
Mek Metapon
1.9K posts

Mek Metapon
@MetaponA
ทวีตเกี่ยวกับการลงทุน เน้นหุ้นอเมริกาจ้า Facebook / Youtube: หุ้นเปลี่ยนโลก
















I’m convinced most $DUOL bears who keep repeating “AI translation will disrupt Duolingo” will eventually accept the opposite conclusion, but only after the stock price makes it socially safe to believe it. Because this business is becoming less about languages and more about learning. And learning isn’t a dictionary problem. It’s a habit formation problem. Habit formation is an optimization loop: streak mechanics, pacing, difficulty curves, motivation triggers, timing, retention, and the thousands of small choices that determine whether a learner stays five minutes longer, or shows up tomorrow. You don’t solve that with a better translator. You solve it with years of proprietary data and relentless iteration on what actually changes behavior. Most investors don’t update their understanding based on what the business is. They update based on what the price is doing. When $DUOL is down, the narrative is “no moat, AI replaces it.” When $DUOL is up, the narrative becomes “behavioral psychology, data moat and new verticals.” Same facts. Different price. Different understanding of the business. And that’s the real game: the psychology of staying loyal to your research, your thinking, and your conviction while the entire market is against you. It’s the hardest skill in investing. It’s what separates multibagger holders from megabagger holders. (Not investment advice. Green owls move in mysterious ways.)


BREAKING $AMD & $META Secret Hyperscaler🧵 @Meta Roadmap to 10s-100s+ GW Hyperscaler This thread will focus more on how META is going to scale this kind of AI Infrastructure by skipping upfront construction cost and cool. And how Dr. @LisaSu going to take AMD's shareholders to 20%+ Market share. Meta and funds managed by Blue Owl Capital formed a JV to develop and own the Hyperion data center (a massive ~2GW up to 5GW where @AMD is already a major supplier) AI-focused campus in Richland Parish, Louisiana, with total development costs around $27 billion). ~Blue Owl funds hold ~80% ownership; Meta retains 20%. ~Meta will lease the entire completed campus from the JV under operating lease agreements (4yr term with options to extend/renew) ~Meta received a one-time distribution of approximately $3 billion from the JV as part of the transaction setup. ~Meta effectively received ~$3 billion net upfront rather than putting money down. This creative financing helps Meta scale its AI infrastructure rapidly without fully burdening its balance sheet, especially given its massive ongoing CapEx (~$60–70B+ annually in recent guidance). The JV owns the campus, Meta leases it back (initial 4-year term + extensions), and future rent payments support the JV's returns/debt service. This is where most analysts misunderstood the clever structure @finkd came up with Blue Owl. Or how this will significantly help $AMD boosting Revenue growth. $AMD should be able to generate $70B-$100B by FY 2026(read more below)! While $META did not disclose the rental detail, but we can expect 6-9% yield for META investment-grade or 7% in this case, which would imply around $2B annual rent obligation plus $1-$1.2B OpEx(for 1GW, or $4B for 5GW). Which is still going to be massively cheaper than renting compute from $MSFT, $AMZN, $GOOGL or roughly $20-$25B renting bill on 1B+ users on MetaAI owning $AMD EPYC Venice and MI455X or Helios rack which collaborated with $AMD. In case of Hyperion campus at 5GW, Blue Owl join venture will handle all the constructure, building, power, cooling and others, while $META will have to cover the cost of GPUs purchase or in this case hybrid of $AMD Helios and $NVDA Rubin Rack. Blue Owl gets higher rental rate and ownership of building, land. If $META owns everything, it will only be $1B-$4B of OpEx from 1GW to 5GW capacity. In this case, $META will pay $3B-$6B OpEx instead. The bonus is much much faster construction where Meta can gets it online very quickly, or as some of the space is already "going live" right now, where more units to go "live" in 2027-2028-2029-2030. We need an example, where you can understand the number a bit more clear. 5GW, at 42% allocation of prior chips to $AMD in 2025 and 58% to $NVDA. If we use the same allocation for 5GW, that would be around 2GW. $META is already a very large customer for $AMD, so I assume the highest discount at large multi-GW scale deployment as Dr. @LisaSu indicated throughout various events. 1GW Helios= $25B. This number could go up to $30-$33B per GW for slower/smaller deployment. 2GW will mean $50B revenue for $AMD. MI500 Racks will probably at 10-15% higher price tag. I will talk about that when we get there. Conclusion: This clever structure allows $META to scale AI Infrastructure much much faster than owning everything as it would be far more capital intensive. Think of $2B Rent/year for every 2-5GW capacity, Meta saves $27B CapEx upfront, or Meta shareholders would punish the stock aggressively. I expect this kind of clever structure to be used in the next 100-200GWs deployment, where $META will generate significant revenue renting out AI compute at lower price to compete with @awscloud , @Azure, @googlecloud, @Oracle and @OpenAI. Current Industry rental price is $3-$10 per hour. $META needs for internal use as well, but will rent excess capacity, similar to the early day of AWS. Meaning, for each GW excess capacity, $META could generate $20-$25B Revenue. However, with $AMD superior TCO, TDP, and lowest Inference/Training cost per million tokens, META would be able to generate $25-$30B per GW excess capacity offering rental rate at 10-15% lower than AWS, Azure, Google Cloud. With $NVDA, $META would be able to generate $5-$10B revenue per GW due to @nvidia Premium position of 70$ gross margin. The most likely outcome will be Split of ~58% $NVDA Rubin Racks to handle 60% training ~42% $AMD Helios Racks to handle 40% training and all Inference with 500-700k additional EPYC Venice to lower Inference cost whether it is renting or internal use! Alright, that is it. Not Financial Advice!















