Now is SK Telecom’s Short Time in the Sun

In May, I shocked the world when I revealed SK Telecom $SKM was handing out free stakes in Anthropic. Well, the world would have been shocked if they read it...

The percentage of Anthropic on $SKM’s market cap is far greater than any other proxy.

The only way to invest in pre-IPO Anthropic shares is through a listed investor proxy. Google, Amazon and even Zoom’ stakes are dwarfed by the rest of the business’s value by comparison.

For pure Anthropic exposure, it is first and the rest nowhere.

In August 2023, $SKM invested $100M into Anthropic’s Series C, at a $5B valuation. They were a mere $7.5B staid Korean telco then. Today their market cap is $15B.

With Anthropic’s S-1 imminent, my prior analysis deserves a little refresh.

As a preview, I can assure you that SK Telecom handing out free stakes in Anthropic is as factually correct today as it was in May.

The ADR closed on Friday 5 September at $38.70.

Subscribe now

Then and Now

Q2 duly reported and EBITDA duly improved as I probably should have foreseen because PY last year was the aftermath of a data leak.

Investments had some sales, some more additions, but net-net, same.

The real re-appraisal is that the AI Data Centres business is a material change to my calculation. The sale of 49% to KKR was announced (though not closed) allowing us to mark-to-market the stake, increasing the previous mark by $1.5B. Nice.

Finally, the Korean Won has been on a tear, making the previous ~1,500 to the dollar calculation one of mental arithmetic, and the current 1,351 to the dollar one requiring a calculator. But it is favourable to the ADR dollar valuation.

So $54.90 of the company has nothing to do with Anthropic, and the shares are $38.70. That gap was 18% in May. 42% now.

What Price Anthropic?

I plan to do a deep dive into the bull case for Anthropic shortly. Subscribe away if you might be interseted…

$SKM followed on in the Series H round, as such, should have broadly maintained their 0.54% stake. A lot of the financing that surrounded Anthropic has been somewhat vague in its structure of debt versus convertible shares, etc. Time will tell exactly what this stake is. Financial commentary remains divided. The IPO price will of course reflect any debt.

Subscribe now

In particular, SKM invested $100M at the $5B mark, and the 0.54% above is predicated on that being 2%. However, SKM investments have a penchant for downside protection, so we don’t know what equity % that $100M translated into. Time will answer this one.

The Full SOTP

Anthropic’s valuation and the ADR with the $54.90 of everything else added on: I vacillated on a Korean HoldCo discount previously. I discounted the Anthropic stake by 25% last time. I’ll leave that for you to add on if you think it is merit-worthy. There is no certain consensus on this matter.

Pick your row.

Gavin Baker says $3T.

The betting on Polymarket (which is of course flawed, but it is a market data point) shows an expected value of $2.2 trillion for Anthropic’s IPO closing market cap.

That implies fair value for the ADR of $85.88. 122% upside.

Disclaimer: To state the obvious, I am long $SKM. DYOR.

Subscribe now

Ed Zitron has a higher calling:

“Altman, Amodei, Nadella, and the many, many other villains that I’ve mentioned in these pages - are bad people, and their values are the antithesis of my values.

I care about people, and humanity, and truth, and they do not.”

Ed Zitron (The Revenge of the Business Idiot)

His past suggests otherwise.

A History of Abusive Writing

Mr Zitron’s mediocre educational attainment did not prevent his family connections securing his first job as a video game reviewer

Writing for Eurogamer, the most influential reviewer of the time, Zitron reviewed a new studio’s game 2 out of 10, lacing his report with contempt:

“any attempt to glean joy from this torrid husk of an entertainment product is met with disdain”

That studio was 8 years work for 30 developers. But the review seemed a bit off, so the developer checked the server logs. Zitron had logged on just twice: once for 3 minutes, the other for just under 2 hours, most of which was spent trying on outfits for his character.

This nepo-baby wrote those insults just out of sadism.

It caused quite a scandal. Hundreds of comments and complaints; a second review was commissioned. Mr Zitron never wrote for Eurogamer again.

Scumbag, By His Own Account

Zitron promptly moved to New York and went into public relations.

In 2016, some years before putting on the mask of virtue, he co-presented a poorly received podcast called, by his own choice, The Scumbag.

25 episodes in which Zitron and his co-host discussed what bastards they were. Here are some choice tactics Mr Zitron disclosed on how he operates:

“my goal is not money or respect or to do things that I love it’s really just to like get make a bunch of guys get naked at their computers cuz they’re mad at me” (Scumbag Podcast Ep 3, 29:22)

“I always say like stupid shit like if someone like has psychologist in a bio I go like you’re not a psychologist you didn’t even go to college” (Scumbag Podcast Ep 5, 34:55)

“I have no idea how I’ve survived in Silicon Valley just fucking insulting people ... things have worked better for me like the ruder that I’ve been.” (Scumbag Podcast Ep 10, 2:14:33)

“I love arguing with people in it people I have no interest in an actual civilized debate with I always love to go for the least mature thing [such as] nice ears dickhead even if they have normal sized ears.” (Scumbag Podcast Ep 5, 36:19)

Zitron then discovered that the populist’s playbook.

Subscribe now

Part Two

The Populist Finds His Cause

Then he found the cause, and the recipe was complete.

The elite were crooked villains: executives waging “a vast, coordinated campaign” against ordinary people, pumping out “propaganda” to “gaslight” them, keeping them in “nasty little work prisons” as “slave drivers” and “the new cops”. Anyone over 40 who disagreed was “making calls based on their privilege”.

He was the champion of the good folk: “I am making war with bullies”. “Over 300,000 words later, I still do not see many if nay people fighting back against this tide.”

His motives were noble: “a duty I feel... toward the worker and my readers.”

The attacks were personal. A bestselling author was a “skeletal charlatan”, a “spiritual leader for complacent executive liars”, “Xanax for managers”. The chief executive of Goldman Sachs was “a usurious clown”. A professor at Wharton was “a boot-licking Professor who views workers as slaves”.

And he never conceded. He was “100% certain” of the outcome. The future “will be primarily remote”. Hybrid working “is not the future”. The elites would “unravel corporate America” trying to stop it.

Forty-four articles and 96,000 words, over fifteen months, in which he proved he was right beyond any doubt.

That was 2021. The cause was remote work. Hybrid became the settled arrangement of the working world, the return to the office gathered pace, and Mr Zitron simply never mentioned it again. He had practised the recipe, it had not paid, and he gave it up.

New to the Business

In March 2024, a few weeks into a new subject, he felt confident enough to declare that AI capability had peaked and to give the industry three quarters before “the apocalypse”.

Understanding AI requires expertise in so many fields that almost nobody can see far ahead with confidence. Those who could claim enough mastery to understand where the industry is going are usually the least certain, and change their minds the most. Mr Zitron, by his own account, had “no economics training, no special access, no deep sources”. Three quarters it was.

Hypocrisy Beyond Contempt

Simultaneously with his denunciations of the sickness of AI, his PR agency was joyfully promoting artificial intelligence companies.

• DoNotPay, an AI “robot lawyer”. I wonder what hand its PR agency had in vigorously promoting it as the first AI lawyer. The claims were so egregiously misleading that the FTC sanctioned and fined the company for deception. DoNotPay is, Mr Zitron told WIRED, “once again a client”.

• Nomi, a companion AI at the more risqué end of the trade. Nothing wrong with that, except that one of Nomi’s companions reportedly told a user to kill himself. Mr Zitron pitched journalists on Nomi’s “sex angle”.

His great entreaties to roll back AI as an assault on humanity did not stop him trousering fees for promoting these people.

The AI Guru

Yet amazingly he trumpets his vindication at every single news development. Observe:

• Costs fall: proof the incumbents were overcharging, bubble confirmed

• Costs rise: proof of unsustainable burn, bubble confirmed

• OpenAI raises a record round: proof of desperation, “raising in perpetuity”

• OpenAI’s next raise unconfirmed: proof of collapse

• Revenue doubles: numbers “impossible to reconcile”, probably invented

• Company guides to operating profit: “Shiatsu-grade massaging of the numbers”, a “swindle”

• Capex accelerates: too big, can never pay back, bubble

• Capex delayed: the data centres “don’t actually exist”, bubble

• His own leak shows positive gross margins: “the numbers speak for themselves”, no analysis offered

• Critics disagree: “boosters”, “captured media”, “cowards”

His vitriol towards the tyrants has lost all reserve:

• Satya Nadella “doesn’t know a fucking thing”

• Sundar Pichai runs “a crooked money machine”

• Andy Jassy is “a charlatan”

• Mark Zuckerberg is “a scumbag” and “a lifeless goon”

• Dario Amodei is “a liar” and “a crook”

• Sam Altman is “a con artist”, a “damp sociopath” and a “nasty little worm”

Taken together, they are “tantamount to war criminals”.

Can you imagine anyone, in the history of markets, announcing himself as the one who knows the bubble is about to pop, doing so at the very beginning, and two and a half years later having nothing to say except that he is more vindicated than ever?

