Podcast işlem fikirleri

Invest Like the Best

Walter Russell Mead - How America Keeps Winning - [Invest Like the Best, EP.490]

8 Eyl 2026299h

Bu bölümde somut çağrı yok

Invest Like the Best

She Knows the 250 People Building AI. Here's What They Actually Believe.

1 Eyl 2026463h

The episode features Sarah (Conviction) discussing the AI investment landscape rather than making specific stock calls. Her most actionable view is a bullish long-term position on AI-driven value creation in biology and robotics — she says she is a 'strong yes' that models can create and capture enormous value in biotech, backing Chai Discovery and Sunday Robotics. She also flags compute independence as a key theme and warns that concentrating the economy in one or three frontier models is a future she does not want, favoring open-source and applied-AI diffusion.

Pozisyonlar · 3

• 54:06Chai DiscoveryLongSarah's firm holds the first check in Chai Discovery, which works with top-10 pharma companies to accelerate R&D, arguing AI models now let companies create and capture enormous value in biology. — Speaker 0–
  Bu işlemin gerekçesi

  “We're the first check-in a company called Chai Discovery, and Chai is working with a number of top 10 pharma companies in really significant ways” — Speaker 0

  1. Speaker claims the conventional view that you can't make money selling software to pharma is now wrong because of models.
  2. Conviction invested the first check in Chai, which has significant contracts with top-10 pharma companies.
  3. Sarah says she is now a 'strong yes' on value creation via AI biology models, implying continued long exposure.
• 20:44Sunday RoboticsLongConviction backed Sunday Robotics, founded by Tony Zhao and Cheng Chi, believing general semi-humanoid robots will be doing things in people's homes in beta by end of this year. — Speaker 0–
  Bu işlemin gerekçesi

  “My partner Pranav and I, were introduced to Tony Zhao and Cheng Chi at Sunday Robotics, when they were PhD students at Stanford.” — Speaker 0

  1. Speaker claims Tony Zhao and Cheng Chi contributed most of the interesting robotics AI ideas over four years.
  2. Conviction invested immediately after the first meeting, citing their data-collection approach to robotics generalization.
  3. She expects humanoid robots in homes in beta by end of this year, supporting a long stance.
• 50:29Nuclear energy (theme)LongSarah's firm has invested in nuclear energy to solve the compute power bottleneck, arguing you need to make nuclear cost-competitive as baseload power to support data centers. — Speaker 0–
  Bu işlemin gerekçesi

  “we've invested in nuclear energy, we've invested in alternative chip architectures.” — Speaker 0

  1. Speaker says compute buildout is bottlenecked on sufficient power including natural gas and nuclear.
  2. Her firm has explicitly invested in nuclear energy as part of its compute-independence thesis.
  3. She emphasizes permitting AP1000 reactors to bring nuclear costs down.

Fikirler · 6

• 3:02OPENAILongSarah works closely with and co-invests alongside major labs including OpenAI, and argues its key priorities of ChatGPT, ads, and coding should be judged on actual competitiveness rather than a grand strategic framework. — Speaker 0+%27,3 bölümden beri
  Bu işlemin gerekçesi

  “I am working closely with and co invested with and have many friends at the labs, the big labs and small.” — Speaker 0

  1. Speaker says she is working closely with and co-invested with many big labs including OpenAI.
  2. She frames OpenAI's priorities as ChatGPT, ads, coding, and co-work expansion.
  3. Rather than betting against the labs, she advises judging each effort's competitiveness, implying continuing engagement.
• 1:02:20ANTHROPICLongSarah says she works with and co-invests with the major labs including Anthropic, and frames its priorities as AGI in a safe way with chat, coding, and co-work expansion. — Speaker 0+%7,7 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI, right, or ASI, in a safe way where they capture a lot of profit.” — Speaker 0

  1. Speaker says she is working closely with and co-invested with many of the big labs.
  2. She names Anthropic among labs whose priorities ladder to safe AGI with profit capture.
  3. She advises judging these efforts' competitiveness rather than shorting the labs.
• 1:02:20GOOGLLongSarah lists DeepMind alongside OpenAI and Anthropic as a major lab whose key priorities are AGI/ASI with profit capture, and her firm co-invests with and works closely with these labs. — Speaker 0+%4,0 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI, right, or ASI, in a safe way where they capture a lot of profit.” — Speaker 0

