1) Headline is very accurate. Anthropic should be #1 company lol, I think around 80% of fortune 100 already use them. It's funny how much media coverage job losses from AI and SaaSpocalype gets anthropic, but people yell "AI Bubble" when they see its valuation.
Bears sound smart, bulls make money
2) My favorite part of the article is this -
“I believe that OpenAI, as many have properly criticized them for, is too distracted. They have too much emphasis on the consumer, pursuing ventures like acquiring TBPN and literally making a phone. (Like, what? Do you not realize that if you reach AGI first, you can make a phone way faster? Lmao)"
THANK GOD SOMEONE SAID IT LMAOOO
The compute advantage is real but Jesus Christ, everyone but Claude is huffing some glue out there. The biggest opportunity in the history of business and:
- Anthropic integrated AI into excel before Microsoft
- Claude cowork and Claude Code was a lot faster than OpenAI, and funny enough I don't even think OpenAI has a rebuttal to Co-work to this day lol
- Google is giving compute to GCP consumers like crazy instead of hogging it to build AGI
- Wtf happened to perplexity, mistral, etc????
- Meta Superintelligence Labs is too behind. Zuck is a terrible leader and swipes the credit card for anything he can't develop himself. Metaverse total disaster lol
- RSI will absolutely take out more competition as time goes on. Frankly, if OpenAI follows through with their shenanigans, their revenue will get leapfrogged so badly a consortium of industry players will have to come together so Anthropic doesn't have a monopoly. Economies of scale are everything here, and Anthropic's 300% faster growth and projected profitability in 2027 vs 2030 is a HUGE, MASSIVE deal.
3) Playing devil's advocate, downsides are that Anthropic:
A) Is a PBC like OpenAI, so profit maximization might take a backseat sometimes
B) The talks Dario gives constantly about replacing jobs are not going to sit well when large scale layoffs actually happen. Not good PR strategy at all
But tbh, none of this will matter. Why?
GLOBAL KNOWLEDGE WORKERS ARE PAID 70T / year!!!! TAM IS INFINITE!!!!
I would be totally shocked if AI companies didn't capture at least 10% of that by 2030. Anthropic is now the biggest player and the leapfrogging will continue for a bit, so 7 times lets say 25% market share on the low end is what, 1.75 trillion / year? That's my bear case. More optimistically could end end mid single to high trillions per year in revenue, and I think margins after scaling will end up pretty decently as well.
Current valuation is a joke. I suspect Aschenbrenner might be fiddling around with Anthropic funding rounds or something behind the scenes tbh, it's one of the best risk rewards in equities ive seen in a very long time. I would say 50-50 SK Telecom doubles in a year, I can't really even see the downside risk in that time period unless Anthropic stops shipping products or something, which would be the opposite of what they've been doing for past 4 years.
The underdog is now the favorite. If Dario took Anthropic from 0 to $900,000,000 in 5 years vs. an infinitely better capitalized competitive market (ex. OpenAI), the beatdown he'll put on the rest of the industry will be insane. Management is soooo underrated in investing and a big reason Lisa Su provided 100x returns at AMD for a decade, Lip Bu Tan got Intel 5x'd in a year (ok ok Gelsinger gave him the foundation but you get the point).
forgot one last thing: Every other stock there's some 1% tail risk AGI / ASII erodes moat. From Lumentum to INTC to Nvidia, what if AI creates a better version of InP or designs CPUs way better than humans?
GPU sales hurt by optical or quantum computing advances that some ASI created? Leather jacket man not gonna be happy. This kinda keeps me up at night somedays lol
So much of terminal value is far out into the future for AI stocks. Especially for fabless companies like ARM / AMD, the entire moat is intelligence X economies of scale. If AI comes for SaaS, there's going to be some guarantee AI EDA tool at some point in the future for hardware companies.