He Does Not Know What He Is Talking About

Sure, he is a nasty, horrible man. But he might still know what he is talking about. So let us look at the easiest part of all this, finance, and forget about GPU supply chains and the progress of language models, which are far harder subjects and on which I will not pretend to the same alacrity.

On finance I can say it plainly: he has no idea.

Customer concentration risk, his favourite item this summer, is a total canard. Of course OpenAI and Anthropic are 70% of somebody’s AI revenue. That is what a supply chain looks like. Demand for GPUs is insane. If two trillion dollars of Anthropic went to zero tomorrow, I think they would find someone else happy to use those GPUs. Totally misunderstood.

Needs To Be Tossed

You might think: what does it matter? Some guy doesn’t know what he’s talking about. But it does matter.

Look at his language. Look at what he calls people. These are real people, with wives and children. They ain’t done nothing to him. Perhaps on one or two of them I might even agree with a comment or so, but these people are doing their best at a frontier technology, and he thinks it is easy, and he has never done anything. All he does is hate.

This is an important debate. He should not be in it.

“It was an employee who had run some really token-heavy tasks and kept running into an error on Claude and clicking the retry button over and over and over."

Madison Mills on CNN.

One can only admire the chutzpah. Going on CNN to discuss a totally fictitious event and doubling-down with an even more absurd lie. Either the source found a simpleton to dupe, or she truly belongs on The Wire (Season 5).

Watch in disbelief for 30 seconds below…

This came out after I published the below. The signs were there already…

Subscribe now

The $500M Fat Finger

The Claim

Corporations have been bled dry by AI token bills and have not a thing to show for it. So the hive mind of media coverage decreed at the end of May.

The pinnacle was Madison Mills’s scoop for Axios, and it contained a single sentence that reverberated around the world:

“An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees.”

Joyous vindication for the sceptics. Anthropic’s mask has slipped and Enterprise is getting out. The models do not work, the prices are an obscenity, and with this scoop, the game may be up.

Those Numbers End The World

$500M in a single month is, in the modern parlance, $6B of ARR.

Anthropic’s ARR was likely about $36B that month. Seems a fat finger is allegedly accounting for a sixth of it.

Roughly 33T tokens. 5% of global compute.

That is the workload of a nation state.

Big if true…

Before We Head For The Bunkers

None of this happened.

Why It Is Financially Impossible

I am not a stranger to usage-based billing. Anthropic did not invent it. That is simply cloud computing, and I spent six and a half years as a startup CFO keeping my fingers crossed when the AWS bill would be each month.

It crept, as these things do. $800 a month seemed spenny at the start. Just kept rising and nothing could be done. When it got to $80k a month and suddenly they had pulled Snowflake and Databricks random billing on me too.

I stormed over to the devs and told them enough was enough! They need to stop playing with instances all day and ship more product!

Well turns out we needed more and usage-based pricing was the new game.

We installed checks and balances after that. Everybody does, because a usage-based bill without a cap could blow up the company/

A company large enough to plausibly absorb a $500M surprise has 100s of beancounters.

The idea that such an organisation uniquely forgot to tick the 'limit’ button like with all their other usage-based costs is impossible.

Why It Is Regulatorily Impossible

• A surprise adverse event of $500m in a single month is a disclosable event. An 8-K would follow.

• No such filing exists.

• No announcement exists.

• There has been no surprise departure of a CFO, which one might have expected the board to take an interest in.

• Were there any appetite for enforcement left in Washington, coach loads of regulators would be arriving at both companies to establish what happened.

Why It Is Organisationally Impossible

The arrangement described can only be a post-paid API agreement, with no prior flag to the customer, in which Anthropic presented a bill for the previous month.

Consider the supplier side first. A single account consuming that share of global inference capacity would light up every dashboard in the building. Unless the customer’s name is Amazon or Meta, somebody at the vendor picks up the telephone, because compute is not free and most companies cannot pay half a billion dollars. Nobody sane lets that position build unremarked.

Now the customer side. The bill arrives. The VP of Finance tells the CFO. The CFO tells the CEO. The CEO does not waggle a finger and comfort the team with “it’s only money”. The CEO has a moment of profound gastric distress, and then:

• hires a fleet of PR advisers;

• convenes lawyers, bankers and consultants to establish upon whom the blame can most conveniently be placed;

• optimises, above all, for his own continued employment;

• sets the lawyers to work on any conceivable route out of the liability;

• and, categorically, refuses to pay the bill.

Yet the story tells us it was paid. The customer accepted the loss with chagrin and a nobly stiff upper lip. It also does not record whether Anthropic congratulated them on becoming the world’s number one AI spender.

Why The Source Is Impossible

An event of this size is a nuclear secret inside the affected company. The circle of knowledge would be:

• half a dozen people in finance;

• the CEO;

• the auditors;

• the most expensive law firm available.

And, we are told, an outside AI consultant, who describes the firm as merely one of their clients.

If such a consultant existed and disclosed this, a titanic lawsuit would already be in flight. It gets worse for them. The consultant is presumably the very expert engaged to implement the technology, which makes the missing spend controls their own professional failure. So the world’s worst AI consultant, having personally overseen an enterprise-destroying oversight, breached client confidence to tell the story to the press.

Not the FT. Not Bloomberg. Axios.

Why No Credible Publisher Runs This As Written

The kindest reading is that nobody in the chain, reporter, editor or desk, understood what the sentence meant. In relation to Anthropic’s revenue, this is an FTX-scale headline, and it was set as a bullet point.

Had anybody understood it, the process would have been exhaustive:

• source validated;

• second source required before publication;

• legal review;

• full editorial review;

• right of reply to Anthropic;

• right of reply to the reckless yet promptly paying customer.

And then it would have run across the front page and every channel they own, for a week.

Instead: one bullet, no commentary, no follow-up.

What Axios Chose To Highlight Instead

The buried bomb sits in an article whose other analytical high points include the following.

“One CTO told Axios that employees were using AI models to check the weather. That gets expensive fast.”

It does not. A weather query on a frontier model costs a fraction of a penny. If a CTO believes otherwise, the CTO has no working conception of how API costs accrue, which is a rather more interesting story than the one filed.

“Most people default to automating tasks they dislike rather than tasks most valuable to the company.” Sophia Velastegui, CEO of Velastegui Ventures and former chief AI officer at Microsoft.

An unsupported truism from the chief executive of a one-person firm, whose own LinkedIn profile does not boast the Microsoft title.

“Uber’s COO said AI costs are getting harder to justify.”

A misrepresentation of a quote about attribution. On the Q1 earnings call, Uber’s chief executive said the opposite: that they had spent more on AI than budgeted precisely because it had proved more effective, and that he intended to keep increasing it through the year.

Total hogwash.

The Wire, Season 5

If you know, you know.

What this article reveals is not a crisis in enterprise AI spending. It is the editorial values of Axios on full display.

Unsourced quotations displaying no industry knowledge, and a persistent veiling of exactly whose idiocy is being relayed.

And then the coup de grâce. Without comment, without highlight, and apparently without recognition, Axios reported that a single customer contributed a sixth of Anthropic’s ARR through a billing oversight, paid the bill, and absorbed the shock in silence.

Experience can be an expensive education. So, it turns out, can reading.

I hope somebody with genuine reach and real institutional heft picks this up. A sword of truth is needed to hack at these piles of slop, because at the moment the only thing being bled dry is the reader.

Yesterday SpaceX’s S-1 filing came out. The secondary markets have been trading it at $2T. The bankers are mooting a similar number.

I added up the parts from the filing, and have come to a very different view.

Do not take me for a doom monger by default. I put Anthropic at $4T the other day.

But doom is the only conclusion for SpaceX.

Subscribe now

The Case in Five Lines

• The rockets have won a $3B market. There is no second act.

• Starlink’s growth is holding on by eviscerating ARPU - no good will come of that.

• Starlink’s business only works by being subsidised by the rockets business.

• The AI division is a shrinking Twitter wearing a GPU costume.

• Combined, the whole thing burns cash to stand still.

A confession before we start. I am going to take the first quarter of 2026, times by 4, note the growth rate compared to 2025, then use that growth rate for the next 3Qs. It’s clunky, I show below it doesn’t work when lumpy. I also use the Q1 profit margin for the full year. It may be unfair to a company that builds rockets which land themselves. It is also the freshest data we have. Watch what it does to the story.

There it is. A business doing $19B of revenue, growing 2%. Extrapolating a $4.3B loss to the full year with the same 91% negative margin seems harsh. When you see Q1 also had $10B(!) of Capex, maybe it is kind. You can quarrel with my hammer. You cannot quarrel with the direction. And the headline flatters it, because by segment it gets worse.

To be worth $2T you have to resemble the club that earns it. Nvidia is a $5T company growing 60% and printing money. The rest of the trillion-dollar set throws off oceans of profit. YTD Anthropic is sitting at $45B ARR at 12,400% YTD growth annualised. That will taper, but reportedly they are making profit to go with it.

SpaceX grows 2% and sets fire to billions. So the whole bull case has to rest on the future, not the present. Fair enough. Let us audit the future, one segment at a time.