  1. Speaker says she works closely with and co-invests with major labs including DeepMind.
  2. She identifies DeepMind's core priority as achieving AGI/ASI safely while capturing profit.
  3. Google owns DeepMind, making GOOGL the listed exposure to that lab.
• 52:50NVDALongWe are seeing consolidated at scale demand for accelerators or, even supply chain independence because, you know, the big buyers of it want it too. — Speaker 0+%2,0 bölümden beri
  Bu işlemin gerekçesi

  “We are seeing consolidated at scale demand for accelerators or, even supply chain independence because, you know, the big buyers of it want it too.” — Speaker 0

  1. Speaker says the market is different today with consolidated at-scale demand for accelerators.
  2. She notes supply-chain independence goals from big buyers change the risk equation for semiconductor companies.
  3. Nvidia is the dominant accelerator supplier at the center of this compute buildout.
• 49:22TSMLongSarah highlights TSMC's near-monopoly in critical compute supply chain components, arguing the dependence on a single leading-edge fab changes the risk equation and drives investment in supply chain independence. — Speaker 0+%4,1 bölümden beri
  Bu işlemin gerekçesi

  “But there's all these component parts... like a very thin sieve in a place that is not necessarily stable or accessible to The US and allies.” — Speaker 0

  1. Speaker says compute independence depends on leading-edge fab capacity, currently concentrated at TSMC.
  2. She notes 'we can't all be stuck on one line at TSMC,' making the supply chain strategically valuable.
  3. This validates TSMC's critical position even as it motivates alternative capacity investment.
• 50:29Data center builders / solar & battery installersLongSarah's firm keeps looking at companies that are essentially data center builders and battery installers to capture the physical capacity buildup for AI compute. — Speaker 0–
  Bu işlemin gerekçesi

  “We like keep looking at people who are essentially like data center builders, solar installer, solar and battery installers of some kind” — Speaker 0

  1. Speaker says her firm keeps looking at data center builders and solar/battery installers.
  2. These play the operational capacity buildup needed for compute independence.
  3. She describes it as a highly operational business between operations, technology, and real estate.

Invest Like the Best

Sarah Guo - What the 250 People Building AI Believe - [Invest Like the Best, EP.489]

1 Eyl 2026467h

Sarah Guo, founder of Conviction, shares her views on AI investment, highlighting the importance of backing exceptional entrepreneurs and the shift toward compute independence and energy infrastructure. She discusses the rise of open-source models and the need for US competitiveness in AI, while also identifying opportunities in robotics, biology, and data center supply chains. The conversation is largely strategic and philosophical, with few direct stock calls, but she implicitly endorses companies like NVIDIA and TSMC as critical to AI compute.

Fikirler · 9

• 35:22NVDALongNVIDIA is central to AI compute, and Guo highlights the necessity of compute for AI progress and the importance of alternative chip architectures. — Sarah Guo+%0,6 bölümden beri
  Bu işlemin gerekçesi

  “The reality is we have over the last three years increasingly competitive open source models from all fronts, Largely China, but definitely also The US and Europe. Most recently, Thinky, Poolside, people waiting for reflection, NVIDIA model” — Sarah Guo

  1. Guo discusses the importance of compute for AI and mentions 'NVIDIA models' among open-source options.
  2. As a leading GPU designer, NVIDIA is the primary beneficiary of AI compute demand.
  3. The discussion implies sustained demand for accelerators, making NVIDIA a key investment.
• 42:01TSMLongTSMC is a critical supplier of advanced chips, and Guo notes supply chain dependence on it, making it a key beneficiary of AI compute buildout. — Sarah Guo+%2,7 bölümden beri
  Bu işlemin gerekçesi

  “I've got my TSMC mug with me. There are parts of that supply chain that are like a very thin sieve in a place that is not necessarily stable or accessible to The US and allies.” — Sarah Guo

  1. Guo mentions 'my TSMC mug' and discusses the thin supply chain in semi manufacturing.
  2. TSMC holds a near-monopoly on leading-edge chip manufacturing, essential for AI accelerators.
  3. Her concern about supply chain independence implies TSMC's importance and investment value.
• 52:25OPENAILongOpenAI is a leader in frontier AI, and Guo discusses the importance of its models and the need for competition, implying she views it as a positive force. — Sarah Guo+%26,6 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI or ASI in a safe way where they capture a lot of profit.” — Sarah Guo