For model companies there's literally no replacement, in some ways it's actually the picks and shovels play of the future while being the complete opposite today. Very crazy idea today but I feel like one days this is gonna get vindicated
I think it might be from their Korean annual report. The thing is they use a fully diluted valuation that includes options that are authorized but not granted, and those are not economically dilutive. The real FD share count is somewhere in the 686 million range, which points to 0.54%. 0.3% is their official disclosure but isn't economic truth
It is what perplexity with got 5.4 said as well. So at 1 Trillion valuation it would be 3 Billion of SKMs market cap or 20% - since SKM doubled over the last year it is likely more than priced in. But did not read your article yet... :)
I think it might be from their Korean annual report. The thing is they use a fully diluted valuation that includestloptions that are authorized but not granted, and those are not economically dilutive. The real FD share count is somewhere in the 686 million range, which points to 0.54%. 0.3% is their official disclosure but isn't economic truth
Hey Jason, loved the Anthropic piece and wholeheartedly agree with the business analysis. Wanted to push back on one piece of the SKM math.
You framed this as Anthropic exposure rather than an SKM pitch, but the option math implicitly bets on SKM trading expensive vs Korean peers. KT and LG U+ average around 3.6x 2027 EBITDA. SKM's $20.07B EV on $3.76B 2027 EBITDA puts the implied core multiple at 5.34x assuming $0 of Anthropic is priced in today. That's a 40%+ premium to peers baked into your breakeven.
Walking through what changes if you assume the market is already pricing some Anthropic in. Take $624B as a defensible midpoint given the BofA exhibit you reproduced at $773B to $1.5T.
At $624B implicit, the stake is worth 0.54% × $624B = $3.37B today. Backing that out of the $20.07B EV leaves $16.70B for the rest of the business, which is 4.44x 2027 EBITDA (still above peer 3.6x but more reasonable).
For the option to break even at the $6 entry, SKM needs to hit $46 at expiration. At 383M shares, that requires roughly $17.6B market cap, or $23.5B EV. Subtract the $16.70B core and the stake needs to be worth about $6.8B at expiration, which back-solves to approximately $1.25T of Anthropic IPO mark, not the $816B you cited.
Still a reasonable bet, but a much higher bar than the piece implied.
Separate question: why do you think the market isn't already pricing the Anthropic stake closer to the $900B last private mark in SKM today? Your framework implicitly requires the market to value Anthropic well below the most recent round, which seems to cut against the BofA exhibit you cited.
One other thing I'd flag: how are you thinking about the path where Anthropic doesn't IPO before Jan 2027 expiry? They just closed a $30B private round at $900B, and there's no S-1 yet. Companies typically don't IPO 3 to 6 months after a mega private round, and even at a CoreWeave-fast S-1-to-trade timeline they'd need to file by October to print before expiration. If timing slips even one quarter, the option could expire worthless even if the Anthropic thesis is dead right.
Curious how you reconcile. Thanks again for all of the time put into this; the Anthropic business model analysis really resonated!
Wow you are good! I thought you only dive into semi but then see this!
My issue is I don't think Anthropic will be definitely the #1. It just keeps rotating. Now engineers are saying Codex is the best. Google is also catching up. Maybe Anthropic leaking their agent harness code on April's fool was their company's biggest failure??
I'll also be more careful on gross margin assumption. Let's assume Anthropic/ OpenAI/ Google are equally good, and they are all compute hungry monster. The demand supply will clearly be imbalance, and therefore giving neocloud and the whole compute supply chain more bargaining power. As long as the frontier model is not a monopoly situation, semi supply chain will get higher % of value creation. Semianalysis put it pretty well.
I see it as a bull case if Anthropic becomes monopoly and reaches singularity. Which I highly doubt to be true
What about the impact of wholesale transfer pricing? As their profit margins go to the moon their compute providers may decide to increase prices because anthropic can afford it
If frontier models converge toward long-horizon agentic tasks, then average context length per query grows faster than efficiency gains from next-gen hardware. NVIDIA GB300 NVL72 delivers 35x cost reduction, but that's per-token at fixed context. With 10x context expansion over the same period, net cost per query can rise, not fall.
Thanks for making this article, and bigger thanks for hearing out suggestion in chat :))
Did = / better than I could do for sure. Georgia tech 2nd year engineer vs Wharton last year diff hehehe
Oh and a few other thoughts:
1) Headline is very accurate. Anthropic should be #1 company lol, I think around 80% of fortune 100 already use them. It's funny how much media coverage job losses from AI and SaaSpocalype gets anthropic, but people yell "AI Bubble" when they see its valuation.