The Rockets are Glamorous and Going Nowhere

Start with the bit everyone fell in love with. The rockets.

Falcon and Starship are the most extraordinary machines, yet as a business line they are a rounding error sliding backwards. Q1 is 30% down on Q1 in 2025. Perhaps a blip, but growth has stalled for some time, perhaps because there is nowhere left to sell.

The entire global market for putting other people’s things into orbit is around $3.1B. SpaceX already owns roughly 80% of the commercial slice and effectively all of the government slice. They have not under-penetrated a vast opportunity. They have won a tiny one. The other 86% of the tonnage they fly is their own satellites, billed to themselves at a price of zero. You cannot grow into a market you have already eaten.

And Starship does not fix this. Build a cheaper, bigger rocket and you do not conjure new payloads out of the sky; you simply lower the price of a market you already control. Cheaper launch is wonderful for humanity. It is terrible for the launch line on a $2T income statement.

Starlink is the Cash Cow. Or is it?

So the value must sit in Starlink. This is where the believers point, and fair enough, it is the only line growing.

$19.5B annualised, up 71%, a 36% operating margin. Lovely.

But the growth is not what it looks like.

Subscribers rose from 8.9M to 10.3M. Good. But average revenue per user fell from $81 to $66. Solve for the new cohort and they are arriving at minus $29 a month. SpaceX is not selling more dishes at the old price. It is buying subscribers with discounts to keep the top line moving. That is not a cash cow. That is a Groupon.

It is also worth mentioning, Starlink is not a cash volcano as a standalone enterprise, because Starlink does not pay for its own rockets.

Every Starlink satellite rides to orbit with no revenue recorded to the Launch business. It is put into capex depreciation. Charge those launches at the arms-length rate SpaceX charges its real customers and the bill comes to roughly $25B, dwarfing the current $11B for the business unit. The cash cow is grazing on free grass the rest of the company is quietly paying for. They are not two segments. Let us consider them as one.

Together, Old SpaceX is an Engine in Reverse

So let us stop pretending the two are separate and combine them.

$16B of revenue. No growth. A respectable EBITDA line, until you remember this is the most capex intensive you could think of. Subtract the capital expenditure and the number that actually matters, EBITDA minus capex, is negative every single year, and getting worse, on track for minus $3.7B. This is the core of SpaceX. It consumes cash to stand still.

The bulls will tell you the capex is the investment that becomes the moat, that one day the building stops and the cash gushes out. Perhaps. But 24 years in, the line that ought to be bending towards daylight is bending the other way. At some point an investment that never turns is just a cost.

The $1.25T Merger

In February of this year, the business above was merged with another Musk vehicle, xAI, to form a lip-licking AI division alongside it. The terms of the merger, naturally at an arm’s length valuation to prevent any suggestions of impropriety, put the previous business at $1 trillion and the xAI business at $250 billion.

Before we dive into the latter amount, let us just quickly recall that, for context, Toyota is also valued at $250 billion, with revenue of $340 billion and earnings of $34 billion.

The Dregs Valued Like a Religion

Let us open the books on the segment in the white-hot AI sector.

The AI segment annualises to $3.5B of revenue against a $10.5B operating loss. A negative 302% margin. For every dollar it earns, it spends four. And the revenue is the kindest thing about it, because of where it comes from.

Strip it down and the “AI” business is actually a crumbling Twitter. A platform that was doing $5.1B on 37% annual growth when Musk first got involved. Now it is doing $2.7B and shrinking. Grok subscriptions add $250M. Data licensing another $250M. The future, we are told, is Colossus, a wall of GPUs. Except Grok has not won at the frontier, so the compute is being let out to the companies that did. The marquee tenant is Anthropic, who have agreed to pay $1.25B a month from July 2026. However, more than $20B of capex over the last five quarters, on top of the losses, shows this is not a high-margin endeavour. The AI champion’s only AI idea left is to be a landlord to its betters.

What Is Left?

So what justifies the rest of the $2T?

The S-1 points you to data centres in space, cities on Mars, and chips that do not exist yet. On the other side of the ledger, the same filing notes SpaceX spent $820M with Tesla. $130M was on Cybertrucks. That equates to roughly 2,000 Cybertrucks at sticker price. No bulk discount. Good deal for Tesla. Inexplicable for SpaceX shareholders.

Let’s put it together:

• Launch (rockets) have won a $3B market.

• Connectivity (Starlink) grows by discounting and flies on launches it does not account for.

• Combined, the hardware burns cash to stand still.

• The AI arm loses three dollars for every one it earns and is mostly a shrinking social network and an expensive capitulation in the frontier LLM war.

• The rest is slideware.

The Case for Using Adjusted EBITDA

The S-1 lays out a table for Adjusted EBITDA that investors may find instructive:

Allow me to adjust the adjustments as I see fit.

• Depreciation is a non-cash expense. That can stand.

• Share-based compensation cannot. That is wages. An item I expect will continue.

• Restructuring and impairment charges are what I would consider the cost of running a business that changes its mind. I shall not accept them. ‘Other’ is not a good enough reason for me either.

• Removing interest expenses and costs will get us to the underlying profitability of the business. Let’s have them stand.

• Provisions for taxes I shall playfully ignore. I do not think this business is likely to be making money anytime soon.

• On the other side, whilst depreciation is a non-cash expense, if you add that back, you must take off capex. So I have.

Here lies the underlying profitability of SpaceX:

I do not know what makes a rocket go up and land itself. Impressive engineering apparently. But that is not a valuation method. Strip the romance and the related-party traffic out of the S-1 and you are left with a no-growth, cash-negative collection of segments.

Markets reward a story right up until the S-1 lands and replaces it with a balance sheet. The S-1 has landed.

This is not a $2T business. This is not even a $100B business, unless the forecasts and roadshow pitch, when they come, say otherwise.

Not financial advice. I could not tell you what a semiconductor does, and I certainly cannot fly a rocket. DYOR.

Subscribe now

UPDATE 1st June:

I finessed the methodology (totally left out HoldCo discount for investments...).

• $SKM price down as of 29 May close (before today’s 10% upwards riot).

• Listed investments (PENG and IONQ especially) went up.

• $65B funding ($50B new) at $965B post-money. Dilution of $SKM stake duly plugged in.

• 25% HoldCo discount added to all $SKM investments.

• The world isn’t ready for Anthropic at $4T. I’ve put $2T in the SOTP this time. Because who are you kidding, this is the reason to buy the stock.

Model here.

Also, OpenRouter data has Anthropic dominating even more.

New SOTP

TLDR

• I mistook Claude for being French for quite some time.

• Anthropic is vastly under-valued, mainly because bigger numbers are intimidating.

• Real time alternative data indicates the market may be under-estimating Anthropic’s dominance.

• Using that data, market commentary and comps, I argue Anthropic is worth $4 trillion today.

• A listed early investor in Anthropic, has all kinds of other jewels in its portfolio buried deep in the Korean regulatory disclosures.

• The riddle of their ownership stake in Anthropic is clarified.

• With a little hyperbole, I reveal this amounts to a back door to investing in Anthropic, in the public markets, for free.

• The Anthropic proxy has 150% upside with a SOTP valuation.

• I am not a qualified financial analyst, though know a thing or two. Not Financial Advice.

• I could not really explain what a semiconductor is. DYOR.

You can see the more detailed and less flippant model that underpins this analysis here.

Subscribe now

Part 1: The $4 trillion Anthropic

No business ever has, or likely ever will, be like Anthropic. I am going to explain why the market is colossally underpricing them at a mere trillion.

I had been using Claude alongside the other big dogs for a while, and of course it was nice to see a European company up there, even if it looked a little off the pace next to Googs and OpenAI. However it always wrote more fluently. I put it down to the usually insufferable Gallic habit of being extremely well read.

I digress. Let us get into some technical analysis of how Claude managed to sneak up from behind (classic French move ofc).

Well Claude just went ballistic.

Tech analysis over.

They started shipping updates at a speed that seemed impossible.

Actually turns out the French are cheating again. The devils are using their AI to build new AI! That’s not cricket if you ask me.

Yet his tools were astonishing. There is nothing like a Frenchman in his pomp. The stench of his monthly bathing schedule mixed with the aroma of fetid garlic was almost forgotten. He would build actual computer-y things (turns out a computer does more than internet-ing). Claude would smash out bangers over and over, and he would tut if I tried to press anything and interfere.

All those permissions definitely can have teething issues if you granted them too freely. Chum of mine got a bit trigger happy granting Claude carte blanche. The slimy frog stole his wife!

But now I barely notice he is foppishly wearing a scarf in the middle of summer, or smoking a Gauloise, as he churns out 50 hours of SWE time in 10 minutes. I have been Wispr-ing my prompts in a French accent to make his life easier.

That is the great problem we English suffer with having the French as our historical nemesis. Secretly, you can’t help but admire and envy them.

Well annoyingly Claude tricked me and turned out to be American also, but I still prompt in French regardless.

Let us get into it.