  1. Guo mentions OpenAI, Anthropic, and DeepMind as top labs and discusses their strategies.
  2. As a major lab, OpenAI is positioned to capture value from AI advances.
  3. Her commentary on the importance of these labs suggests a positive outlook.
• 52:25MSFTLongOpenAI → Microsoft is a major investor in OpenAI and provides Azure cloud services, making it a key listed beneficiary of OpenAI's success. — Sarah Guo-%2,2 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI or ASI in a safe way where they capture a lot of profit.” — Sarah Guo

  1. Guo discusses OpenAI's importance and its compute needs.
  2. Microsoft has a significant equity stake in OpenAI and provides the cloud infrastructure.
  3. Positive views on OpenAI translate to Microsoft's success.
• 52:25ANTHROPICLongAnthropic is a leading AI lab, and Guo includes it among the top labs that will shape the future of AI. — Sarah Guo+%6,4 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI or ASI in a safe way where they capture a lot of profit.” — Sarah Guo

  1. Guo names Anthropic as a top lab alongside OpenAI and DeepMind.
  2. Anthropic produces competitive frontier models.
  3. Her discussion implies she views Anthropic as a significant player.
• 52:25AMZNLongAnthropic → Amazon, through AWS, is a major cloud provider and investor in Anthropic, making it a key beneficiary of AI growth. — Sarah Guo-%2,7 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI or ASI in a safe way where they capture a lot of profit.” — Sarah Guo

  1. Guo discusses the importance of AI compute and the major labs.
  2. Amazon invests heavily in Anthropic and provides AWS cloud services.
  3. Positive outlook on AI and Anthropic benefits Amazon.
• 52:25GOOGLLongDeepMind → Alphabet's DeepMind is a top AI lab, and its cloud computing arm benefits from AI compute demand. — Sarah Guo+%3,2 bölümden beri
  Bu işlemin gerekçesi

  “Let's assume the priority for OpenAI and Anthropic and DeepMind is AGI or ASI in a safe way where they capture a lot of profit.” — Sarah Guo

  1. Guo names DeepMind among the top labs.
  2. Alphabet owns DeepMind and operates Google Cloud for AI workloads.
  3. AI progress benefits Alphabet through its AI leadership and cloud revenue.
• 35:22METALongMeta is a leading open-source AI model contributor (Llama), aligning with Guo's view on open-source ecosystem importance. — Sarah Guo+%17,2 bölümden beri
  Bu işlemin gerekçesi

  “The reality is we have over the last three years increasingly competitive open source models from all fronts, Largely China, but definitely also The US and Europe. Most recently, Thinky, Poolside, people waiting for reflection, NVIDIA model” — Sarah Guo

  1. Guo highlights the importance of open-source AI models.
  2. Meta develops and releases Llama models, a major open-source effort.
  3. Her support for open-source suggests a positive view on Meta's AI strategy.
• 35:22Mistral AILongMistral is a European open-source AI model provider, and Guo mentions it as an example of competitive open-source models. — Sarah Guo–
  Bu işlemin gerekçesi

  “The reality is we have over the last three years increasingly competitive open source models from all fronts, Largely China, but definitely also The US and Europe. Most recently, Thinky, Poolside, people waiting for reflection, NVIDIA model” — Sarah Guo

  1. Guo lists Mistral among open-source model developers.
  2. Mistral creates competitive open-source AI models.
  3. Her positive view on open-source implies a positive view on Mistral.

Invest Like the Best

Ex-NVIDIA Engineer: Why AI Is About to Get 1000x Cheaper

25 Ağu 2026631h

Pozisyonlar · 2

• 1:16:31NVDALongSpeaker 0 says he is bullish on NVIDIA short term and that you should 'never bet against them' because they will reinvent themselves and remain best-in-class at both low-latency and throughput-optimized inference. — Speaker 0+%4,9 bölümden beri
  Bu işlemin gerekçesi

  “I am bullish on NVIDIA in the short term, and NVIDIA should never bet against them.” — Speaker 0

  1. Speaker 0 explicitly states he is bullish on NVIDIA in the short term and warns against betting against the company.
  2. Despite a contrarian view on lackluster per-watt improvements and TSMC risk, the speaker still frames NVIDIA as a short-term long.
• 46:34AMDLongSpeaker 0 says AMD chips are good and that misperceptions about their inferiority create alpha for buyers willing to optimize them, though supply is increasingly allocated. — Speaker 0+%18,8 bölümden beri
  Bu işlemin gerekçesi

  “there is alpha and just like other people have this perception that AMD is not as as good as NVIDIA. That's music to my ears.” — Speaker 0

  1. Speaker 0 says 'AMD, I think great chips overall' and that the perception AMD is not as good as NVIDIA is 'music to my ears' because it lets him buy cheap capacity.
  2. He notes AMD lacks the same out-of-box kernel optimization as NVIDIA, so a strong kernel team can extract more performance and buy at a discount.
  3. He also notes Meta and OpenAI have bought a lot of AMD chips, meaning supply is tightening, but the long-tail opportunity remains.

