened when others did it
SpaceX SPCX recent IPO, June 2026. Priced at $135. Opened to the public at $150. Closed day one at $160.95, and was above $200 four days later. It joined the Nasdaq-100 on 7 July and fell about 6% that day. So far, both the allocation-lottery winners and the day-one buyers are fine.
So far. Jay Ritter's long-running research says most companies that go public underperform the market over their first three years.
Facebook, May 2012 - the company I'd later work at, now Meta META. Priced at $38 in May 2012. Below $18 by September. About 15 months to climb back above $38. The people who held were rewarded many times over. The people who bought the hype in May and sold in the September panic lost big time. The learning here (and honestly, almost everywhere - DO NOT PANIC 🧘).
The shovels, this Monday. On filing day, TeraWulf WULF rose 8.1%, Hut 8 HUT 4.2% and CoreWeave CRWV 3.5% (CoinDesk). The shovel sellers got paid the day the paperwork landed. The same sensitivity works in reverse.
I myself bought some, and benefited already. I also plan to buy a small number of ANT once it’s out, as a sign of trust to the company.
What I'm not claiming yet: that ANT will trade the way SpaceX did. Two IPOs and one filing day are not a pattern. The smaller true thing: the price you pay and the rules you set beforehand are the two parts of this you control.
Same kind of event, very different outcomes. The difference was the price people paid and whether they had rules before the noise started.
If you started a "Money calls" note after my conversation with Marina Davidova in Edition 8, this is the perfect next line: what you decide about ANT - buy, wait or skip - and why. As Marina put it, "experience is an action plus the time it takes to see the result." In six months you'll know whether it was a good decision or just a lucky one.
Your one next step
Tonight, open the factsheet of your biggest ETF and add up five weights: Amazon AMZN, Alphabet GOOGL, Microsoft MSFT, Nvidia NVDA, Broadcom AVGO.
That's how much of your money already rides on the companies that get paid before Anthropic does. Ten minutes, and the question changes from "should I buy ANT?" to "do I want more of this, and in which layer?"
That’s it for a long and number-heavy edition #9 today.
Hit reply and tell me: your number from tonight's ETF check - and which layer you'd want more of, if any. I'll share the range, anonymously, next week. I read every answer, and the best ones become future editions. Ready when you are 🤍
With love and excitement for this IPO,
Eugenia
Know someone this would help? Forward it, or send her the sign-up link.
Are we connected on LinkedIn? That's where I write about business, startups and AI. The full archive is on the website.