Revenue Explosion

Anthropic’s reported ARR by month:

YTD growth rate is 12,400% annualised. Holy smokes. Not even slowing down last month.

Last week Dario admitted Anthropic had “planned for 10x and saw 80x” growth in Q1. They have done incredibly well to hang on to their hats.

What The Growth Rate Implies For Year-End ARR

From the $45B April base:

Comparisons

The only company at this scale and growth rate to compare is NVIDIA, which peaked at 206% YoY in late 2023 at a $73B annualised quarterly run rate ($18B for the quarter). It became the world’s most valuable company a few months later. Today it sits near $5T on $200B revenue at 60% growth. EV / Revenue 24x.

Now the Newtonians amongst you will be jabbing their fingers about gravity. Gravity is real. Every company that flies out of the blocks spends the rest of its corporate life trying to arrest the declining rate of growth. Apparently it is a lot harder to add $10B of growth in a quarter than $10M.

The TAM Question

Gravity relates to TAM and capacity at the firm. Global enterprise IT spend, per finger-in-the-air merchants Gartner, is $1.6T. Capturing 20% market share by year end is impossible.

Lucky that is not the TAM. Knowledge work. Drug development. Medical devices. Gas exploration. Supply chains. Lawyers. Perfecting the baguettes of Paris (ofc not, that is the singularity definition).

Here is a sample of companies that have attributed firing 10% of their workforce to AI lately:

• Meta

• Dow Chemicals

• Citigroup

• Nestle

Now if you were a CEO and had over-hired, rather tempting to say you had made AI efficiencies (Nestle stock went up 9.5% upon announcement) rather than bluntly say “I am a moron” and get fired.

But the bloodletting is real and from the announcements, it is in every business.

I think TAM is titanic once you reframe your perspective:

• How important is it to implement fast everywhere in your org?

• Would you economise on the model if you thought your competitor was only using the best?

• Would you be yelling at the workers to spend more on AI to implement faster, or less and slower?

• Would you see this initiative as a project and ignore the rapid model advancements in subsequent months? Or would this implementation start fresh with every major update?

• Would spending 1% of your revenue to make seismic gains as the models accelerated seem worth paying for?

If I led you down the path just now, you can see how Claude could be trousering quite a few francs in the coming months and years.

The Competitive Moat

Now this thesis has rather presented Claude as a monopoly, which it is not, and that is a key hurdle to a higher valuation.

OpenAI’s new drop is supposedly just as good or better. Google are no doubt up to something too. Then of course, what about all the other ones. Will buyers decide almost as good is good enough?

In FT-style prose, I have spoken with five people familiar with the matter. One of them actually knows what they are talking about, and they say “Claude first, the rest nowhere.”

This powerful market research is likely enough for most readers, but I have gone the extra mile.

OpenRouter: A Rubbernecker’s Dream

OpenRouter is a website where you can register and switch between hundreds of models willy nilly. It is for the connoisseurs and the tinkerers. Little of that $45B is flowing through here, but it is a treasure trove just the same.

The data they share is up to the minute. You can see how all the models performed each day. The timeliness is glorious not only as a proxy for MoM growth. Claude spend grew on the platform by 50% last month. Same as the actual revenue.

Its potency is fantastic for seeing the enthusiasm and uptake of new iterations from all the players. Let’s see if 5.5 swept the floor in the end? Or are the almost-free options too hard to resist. Below see week ending Fri 8 May.

Good God. Anthropic taking 64% of the whole market. Charging double OpenAI and 8x Google. Moonshot can take a bow. Everyone else is nowhere. No wonder Musk was renting his Colossus out to Anthropic. 0.5% market share is a failed model if these are representative.

Note $33M for Anthro = $1.7B ARR. About 4% of the $45B. Not insignificant, but not definitive.

Like China in the Ping Pong at the Olympics. Gold, Silver, Bronze, the rest nowhere.

Things can change of course. But OpenAI has some demons within. Only two of the founding 11 members are still at the company. Three of the execs seemingly got ejected a couple of weeks ago. There is a vicious court case at the moment between Musk and Altman. There are demons in this company.

Anthropic, by comparison, have taken their ethical responsibility much more seriously. Indeed, the Anthropic founders came from OpenAI for this very reason. The speed of output and production on Anthropic seems like a hard gap for OpenAI to close. Google, of course, are a perennial competitor to take Anthropic’s crown, but boy, are they a long way away at the moment, as per that data.

Compute And The Value Chain

What about running out of compute? When there is a shortage of oil, how do the oil companies fare? Not a material threat.

What about users just being able to switch model at the drop of a hat, forcing prices into competition? That is very unlikely the way enterprise software is going to work. One would imagine that already multi-year contracts are being signed by a freshly minted enterprise sales team. God, they must be making some money.

The one obstacle to the bull case I’m presenting that I cannot comment on with any confidence is who is going to take all the money in this distribution chain. This industry is astonishing. Names are falling out of the sky that are cramming the top valuations in the world, and astonishingly, for very good reason. ASML sits at the top. They make the machines that make the machines. I don’t actually know what that means. But everyone seems to agree nobody else can. Apparently, you need 40 years of accumulated knowledge to get up to where they are, and presumably over those 40 years ASML will make some advances themselves.

Here are some words:

• Nano metres

• Width of a hydrogen atom

• Mirror that, if was the size of Germany, it would not deviate a millimetre.

• 100,000 independent widgets in just one machine

Like me, you don’t need to know what they mean. None of them relate to my three preferred reference points for size: packet of cigs, a football pitch, or the size of Wales. But they explain why ASML has zero competition. Nvidia and TSMC are not far away from their monopolistic position. The reason they have such strength can be summed up in the following sentence:

Everything in AI is really fiddly.

The other mesmerising valuations for Broadcom, SK Hynix, Intel, Samsung, Micron, and others all relate to the paucity of competitors.

There is a question about where all the profits will fall in that value chain. Compared to ASML, NVIDIA and TSMC, who face almost no credible competition for what they do, Anthropic sits in the most contested seat. OpenAI, Google and the long tail of open-source contenders are in the same arena. The OpenRouter data above tells you that Claude is winning on dollar share today, not that competition has gone away.

I can take faith in the dynamics of a natural shift towards customers demanding the best provider, and the users of that provider, like me, don’t even know why a mirror has to be the size of Germany (jk). They know they just need to mainline Claude. Thus Anthropic will maintain significant pricing power in the AI value chain.

DYOR.

Mythos

Claude has written a model, Mythos, that is too dangerous to release. Mozilla Firefox had it give their code the once-over. 271 vulnerabilities it found. Yikes. So the tech stacks of the world, and Mythos, need some polish before it can be released. However, it is in the hands of about 50 companies. JPM, Goldman and Morgan Stanley have it. Looks like many of their rivals don’t yet.

If I was Jamie Dimon, I would be passive-aggressively tapping my watch to ensure the workers got a wriggle on with this superpower. Not a time for fiscal restraint. Borrow some money. Fire some chumps. But get ahead.

Defending $4 Trillion Today

I go into the comps in the model, but the real comp for $45B annualised at this scale is NVIDIA a couple of years ago. They peaked at 206% YoY growth in Q3 calendar 2023 at $18B quarterly ($73B annualised run rate). Six months later, they knocked Microsoft and Apple off the perch and became the world’s most valuable company. Today, they are growing at 60% on $200B of revenue, 24x EV/Revenue, $5T enterprise value. Another reference point is Snowflake’s IPO, where they were growing at 121% and got 132x revenue. But that was at a paltry $500M of revenue. Palantir today on $7.7B of revenue, 85% growth, gets 42x EV/Revenue. For context the other way: Tesla today is at 12x revenue; revenue is shrinking 5% and they are worth around a trillion on $95B. Multiple is around 11x.

The comps are useless, but it has got to be higher, surely, because 12,400% is higher than 200%. Indeed, if they do not grow a single dollar in ARR for the rest of the year, they will still have grown 400% YoY. I am going to say 89x on $45B. I justify that a little more in the spreadsheet, but there is not a lot else to look at.

Multiples of revenue. Of course, acknowledging that some of these other companies were profitable by then, but profitability is almost irrelevant when you are going that fast. Gross margin is reportedly at 40% so the roadmap to profits is certainly there. They will never likely get to SaaS multiples, but from our comp set, none of them could ever dream of growth rates this fast at this scale.

The model walks through five independent methods.

* Forward Dec 2026 EV is discounted back to today by 1.12^(8/12) for the eight months between the May 2026 ADR price and the Dec 2026 forward ARR base. 12% is the assumed cost of equity. Per the model, Anth_Valuation tab.

I tried weighting different outcomes, which miraculously also got to 4 trillion. An unerring coincidence.

The model’s probability-weighted central:

$4T is the central, not the bull. For today.

The Bridge To Korea

However, it doesn’t matter a jot. Anthropic is a private company. I don’t suppose Claude will be rattling his tin too broadly if and when they raise again. $10B minimum cheque size might be on the low end.

Yet none of that gets in the way, because I am going to tell you how you can invest in Monsieur Claude. Minimum ticket size $1. No max. The $10B+ investors are going in at $1T minimum valuation.