Bears sound smart, bulls make money
2) My favorite part of the article is this -
“I believe that OpenAI, as many have properly criticized them for, is too distracted. They have too much emphasis on the consumer, pursuing ventures like acquiring TBPN and literally making a phone. (Like, what? Do you not realize that if you reach AGI first, you can make a phone way faster? Lmao)"
THANK GOD SOMEONE SAID IT LMAOOO
The compute advantage is real but Jesus Christ, everyone but Claude is huffing some glue out there. The biggest opportunity in the history of business and:
- Anthropic integrated AI into excel before Microsoft
- Claude cowork and Claude Code was a lot faster than OpenAI, and funny enough I don't even think OpenAI has a rebuttal to Co-work to this day lol
- Google is giving compute to GCP consumers like crazy instead of hogging it to build AGI
- Wtf happened to perplexity, mistral, etc????
- Meta Superintelligence Labs is too behind. Zuck is a terrible leader and swipes the credit card for anything he can't develop himself. Metaverse total disaster lol
- RSI will absolutely take out more competition as time goes on. Frankly, if OpenAI follows through with their shenanigans, their revenue will get leapfrogged so badly a consortium of industry players will have to come together so Anthropic doesn't have a monopoly. Economies of scale are everything here, and Anthropic's 300% faster growth and projected profitability in 2027 vs 2030 is a HUGE, MASSIVE deal.
3) Playing devil's advocate, downsides are that Anthropic:
A) Is a PBC like OpenAI, so profit maximization might take a backseat sometimes
B) The talks Dario gives constantly about replacing jobs are not going to sit well when large scale layoffs actually happen. Not good PR strategy at all
But tbh, none of this will matter. Why?
GLOBAL KNOWLEDGE WORKERS ARE PAID 70T / year!!!! TAM IS INFINITE!!!!
I would be totally shocked if AI companies didn't capture at least 10% of that by 2030. Anthropic is now the biggest player and the leapfrogging will continue for a bit, so 7 times lets say 25% market share on the low end is what, 1.75 trillion / year? That's my bear case. More optimistically could end end mid single to high trillions per year in revenue, and I think margins after scaling will end up pretty decently as well.
Current valuation is a joke. I suspect Aschenbrenner might be fiddling around with Anthropic funding rounds or something behind the scenes tbh, it's one of the best risk rewards in equities ive seen in a very long time. I would say 50-50 SK Telecom doubles in a year, I can't really even see the downside risk in that time period unless Anthropic stops shipping products or something, which would be the opposite of what they've been doing for past 4 years.
The underdog is now the favorite. If Dario took Anthropic from 0 to $900,000,000 in 5 years vs. an infinitely better capitalized competitive market (ex. OpenAI), the beatdown he'll put on the rest of the industry will be insane. Management is soooo underrated in investing and a big reason Lisa Su provided 100x returns at AMD for a decade, Lip Bu Tan got Intel 5x'd in a year (ok ok Gelsinger gave him the foundation but you get the point).
Haha yeah that's funny we basically think the exact same way
Yup :)
forgot one last thing: Every other stock there's some 1% tail risk AGI / ASII erodes moat. From Lumentum to INTC to Nvidia, what if AI creates a better version of InP or designs CPUs way better than humans?
GPU sales hurt by optical or quantum computing advances that some ASI created? Leather jacket man not gonna be happy. This kinda keeps me up at night somedays lol
So much of terminal value is far out into the future for AI stocks. Especially for fabless companies like ARM / AMD, the entire moat is intelligence X economies of scale. If AI comes for SaaS, there's going to be some guarantee AI EDA tool at some point in the future for hardware companies.