Fikirler · 2

• 1:19:56MUKaçınThere's no easy way to bring on a lot more fabs of memory. And those guys have been So it's gonna be a while until we know. — Speaker 0+%8,1 bölümden beri
  Bu işlemin gerekçesi

  “There's no easy way to bring on a lot more fabs of memory. And those guys have been So it's gonna be a while until we know.” — Speaker 0

  1. Speaker 0 identifies HBM capacity as one of the three to five critical bottlenecks in modern chip supply and says there is no easy way to add memory fabs.
  2. He notes the 'boys in Boise' (Micron) don't love huge capex for cyclical reasons and have been burned before, implying supply discipline.
• 1:16:58TSMKaçınMy contrarian take is that it wouldn't be that bad. Supply would take a shock for sure, but the best processes that we have in the West, like Intel, not that far behind, at worst, maybe 2x worst performance per watt. — Speaker 0+%3,7 bölümden beri
  Bu işlemin gerekçesi

  “My contrarian take is that it wouldn't be that bad. Supply would take a shock for sure, but the best processes that we have in the West, like Intel, not that far behind, at worst, maybe 2x worst performance per watt.” — Speaker 0

  1. He notes TSMC 5nm vs 4nm vs 3nm vs 2nm doesn't show dramatic per-watt improvements, and Intel is at worst 2x behind.
  2. This is a skeptical view on the premium assigned to TSMC's technological lead rather than an explicit short or avoid recommendation.

Invest Like the Best

Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]

25 Ağu 2026635h

Neil Movva, founder of SAIL Research, argues that the future of AI inference lies in background, long-running agents where cost per token matters more than latency. He advocates a 'scavenger strategy' of buying cheaper, non-NVIDIA chips and unreliable power sources, and believes open-source models will commoditize the frontier labs' premium. He is bullish on AMD and bearish on NVIDIA's performance-per-watt improvements, though he warns memory (HBM) will remain a bottleneck.

Pozisyonlar · 1

• 45:56AMDLongAMD GPUs are undervalued because people haven't optimized kernels for them, but SAIL can exploit that inefficiency to buy cheap compute and resell tokens at a margin. — Speaker 2 (Neil Movva)+%20,0 bölümden beri
  Bu işlemin gerekçesi

  “There is alpha and just other people have this perception that AMD is not as good as NVIDIA. That's music to my ears.” — Speaker 2 (Neil Movva)

  1. Neil claims AMD chips are great but underutilized because 'people don't understand how to program them very well'.
  2. SAIL's kernel expertise enables extracting more performance from AMD chips, making them a cost-effective alternative to NVIDIA.
  3. Thus, buying AMD represents a direct bet on the arbitrage opportunity in inference compute.

Fikirler · 9

• 33:12CBRSLongNeil views Cerebras as a valuable accelerator for memory-bound operations, despite its limitations with KV cache, and sees a hybrid future where it complements GPUs. — Speaker 2 (Neil Movva)+%4,8 bölümden beri
  Bu işlemin gerekçesi

  “You would like to put the attention possibly on the GPU and the MLP on the Cerberus chip. And I believe this is what's happening with NVIDIA and Grok.” — Speaker 2 (Neil Movva)

  1. Neil believes Cerebras excels at fast access to memory, which suits MLP layers that are compute-bound.
  2. He notes a likely partnership with NVIDIA (and Grok) to split MLP and attention workloads, indicating a durable role for Cerebras.
  3. The trade is a long on Cerebras as a niche but essential compute component.
• 1:11:58NVDAShortNVIDIA's performance-per-watt gains across generations are minimal, so its premium may not justify long-term advantage, and it's vulnerable to cheaper alternatives. — Speaker 2 (Neil Movva)+%5,1 bölümden beri
  Bu işlemin gerekçesi