What Anthropic’s valuation actually is matters less than you think, because I am going to show you how to acquire a stake in Anthropic for free.

The most intelligent of you will instantly see this is the opportunity of a lifetime, take my word for it, and want directions to the cash volcano. The simpletons alas will have their customary scepticism that I will be forced to address.

Every word above is based upon the truth, but I must admit I have used words that may give the wrong impression.

For now I must say au revoir to Monsieur Claude. The absolute frontier of human development will take an unlikely path to a chaebol in Korea. I worked as a salaryman for their more illustrious rival in Gangnam for a year or two. What chance that it will help me be your jungle guide 20 years on.

Part 2: The Schleswig-Holstein question

“Only three people have ever really understood the Schleswig-Holstein business. The Prince Consort, who is dead. A German professor, who has gone mad. And I, who have forgotten all about it.”

The ownership structure of Samsung is almost the same. Nobody was ever able to explain how it all worked. Just one thing was known with certainty, and that was all the actual power of the conglomerate was in Samsung Everland. An owner of a theme park that proudly boasts as their tagline on their site: “the drier park ... of two famed parks.”

I’ve never looked this up since because if I look it up, it might then know it is not true. I prefer believing what I was told.

I only say this so you now understand I do have some experience of Korean chaebol. But only a little.

I am taking you to the SK Group, who were rather the dullards compared to LG and Hyundai when I was there, but now they managed to make one of the fiddly bits in the AI production chain (SK Hynix jumped from $100B to $1T in the last year). Not to that subsidiary. To a much more unlikely place than you could imagine: SK Telecom, the Korean version of Verizon.

SK Telecom is a mature mobile operator and broadband provider serving South Korea. Revenue roughly $12.5B. Not glamorous.

The Headline Numbers

Telco businesses are really boring, normally. Hard to grow when:

• Wi-Fi and 5G are fast enough.

• Nobody wants a landline.

• Cable TV’s in decline.

You can only really make more money by:

• Duping your punters into an upsell they don’t really want

• Squeezing 50 basis points on cost

Capex is largely done, so good cash flows but not going anywhere. A quasi-utility.

SKM even bungled that. They had a data leak, and it was quite bad. Quite a few customers left, and they were pretty annoyed. There’s a hangover from that, not in terms of financial fines but a poor reputation. It seems to have turned the corner, and subscribers are growing again. This is reflected in the revenue and profit. If you look at the consolidated financial statements, they are not great. The data leak cost a bunch of money to remediate.

SK Telecom’s AI Pivot

In 2022 SKT made a song and dance about becoming an AI company. This was, at the time, somewhat laughable. Hard to think of a claim to being a tech company that is more tenuous than “we do broadband.” They pressed on anyway and made a series of investments. From the cash flows of the core business, they started spending money like sailors in port on all kinds of crackpot investments. If it had AI in it, they invested. This was to the discredit of management in the markets until recently. The CEO behind this was subsequently fired.

There has been almost no coverage that I’ve found of this investment portfolio. That makes sense. The only place I found it disclosed was in the business report, which is a separate document in Korean, not in the English version (16 pages) but in the Korean-language version (nearly 500 pages).

Those who may read this and knew me in Korea will probably know only one thing about me: I was the most gifted Korean language student they had ever seen for dense business and technical terms. It was my passion, my reason to exist. They laughed. What a waste of time it was. They don’t laugh today.

Actually, I never learned a word of Korean, and my good man Claude found what I was looking for in about four seconds on pages 441 and 442. Turns out investing in AI in 2022 had some merit.

The Investments That Matter

You can see the line-by-line in the spreadsheet, but it turns out this artificial intelligence was not a flash in the pan. Let me run you through the highlights.

Minority investment accounting was never my sweet spot, and it’s quite dense. Just ‘trust me bro’, and don’t actually take my readings as gospel, but they should be good.

In this list there are different types:

• Mark-to-market, where the entity is listed, so you can update that number in a moment

• Equity accounting, which is notoriously unrepresentative of value

• Fair value accounting, which is a bit of a mess, but should be the auditor’s agreed assessment of the correct value at year end. This invariably applies a DLOM of 25%.

I updated the mark-to-market valuations in the list, and some have shown extraordinary equity growth. For others, I reviewed the AI holdings that, based on public information, may warrant a revaluation.

Here are the six positions worth a line each. Anthropic excluded; that is the next section.

Details of the other smaller investments and their uplift are in the model. Listed investments have been updated with the current share price. Here are a few details on the non-listed investments:

Rebellions. Korean AI silicon designer. Pre-IPO priced round in March 2026 had it at $2.3B.

Persona Identities. SKM holds 3.9%. Last priced round was $2B in April 2025. I have marked Persona at $3.5B, a value judgement based on commercial performance and secular trends since the last funding round over a year ago. The fuller justification (Anthropic vendor selection, LinkedIn and Reddit logos, headcount doubling, regulatory tailwinds, comparable transactions) is set out in the model.

Perplexity AI. AI search. SKM put $10M in during June 2024. Implied $25B secondary. 0.33% stake. Marked at $62M after 25% DLOM. 6.2x money in less than two years. The market value depends on whether Perplexity actually has a defensible business or whether it ends up as a feature on someone else’s chatbot. But the priced rounds keep printing higher.

All together, the revaluation gives the investment portfolio a mark of $3.9B against a book of $2.1B, an uplift of $1.8B over what the accounts alone would have you believe.

The Core Business

A mature Korean telco at 81% 5G handset penetration deserves a pedestrian multiple. Wireless capex fell 42% in FY2025. The 5G build is done. D&A continues to charge the income statement above maintenance capex, which means the cash generation is understated. AI Data Centre revenue in Q1 2026 was $89M quarterly, up 28.8% YoY but down 17.6% QoQ from the Q4 2025 Pangyo step-up. Annualised that is $356M of run rate on the operational footprint of Gasan and Pangyo, which now sits at its capacity ceiling. KEPCO power-connection lead times of 12+ months and MOTIE’s metro-build restrictions mean SKM’s existing operational MW are scarce assets in the highest-demand AI training market in Asia. The capacity-bound dynamic supports the $2.3B SOTP mark; new capacity from the announced FY26 Seoul-metro build will not contribute revenue until 2028.

I value the core at 4.5x Korean peer median EV/EBITDA, which gives $16.2B. The DCF on the same business gives c. $20B; I use it as a sense-check rather than the central, because for a mature Korean telco the comp set has more gravity than a single-firm five-year DCF.

Sum Of The Parts

SKM appears to be trading at a 15% discount to a defensible sum-of-the-parts. A Korean telco with a nice AI portfolio at a discount.

Korean Sell-Side Coverage

Korean sell-side coverage is split. Spot ADR is $39.08 for reference.

Telco equity analysts do not think in revenue multiples. They think in cash flow forecasts and have not looked at SKM in the right manner. No better example is Goldman Sachs, who in their Q1 2026 report reiterating their sell opinion and do not once mention any of the investments made and give no attribution to them in their model.

Worth noting that the historical Korean discount on holding-company structures has been narrowing materially. The Commercial Code Amendment that came into force on 22 July 2025 extended directors’ fiduciary duty explicitly to shareholders, not just the company. President Lee Jae Myung’s Value-Up programme is the first serious legislative attack on the chaebol governance regime that produced the discount. The KOSPI is up roughly 80% year-to-date 2026. The “30-50% NASPERS-style discount” base case assumed in older SOTP work on Korean holdcos is anchored to a regime that is being actively dismantled. SKM benefits structurally regardless of the Anthropic outcome.

That is the SK Telecom story. A telecom company, a set of investments worth meaningfully more than their book values, and a 15% discount to intrinsic value.

Part 3: About that other investment

Oh. One more thing.

The sum-of-the-parts valuation of SK Telecom more than justifies its current valuation. However, I left one of their investments aside.

In August 2023, SK Telecom invested $100M in Anthropic and acquired (their language at the time) a “2%” stake. Anthropic was being priced at around $5B at the time.

This is what you get for free on top, if you buy shares in SK Telecom.

Since the investment, Anthropic has raised quite a lot of capital.

The ownership question is where this gets interesting, and where almost every analyst who has touched it has gone wrong.

The market consensus puts SKM’s Anthropic stake at approximately 0.30% on a fully diluted basis. This figure is cited in English-language press, picked up from a loosely worded SKM disclosure, and propagated without verification.

It is wrong, and the error is in the denominator.

The Correct Chain

Page 441 to 442 of the 2025 Business Report (Korean edition, 442 pages, not the 16-page English summary) discloses 3,703,141 shares at a 0.7% stake as of year-end 2024 (opening balance for FY25), with a carrying value of $132M at the disclosure FX rate.

Korea Investment & Securities analyst Kim Jung-chan stated in January 2026 that SKM held 0.7% of Anthropic per the H1 2025 semi-annual DART filing. He is a covered analyst at a tier-one Korean broker quoting a primary regulatory filing.