For model companies there's literally no replacement, in some ways it's actually the picks and shovels play of the future while being the complete opposite today. Very crazy idea today but I feel like one days this is gonna get vindicated
Again, I thought I was the only one that thought this. We literally think the exact same. lmfao
skm disclosed officially that they only holds 0.3%
I think it might be from their Korean annual report. The thing is they use a fully diluted valuation that includes options that are authorized but not granted, and those are not economically dilutive. The real FD share count is somewhere in the 686 million range, which points to 0.54%. 0.3% is their official disclosure but isn't economic truth
That makes sense. But the diverge between FD and OS shares is too large to believe 🤔 ……
can you lmk where they said this? thanks
It is what perplexity with got 5.4 said as well. So at 1 Trillion valuation it would be 3 Billion of SKMs market cap or 20% - since SKM doubled over the last year it is likely more than priced in. But did not read your article yet... :)
I think it might be from their Korean annual report. The thing is they use a fully diluted valuation that includestloptions that are authorized but not granted, and those are not economically dilutive. The real FD share count is somewhere in the 686 million range, which points to 0.54%. 0.3% is their official disclosure but isn't economic truth
Congrats to @aaaaaaaa for winning the giveaway
Hey Jason, loved the Anthropic piece and wholeheartedly agree with the business analysis. Wanted to push back on one piece of the SKM math.
You framed this as Anthropic exposure rather than an SKM pitch, but the option math implicitly bets on SKM trading expensive vs Korean peers. KT and LG U+ average around 3.6x 2027 EBITDA. SKM's $20.07B EV on $3.76B 2027 EBITDA puts the implied core multiple at 5.34x assuming $0 of Anthropic is priced in today. That's a 40%+ premium to peers baked into your breakeven.
Walking through what changes if you assume the market is already pricing some Anthropic in. Take $624B as a defensible midpoint given the BofA exhibit you reproduced at $773B to $1.5T.
At $624B implicit, the stake is worth 0.54% × $624B = $3.37B today. Backing that out of the $20.07B EV leaves $16.70B for the rest of the business, which is 4.44x 2027 EBITDA (still above peer 3.6x but more reasonable).
For the option to break even at the $6 entry, SKM needs to hit $46 at expiration. At 383M shares, that requires roughly $17.6B market cap, or $23.5B EV. Subtract the $16.70B core and the stake needs to be worth about $6.8B at expiration, which back-solves to approximately $1.25T of Anthropic IPO mark, not the $816B you cited.
Still a reasonable bet, but a much higher bar than the piece implied.
Separate question: why do you think the market isn't already pricing the Anthropic stake closer to the $900B last private mark in SKM today? Your framework implicitly requires the market to value Anthropic well below the most recent round, which seems to cut against the BofA exhibit you cited.
One other thing I'd flag: how are you thinking about the path where Anthropic doesn't IPO before Jan 2027 expiry? They just closed a $30B private round at $900B, and there's no S-1 yet. Companies typically don't IPO 3 to 6 months after a mega private round, and even at a CoreWeave-fast S-1-to-trade timeline they'd need to file by October to print before expiration. If timing slips even one quarter, the option could expire worthless even if the Anthropic thesis is dead right.
Curious how you reconcile. Thanks again for all of the time put into this; the Anthropic business model analysis really resonated!
How much ARR is it even possible for Anthropic to achieve this year given compute constraints?
Thanks for the great article Jason! Question: doesn't zoom have a bigger share of Anthropic? Wouldn't it be a better bet? Thanks!
Wow you are good! I thought you only dive into semi but then see this!
My issue is I don't think Anthropic will be definitely the #1. It just keeps rotating. Now engineers are saying Codex is the best. Google is also catching up. Maybe Anthropic leaking their agent harness code on April's fool was their company's biggest failure??
I'll also be more careful on gross margin assumption. Let's assume Anthropic/ OpenAI/ Google are equally good, and they are all compute hungry monster. The demand supply will clearly be imbalance, and therefore giving neocloud and the whole compute supply chain more bargaining power. As long as the frontier model is not a monopoly situation, semi supply chain will get higher % of value creation. Semianalysis put it pretty well.
I see it as a bull case if Anthropic becomes monopoly and reaches singularity. Which I highly doubt to be true
What about the impact of wholesale transfer pricing? As their profit margins go to the moon their compute providers may decide to increase prices because anthropic can afford it
If frontier models converge toward long-horizon agentic tasks, then average context length per query grows faster than efficiency gains from next-gen hardware. NVIDIA GB300 NVL72 delivers 35x cost reduction, but that's per-token at fixed context. With 10x context expansion over the same period, net cost per query can rise, not fall.