  “you look at Hopper to Blackwell to Rubin and you compare like for like what is the performance per watt... it hasn't improved all that much.” — Speaker 2 (Neil Movva)

  1. Neil claims that from Hopper to Blackwell to Rubin, performance per watt hasn't improved dramatically.
  2. He suggests that competitors like Intel are only ~2x behind, narrowing the moat.
  3. This implies that NVIDIA's dominant position may erode, leading to a possible short thesis.
• 1:12:27INTCLongIntel's process technology is close to TSMC, making it a viable alternative in a geopolitical disruption, but Neil doesn't provide a direct buy recommendation. — Speaker 2 (Neil Movva)+%24,3 bölümden beri
  Bu işlemin gerekçesi

  “the best processes that we have in the West, like Intel, not that far behind, at worst, like maybe two x, worst performance per watt.” — Speaker 2 (Neil Movva)

  1. Neil argues that if access to TSMC is lost, Intel is not far behind in performance per watt.
  2. He frames Intel as a strategic backup, not a primary investment.
  3. This could be a long-term hedge, but the call is speculative.
• 1:14:40MULongHBM memory is a major bottleneck, and companies that supply it like Micron will benefit from rising demand and pricing power. — Speaker 2 (Neil Movva)+%9,1 bölümden beri
  Bu işlemin gerekçesi

  “There's no easy way to bring on a lot more fabs of memory. So it's just gonna be a while until we have [enough].” — Speaker 2 (Neil Movva)

  1. Neil identifies HBM capacity as one of the most critical bottlenecks in chip supply.
  2. He notes that memory companies are reluctant to expand due to cyclicality, keeping supply tight.
  3. This scarcity directly benefits memory chip manufacturers like Micron.
• 1:12:13TSMLongTSMC remains the critical enabler of AI compute, but Neil's view is that its efficiency gains are marginal, so its value is more as a monopoly chipmaker than an innovator. — Speaker 2 (Neil Movva)+%4,7 bölümden beri
  Bu işlemin gerekçesi

  “If you look at TSMC five nanometer versus four versus three versus two, the performance per watt on these chips doesn't change like a dramatic amount.” — Speaker 2 (Neil Movva)

  1. Neil acknowledges TSMC's importance but notes that performance per watt improvements across nodes are not dramatic.
  2. This implies TSMC's pricing power may be limited, but its indispensability keeps it a safe long.
  3. The trade is based on TSMC's foundational role, not on innovation.
• 1:07:30ANTHROPICKaçınAnthropic's premium for being months ahead may not last as open-source models catch up, and enterprise adoption lags, so its current pricing power is under threat. — Speaker 2 (Neil Movva)+%5,8 bölümden beri
  Bu işlemin gerekçesi

  “I don't know that the premium for being three to six months ahead is gonna last that long.” — Speaker 2 (Neil Movva)

  1. Neil claims frontier labs like Anthropic charge a premium for being 3-6 months ahead, but enterprises don't move that fast.
  2. Open-source models are increasingly capable, reducing the moat.
  3. This could lead to margin compression for Anthropic.
• 1:07:05OPENAIKaçınOpenAI's closed-source model may face competition from open-source alternatives, making its long-term dominance uncertain. — Speaker 2 (Neil Movva)+%28,6 bölümden beri
  Bu işlemin gerekçesi

  “I don't think it's fundamentally possible to prevent the diffusion of information or model capabilities. It will happen.” — Speaker 2 (Neil Movva)

  1. Neil argues that distillation from closed models is inevitable, aiding open-source rivals.
  2. He believes open source will always exist and may erode closed labs' advantages.
  3. Thus, he is not optimistic on OpenAI's market position.
• 1:15:18CRWVLongNVIDIA's strategy of empowering neo-clouds like CoreWeave creates a supportive ecosystem, and CoreWeave's scale and NVIDIA partnership make it a winner in compute supply. — Speaker 2 (Neil Movva)-%6,9 bölümden beri
  Bu işlemin gerekçesi

  “Jensen is really good at making his friends billionaires. He's made CoreWeave a billion dollar company, a many billion dollar company.” — Speaker 2 (Neil Movva)