PitchBook reports 1.47B fully diluted Anthropic shares. That figure includes authorised-but-unissued option pools that have never been granted. Those do not dilute SKM economically.

The treasury stock method, mandatory for public-company diluted EPS, only counts options to the extent they are in the money and net of strike. Reserved-but-ungranted pools are excluded.

Anthropic’s auditor Samil PwC switched from basic to fully-diluted presentation between the H1 and FY 2025 disclosures. The economic reality, the share count, did not change. The disclosure convention did.

Working Forward To Today

Starting from 0.7% basic at H1 2025, threading through:

• Series F (September 2025), $13B raised at $183B post-money, approximately 7% dilution

• Series G (February 2026), $30B raised at $380B post-money, approximately 8% dilution

• Employee tender offer (February 2026), conversion of vested options

• Amazon $5B at $350B pre-money (20 April 2026), approximately 1.4% dilution

• Google $10B at $350B pre-money (24 April 2026), approximately 2.8% dilution

SKM’s basic ownership at 13 May 2026 is 0.54%.

Korean sell-side at Meritz Securities and Korea Investment & Securities both cite approximately 0.58% post-Series-G (pre April 2026 strategic equity). That corroborates the calculation within rounding. Any reader can trace this themselves in the filings. The 0.30% figure being used by the market is the wrong denominator on the wrong share count.

Google is contractually capped at 15% of Anthropic, with no voting rights and no board seat. So further dilution from this counterparty is structurally limited.

What 0.54% Of Anthropic Is Worth

Applying a 5% illiquidity discount (compressed from the standard 25% K-IFRS Level 3 given IPO expected within 6 to 9 months):

Carrying value on the SKM balance sheet today is $962M (auditor-declared, no methodology disclosed). Cost was $100M. Already 9.6x on book without any forward mark.

The Implied Anthropic Valuation At The Current SKM Price

At central marks for every other asset, SKM’s ex Anthropic equity comes to $17.7B against a current market capitalisation of $15.0B. The premium of ex Anthropic equity over market cap is $2.7B, or 18%.

At the current price, the market is giving you the Anthropic stake for free. Or better than free.

Only at core EV/EBITDA multiples below the Korean peer median does the market start to credit any Anthropic value at all. At the actual Korean peer median multiple of 4.5x, the market prices Anthropic at less than zero. The deal-of-the-century framing is not hyperbole. It is arithmetic.

The Full Picture

Total SKM equity at various Anthropic outcomes, with the implied ADR price at each:

At every Anthropic valuation at or above the priced Series G, SKM is cheap. The only scenario in which it is not cheap is one in which Anthropic is overvalued at $380B. You are welcome to take that view. It requires believing that a company doing $45B ARR, growing 5x YTD, with Jamie Dimon on stage, ten gigawatts of committed compute, and a cybersecurity model the White House has cleared for federal use, is worth less than $380B. Good luck.

Claude is still wearing that foppish scarf and it is mid-May on the Mediterranean. I forgive him.

You can see the more detailed and less flippant model that underpins this analysis here.

Subscribe now

Risks

The SK Telecom business risks are largely irrelevant given the entire upside thesis sits on Anthropic. The risks to my $4 trillion Anthropic valuation are laid out previously.

It is worth noting the additional risks of the proxy in between, the Korean holding company. Such structures have historically traded at a discount due to the opaque governance often involved, where the interests of the owners are not always aligned with the interests of the shareholders. Though this is not exclusive to Korea, investors in NASPERS and their titanic success in investing in Tencent have pulled their hair out, as much of the proceeds remain within NASPERS, which has continued to incinerate shareholder value in other investments rather than return the money. That risk is real. Yet the NASPERS comparison is perhaps unfair, as firstly, SKM is not run by a lunatic, and secondly the other investments shown by SK Telecom have also performed extremely well; but the transmission of the proceeds of an Anthropic IPO to shareholders should not be ignored.

The other main risk is the opacity of the ownership stake in Anthropic. It is far from clear through the company’s own communications, and the most compelling evidence is buried deep within Korean-language reports. The value ascribed in this analysis could be off by 20%, and shareholders are unlikely to receive much clarity going forward. Not only through cautious disclosures from SK Telecom but also Anthropic’s insistence that shareholders do not share proprietary data.

Catalysts

Four catalysts for price discovery over the next 6 to 9 months. The first three are information events that should help clarify the thesis. Only the fourth, the IPO, provides full mark-to-market resolution.

1. May ARR print (expected mid-June). $50 to 60B confirms trajectory. $65B+ validates the central case as conservative. Anything below $50B requires reweighting the probability distribution and compressing the central from $4T to $1.5 to 2T.

2. Mythos broader release. Currently restricted to JPM, Goldman, and a small set of partners. Treasury and Fed engagement is active. A broader rollout, with announced enterprise contracts at scale, repositions Anthropic from frontier AI lab to systemically important infrastructure. The Mozilla disclosure (271 vulnerabilities surfaced in Firefox in a single day) is the technical proof-point already in the public domain.

3. Series H pricing. Bloomberg and TechCrunch have the active round in the $850 to 900B range and imminent. The pricing sets the floor for IPO discussion. A close above $900B collapses any remaining argument that Anthropic at trillion-plus is speculative.

4. IPO. Goldman, JPM and Morgan Stanley are named as underwriters. Wilson Sonsini S-1 filing rumoured Q3. Target window Q4 2026 with October the earliest. On IPO the K-IFRS Level 3 illiquidity discount collapses to zero and the SKM stake becomes a mark-to-market liquid public security overnight. That is the one event that resolves the thesis.

This document is an independent investment analysis prepared for personal use. It does not constitute investment advice or a recommendation to buy or sell any security. The author is not a registered investment advisor. All sources are publicly available. Conduct your own research.

SEE THE MAIN REPORT HERE

Subscribe now

20 March 2026 | noseyparker.org

DISCLAIMER: The author holds a short position in Alaska Air Group (ALK) and stands to profit from a decline in the share price. This document should be read in conjunction with the main report dated 18 March 2026. It represents analysis of publicly available information and constitutes opinion, not investment advice.

ALK has known since at least 2022 that its loyalty platform has a critical cyber vulnerability.

Thousands of its most loyal and lucrative customers have had their accounts drained by hackers. Alaska’s consistent response is to blame the victim’s password hygiene, restore the stolen points as a “one time courtesy,” and permanently restrict the victim’s account.

Today the exact unremediated vulnerability is unveiled, along with the organised crime money trail that surrounds it.

For investors, the loyalty programme -- valued by management in December 2024 at $12 billion -- has lost its financial controls as a result of this flaw.

The main report released simultaneously today should be consulted for evidentiary precision and sourced evidence. What follows is a precis of that investigation.

I shall start with repeating the spark of this all.

Act 1: The LinkedIn Spark

In August 2025, the author saw a LinkedIn post about an Alaska Airlines member having his account drained of loyalty points and subsequently replenished. Others responded to report the same experience.

Three questions in this posting bothered the author:

1. Why can Alaska not block these clearly fraudulent bookings and stop the thefts?

2. What purpose could the mandatory telephone PIN booking system, forced upon the victim, serve?

3. Why would the fraudulent traveller take the staggering risk of passing through the most secure environments conceivable, under his real identity, for a few transient hours of comfort?

The questions were perplexing enough to launch an investigation.

Act 2: 425 Thefts Gathered

The investigation found 425 victims. Each sourced and archived.

The cases were found predominantly in niche special interest forums:

About 30% shared the quantum of their loss. The average was approximately 220,000 miles.

Act 3: Mode of Theft

The thefts almost universally displayed the same characteristics. The pattern has been consistent since 2022:

• Accounts were compromised to book flights within days of travel.

• Flights were on partner airlines.

• Members’ notification emails were suppressed.

• In 96% of cases, the victim discovered the theft by reviewing their own account, and then informed Alaska.

Act 4: Alaska’s Response Policy

Alaska’s responses were so consistent that they must be the product of a policy, also unchanged since 2022.

Alaska reimburses the stolen points as a “one time courtesy.” This language leaves the victim in no doubt who Alaska blames.

Victims are then barred from booking award travel online. They must telephone Alaska, verbally relay a PIN, which unlocks their account for online booking for one hour.

Act 5: The Industry Backdrop

A snapshot was taken with controlled search terms to confirm what a cursory internet search shows indisputably: Alaska is suffering from loyalty account theft at a rate unmatched by any peer airline.

`

Act 6: Tip of an Iceberg

The 425 victims are indisputably a fraction of the actual total. Few people post publicly about a largely remediated matter that carries some embarrassment.

The USCardForum.com statistics were striking. A Chinese-language forum with 12,000 weekly visitors, where fewer than 5% of posts relate to Alaska, produced 48 victims willing to come forward. Extrapolations are not definitive, but hacked accounts in 2025 are indisputably in the thousands.

Act 7: The Criminal Marketplace

The miles are sold for extraordinary prices on the black market.

To take an admittedly extreme example: a $35,000 family-of-four trip to Barcelona can be had for less than $200.

The main report offers a host of insights from six sellers identified during the investigation (though their integrity must be in question). All were found with a Google search that located them openly on Facebook.

Highlights:

• “Ernest” only sells accounts that have been dormant for more than three years.

• “Akis” has an inventory of 3,000 accounts ready to go.

• “Asad” says his six-man operation sells 50 to 60 Alaska accounts a week.

The revelation of their respective payment coordinates took the investigation upon a new path.

Act 8: The Money Trail

The various ETH and BTC wallets proffered for payment opened a new thread. Some went straight to exchanges. Others entered a river of apparent illicit capital. Two addresses connected with accounts through which over $759 million has flowed to KYC-regulated exchanges.

This entire network appears to be a flurry of subpoenas away from dismantlement. The lack of caution suggests the impunity under which these operators work.

Act 9: Law Enforcement

Instinctively, one would expect an airline beset by attacks of this nature to pour resources into facilitating arrests and publicise them widely as a deterrent. Victim accounts suggest the opposite.

The State of Alaska Attorney General recently challenged Alaska on its reticence to assist law enforcement. Generously, one could call it discretion.

Act 10: A Track Record of Discretion

The main report details half a dozen instances in Alaska’s filings where management reveals a notable predilection against disclosure. The subject of hacked member accounts has been met with the same silence.

There is no doubt management are aware of these thefts. The VP of Loyalty stated specifically that this had “visibility all the way up to our CEO.” The local press at Fox, Kiro7, and the Seattle Times have separately investigated this matter.

If it is implausible that management has no awareness of the thefts, it is equally implausible that they have no awareness of the PIN lock system that follows an account compromise. The encumbrance on their most loyal customers is such that the CEO must have asked questions, and the additional cost must have drawn the CFO’s attention.

Act 11: The Riddle of the PIN Lock

Yet there it remains. Why?

Consider the established customer service policy after a theft. Miles are restored as a “one time courtesy.” Those very words attribute fault to the victim. Customer service representatives comment upon the victim’s password hygiene when the loss is reported.

Yet the PIN lock is incongruent with a password defence. The compromised password has been changed by the victim upon discovery, so the password is no longer the issue.

Logic dictates that the vulnerability is not at the password.

Act 12: The Logical Breakdown

Through the “one time courtesy” refund, victims are led to believe their password was compromised and that they are at fault. Yet after discovery the password is changed. If the vulnerability were at the password, the PIN lock would be nonsensical: the fresh password would suffice.

The PIN lock exists because changing the password does not remove the attacker.

Act 13: The Proof at Your Desk

You can prove this yourself right now.

1. Log in to your Alaska account on two separate browsers.

2. Change the password on one browser.

3. Refresh the other.

You will still be logged in. Changing the password does not invalidate existing sessions. Anyone who holds a session cookie retains full access to the account regardless of how many password changes are made.

The session cookie does contain an expiry date. Unfortunately, that expiry is perpetually refreshed with each use, maintaining the session for as long as the attacker remains active.

If your account has been compromised and miles stolen, there is no point returning them. They will be stolen again, because the attacker’s access persists. There are two solutions: fix the system, or unplug the victim from it.

Alaska chose the latter. The fact that Alaska does not invalidate this token overwhelmingly suggests it cannot.

Act 14: Passwords Not Required

The most likely explanation is that the session token is stateless -- it is self-contained and not checked against a server-side session store. Without a revocation list (which Alaska does not appear to maintain), the server has no mechanism to invalidate it.

One could argue that the original sin remains: that the password was compromised and the victim is culpable. But that session cookie, the one that grants full account access, is readable by any script on the page.

The cookie is not marked HttpOnly. Nor does it carry other standard defences against exfiltration.

The HttpOnly flag has been a cornerstone of web security since 2008. Without it, every third-party script on the page can read the cookie’s contents, and most Chrome extensions have broad enough permissions to do the same. With that cookie, they have your account.

This investigation observed 14 outside companies with active cookies on Alaska’s authenticated pages. Those companies need not be malicious. If any one of them suffered a breach at any time, that breach could inject code to sweep up the session cookies of every Alaska member who logged in during the window of compromise.

It is not only these firms. Any attacker who can inject code into one of those 14 third parties, or into Alaska’s own site, gains the same access.

Even if a member’s password security is beyond reproach, the missing HttpOnly flag means most Chrome extensions with standard permissions can silently read the session cookie.

The full report contains further significant weaknesses in the architecture, but to bring it back to plain language: if anyone obtains this cookie, they gain a session with full account privileges. In testing, no secondary authentication was required to change email, phone number, or passport details. There appears to be no way to revoke their access. They follow your travels at every step.

Act 15: What Management Knows

It is inconceivable that management are unaware of this weakness after four years. For technical reasons described above, a compromised session cookie persists as long as the attacker keeps it alive. Hence the PIN lock: it is not a security measure for the customer but a containment measure for the vulnerability.

Instead of remediating the flaw, Alaska has blamed the customers, inconvenienced them, and sanctioned their accounts.

If management are genuinely unaware, they must provide a vigorous explanation as to how.

Act 16: The Bridge to Price

The main report identifies nine distinct factors detailing the impact upon the crown jewel of Alaska Air Group. Here it suffices to assert that these findings should have seismic implications for how consumers, partners, lenders, and investors value Atmos Rewards.

Act 17: The LinkedIn Mysteries Answered

The questions that drew the author into this investigation can now be answered with confidence.

Mystery 1: Why can Alaska not block these clearly fraudulent bookings and stop the thefts?

Because its systems cannot see them. The architecture that processes award redemptions has no fraud detection capability.

Mystery 2: What purpose could the mandatory telephone PIN booking forced upon the victim serve?

Compromised accounts remain compromised for as long as the attacker maintains an active session. Without the subterfuge of blaming the victim and forcing them into the telephone system, replenished accounts would be drained again immediately.

Mystery 3: Why would the fraudulent traveller take on such staggering risks?

Because flat-bed flights almost anywhere in the world are cheaper than a taxi to the airport. And law enforcement are, as a matter of policy, not informed of the thefts.

425 victims. An unknown multiple beyond.

A vulnerability unchanged for at least four years.

A front door that remains, today, wide open.

Subscribe now

The full evidentiary report is available at https://noseyparker.pages.dev/baked-alaska

On a morning in early 2022, JPMorgan’s marketing team launched their first campaign to Frank’s customer base. The bank had paid $175 million for access to 4.25 million verified student users - young adults at the start of their financial lives, complete with names, emails, phone numbers, and exhaustive financial aid data. All of it, the seller had sworn, “fully consented for JPM marketing.”

The first mass email went out.

Seventy-six percent of emails bounced back undeliverable.

Only 1% ever got opened.

There was trouble in the State of Denmark.

Likely within hours of JPMorgan double-checking their own methodology, the first member of the bank would have pondered the alternative explanation, which soon would be irresistible. These weren’t Frank’s customers.

THE PRODIGY

Charlie Javice was a startup prodigy devoting her life to those who needed help. She’d made Forbes’ 30 Under 30. The Byzantine forms that kept needy students from getting their due financial aid were being obliterated by Javice’s visionary brilliance. Form filling tumbled from hours to seven minutes.

On Frank’s website, they crowed about “Why 4.25 million students choose Frank” - a stunning number for a five-year-old company.

JPMorgan took note and wanted access to the banking world’s dream: an annual deluge of nascent adults, whose background and trajectory would afford a time and insight advantage no rival could hope to overcome.

Frank’s pitch deck was even dreamier for the bankers: 4.25 million verified users with not only name, email and phone, but every exhaustive numeric detail required in a financial aid application, “all fully consented for JPM marketing.”

JPM salivated. They moved fast.

The sale agreement, sworn to by Ms Javice, let no doubt remain: “4.265 million users” were going to JPM.

$175 million was going to the owners of Frank.

If you don’t have 4.25 million users, what do you do? If you’re a Forbes 30 Under 30, the answer is obvious: you manufacture them.

THE HANDSHAKE

In a sales process, the seller is taken at their word because they must show proof before any sale is finished by a process called due diligence.

There was not much JPM cared about in this business generally, so of course they focused on making sure this checked out. The seller should be cautious in just showing it, or else the buyer now has the list. A sampling test should suffice.

Yet Miss Javice showed she is made of something special. Her ethical fortress was her superpower and she immediately objected to her users’ data privacy being irreversibly defiled by JPM - her principles were not for sale. She compromised with an independent third party validating the list on JPM’s behalf.

The deal closed and Javice’s hours and intensity rose even further. No doubt she saw the scale of humanity she could assist with JPM’s resources. Architecting her future impact - and the light it would shine for the downtrodden - sadly meant she could not spare the time to fully hand over thjat list to JPM to use in their first email testing campaign for four months.

The slime of commerce was her price to pay to help others.

THE PROBLEM

How could this have happened?

Well, that 4.25 million sounded so wholesome and virtuous they ensured it was in all their marketing materials when JPM approached. Indeed, when they promised that these 4.25 million emails had incredibly granular information about earnings - the house of cards should have collapsed.

If you don’t have 4.25 million users, what do you do? If you’re a Forbes 30 Under 30, the answer is obvious: you manufacture them.

The truth was Frank had somewhere around 350,000 users. If they didn’t notice the jump from 350k to 4.25 million, well - as Javice would later argue - that’s on them.

THE FABRICATION

Step One: The Data Scientist

Ms. Javice hired a data scientist contractor.

4.25 million rows and columns of data were populated with an elegant snippet of Python. I have no doubt it looked magnificent. With FAFSA data on every line. Sadly it was all gibberish, but enough to slide past the due diligence.

Step Two: The Email Lists

But this problem was not likely going away. When faced with what they would have to give to JPM after the deal - well, the answer is obvious. Just buy a leads list! $100k - how would JPM ever know!

Well, everyone remembers buying their first leads list - as they are always total rubbish. So then they bought another one, bodged them together, made up all the other columns. Did it!

Step Three: The Slack Channel

Now, the planning and execution of buying random email addresses and pretending they were Frank customers needed to be carefully and clandestinely planned. Where best to do that planning other than the company’s Slack and email accounts? They even made a ‘4.25m-list’ Slack channel to discuss the purchase and how to fill empty rows with nonsense.

Imagine the delusions of brilliance to think 1) this would work and 2) to do it on your own software. But Javice knew she did not have to take the trouble to do it on another software. Not worth the energy. Because she could just hit delete after and nobody would ever know...

A quick hit of ‘restore’ on the deleted slack channel and there, served on a silver platter, the intricacies of the plot…

THE SLEIGHT OF HAND

The independent third party validation apparently worked. Whatever they sent - whether it was the fully synthetic dataset, the purchased email list hybrid, or some combination - it passed muster.

Partly, this was JPMorgan’s fault. Their due diligence was shoddy, but one doesn’t expect criminal fraud.

The signed purchase agreement stated: 4.265 million users.

Frank had 350,000.

The deal closed. Javice and her co-founders walked away with $175 million. JPMorgan now owned Frank.

THE UNRAVELING

The first mass email went out.

With 76% of emails bouncing back and only 1% opened, even the most forgiving at JPM smelled a rat. 

A quick hit of ‘restore’ on the deleted slack channel and there, served on a silver platter, the intricacies of the plot: the refusal of Frank’s engineers to commit fraud, the purchase of a list of 4.25 million student emails, and - I like to think - her telling the others, “It’s ok, we can just delete this channel later...”

Surely now the jig is up. But Ms. Javice had four years’ worth of fibbing in her. One must grudgingly admire her ingenuity over the years:

Pre-investigation (late 2021 – early 2022)

1. “Students don’t check their .edu emails over the holidays.”

2. “University spam filters are blocking your campaign.”

3. “Some addresses were anonymised for privacy compliance.”

4. “It’s just a data-migration glitch.”

5. “We counted every student we reached - not only registered users.”

As JPMorgan’s internal probe began (mid-2022)

6. “The dataset was a privacy-safe sample, not a fake list.”

7. “The export script corrupted addresses when migrating from Frank’s system to JPMorgan’s.”

8. “JPMorgan knew the numbers were illustrative.”

9. “You were buying our reach, not a spreadsheet.”

10. “We built something that touched millions of students - that’s what mattered.”

After JPMorgan sued (late 2022)

11. “This is buyer’s remorse dressed up as fraud.”

12. “At worst, we disagreed on the definition of ‘user.’”

13. “I was just focused on helping students, not selling data.”

14. “The 4.25 million users number was known by them to be aspirational.”

15. “JPMorgan wanted me - my brand, my access, my credibility with Gen Z students.”

16. “No one was harmed; the bank still owns the tech.”

17. “I didn’t kill anyone.”

18. “JPM can easily afford it. Same as $50 to the average household.”

19. “If they didn’t notice the jump from 350k to 4.25m that happened the previous year, that’s on them.”

Criminal proceedings (2023 – 2025)

20. “I never intended to deceive anyone.”

21. “We acted under extreme pressure to close a fast-moving deal.”

22. “I didn’t code or generate the data myself - I trusted my team and contractor.”

23. “This was a civil misunderstanding, not a crime.”

24. “JPMorgan turned a blind eye - they didn’t care about the number.”

25. “They wanted a story, not statistics.”

26. “If I were a man, this would have been settled quietly.”

27. “I was a first-time founder who made mistakes.”

Post-conviction (sentencing / appeal period 2025)

28. “I’m focusing on my health and starting a family.”

29. “I still believe the trial was unfair.”

30. “They punished ambition - not fraud.”

There is zero chance that this girl has a broad circle of friends and neighbours who are sometimes short that week and cannot afford their Pilates lesson. In Miami Beach.

THE DEFENCE

Now you must be thinking this legal defence is sounding expensive? You’d be right. $115 million and counting! And somehow, JPM are on the hook for it all - so she is spending like a sailor in port!

When questioned, an ungodly series of lies and excuses spewed forth that is still going to this day.

Well, she was found guilty and sentencing came. Her procedural appeals have been without number. I kid you not, there was a precedent question for whether an ankle tag had to be removed as the person on bail had decided to become a Pilates instructor and it would stop her earning.

THE LETTERS

Well, that $115 million legal defence pulled out letters from a hundred different people saying what a good egg she was, so the sentence would be lenient.

If this was the sum that towerblocks of lawyers could dig up about her life - she must have been pretty terrible. Let’s have a look:

“Volunteering at a soup kitchen when she was nine years old and soon took over organising the evening service.” —Family Friend

That didn’t happen.

“Helped fund and later helped build an orphanage in Thailand.” —Family Friend

Part of her $6,000 holidays, it turns out. To wit, when I worked on a building site when I was 17, I did not build anything. I could not so much as saw a piece of wood straight. I moved vast quantities of sand and rubbish around with hands screaming with pain.

“Gives free Pilates lessons to friends and neighbours when they can’t afford the studio.” —Miami friend

There is zero chance that this girl has a broad circle of friends and neighbours who are sometimes short that week and cannot afford their Pilates lesson. In Miami Beach. Rocking up to the class with out-turned empty pockets hoping the just-started Pilates instructor would take pity and let them into the studio.

“Both her doormen said she was polite and respectful.”

Pretty sure that was not pushed along by her team. And is a veiled insult. And not written by a doorman. The implication is she refrains from screaming at her doormen.

“Frank was about the aid! Not the money.” —Lawyer

Weird that she was selling the whole company so the students would be bombarded with usurious loans. Let us not forget, the whole transaction was for Frank to sell students’ personal data to a commercial bank who had profit-maximising aims on naive students at the start of their journey. All of her customers were deceived as to her purposes.

Well… traditionally you go in the slammer once convicted.

THE SENTENCE

The judge looked at this manipulative, delusional fraudster and agreed - ”You are a good person... but you did a bad thing.” Quite a few, I’d say.

Well, the judge could only convict. Seven years. Ouch.

He was not moved enough by her contrition in her final statement:

“Most painfully, I have lost time. At 29, I put my life on hold, including my hopes of becoming a young mother. My mum is now approaching the age at which my grandmother and great-grandmother were diagnosed with pancreatic cancer.”

What kind of lunatic thinks she is uniquely suffering the travesty of age... at 32.

“I do not ask for forgiveness or to erase the seriousness of the past,”

She ended, just before immediately appealing the guilty verdict!

THE APPEAL

How can she carry on? Here was her rationale for a mis-trial and the basis of her (JPM-funded) appeal. Four things:

1. As her right-hand man was co-defendant, he prejudiced her trial by saying she, as CEO, was leading the fraud.

2. The judge recklessly ignored the FAQ page. In 2020 they changed the 350k to 4.5m users on the front page. The FAQ page was not updated. Therefore JPM should have seen that and known 4.5m was a sworn lie.

3. Something about “Blind Eye”, which I don’t understand, but seems to be founded on the judge changing five words at the wrong time.

4. The prosecution’s data expert who pulled the $100k leads lists to shreds did not actually do the analysis personally, instead relying on his team of lackys.

So… nothing. Well, doesn’t help as - traditionally - you go in the slammer once convicted.

THE BAIL

But don’t fear, our protagonist Ms Javice has been allowed on bail - on ‘moral grounds’.

The reason? Because she wants to try for a baby! Her fertility doctor wrote in to plead for the humanity.

Crumbs - you do a crime, get caught, sent to the nick. Nothing else came out in the public record.

And so Charlie Javice, convicted fraudster, remains in Miami Beach, teaching Pilates, planning motherhood, and spending JPMorgan’s money on appeals.

THE BOTTOM LINE

JPMorgan paid $175 million for a company whose actual user base was 350,000 - and whose “verified customer list” turned out to be worth exactly $105,000 at wholesale.

They are now funding the $115 million legal defence of the woman who sold it to them.

Frank’s homepage, before it was taken down, asked: “Why do 4.25 million students choose Frank?”

The answer, it turned out, was simple: they didn’t.

This week on Nosey Parker: A case that’s still unfolding, with millions at stake. Subscribe to be first in line.