  1. Neil notes NVIDIA has made CoreWeave a multi-billion-dollar company, showing NVIDIA's commitment.
  2. CoreWeave's access to NVIDIA chips and its role as a leading inference provider positions it well.
  3. The trade is based on the continued growth of AI compute demand.
• 1:01:42SPCXShortxAI's cluster utilization issues indicate operational inefficiencies that could undermine its cost structure, making it a potential underperformer. — Speaker 2 (Neil Movva)+%12,9 bölümden beri
  Bu işlemin gerekçesi

  “we all make fun of x AI for having, you know, some challenges with total flop utilization on its clusters.” — Speaker 2 (Neil Movva)

  1. Neil mentions that xAI has had 'challenges with total flop utilization' on its clusters.
  2. This suggests their massive compute investments are not being used effectively.
  3. Inefficient compute leads to higher costs per token, reducing competitiveness.

Invest Like the Best

What Happens When the AI Boom Runs Out of Money

18 Ağu 2026799h

Ben Thompson argues the AI capex boom faces a timing mismatch that could trigger a funding blow-up, while remaining long-term bullish on AI's economic impact. He sees Amazon as the most compelling large-cap setup due to its build-for-itself model, and calls out Nvidia's margin sustainability as threatened by hyperscaler competition and circular financing that effectively hides price cuts. He also flags memory makers and TSMC as structurally risky, with TSMC's conservatism having unintentionally saved Intel by creating acute compute scarcity.

Pozisyonlar · 1

• 47:39AMZNLongBen says Amazon is the most compelling megacap because it builds for itself first, then sells the infrastructure externally, making its core business impervious to AI and its new business lines organic. — Speaker 0-%2,9 bölümden beri
  Bu işlemin gerekçesi

  “The answer is always Amazon. And the reason, because what Amazon is so compelling is the extent to which they build for them. They are their first best customer.” — Speaker 0

  1. Amazon's core retail and logistics moat is deepened by AI rather than disrupted, and AWS provides the capital-light platform to sell AI capacity.
  2. The trade is long AMZN as the most diversified and structurally advantaged AI infrastructure and commerce play.

Fikirler · 11

• 57:26METALongMeta's probably the most interesting — Speaker 0+%19,1 bölümden beri
  Bu işlemin gerekçesi

  “Meta's probably the most interesting” — Speaker 0

  1. Ben says Meta's ad business is an unmatched verification engine where human clicks and purchases validate generated content at global scale, making it ideal for training AI models.
  2. He notes Meta pays nothing for content and could use AI to improve ad matching, with even a few percentage points of improvement worth billions.
  3. The view is that Meta's founder-led push onto the frontier is rational and potentially highly profitable despite near-term spending concerns.
• 1:15:36NVDAKaçınNvidia's position is, I think definitely unnatural... they're actually not maintaining their margins because who is buying this whole question of circular financing — Speaker 0+%0,3 bölümden beri
  Bu işlemin gerekçesi

  “Nvidia's position is, I think definitely unnatural... they're actually not maintaining their margins because who is buying this whole question of circular financing” — Speaker 0

  1. Ben claims Nvidia's deals to backstop neo clouds and take equity represent an implicit price cut that reduces holistic profitability even if margins look high.
  2. Hyperscalers are selling their own chips (TPUs, Trainium) as commodities externally, threatening Nvidia's differentiation and pricing power.
  3. The risk is that power abundance and competition erode Nvidia's margins, making the stock vulnerable despite current dominance.
• 46:27INTCLongBen argues Intel is ultimately saved by the acute compute scarcity caused by TSMC's underinvestment, as big tech companies will be forced to bring Intel's foundry up to speed to diversify away from Taiwan risk. — Speaker 0+%12,1 bölümden beri
  Bu işlemin gerekçesi

  “The scarcity is what ultimately saved Intel. And I expect at some point in the near that they're gonna announce like some major partner for the first time.” — Speaker 0

  1. Ben says TSMC's conservatism created a massive compute shortage, and big tech is foregoing so much revenue that they will go through the pain of working with Intel.
  2. Intel needs customers to build a real foundry business, and the scarcity creates the economic incentive for them to do so.
  3. He expects Intel to announce a major partner soon, which would be a big deal for the stock.
• 54:41AAPLLongfrom my perspective, I'm fine with Apple not doing AI. I want them to keep making great devices. — Speaker 0+%8,7 bölümden beri
  Bu işlemin gerekçesi

  “from my perspective, I'm fine with Apple not doing AI. I want them to keep making great devices.” — Speaker 0

  1. Ben says Apple's aggregator position means suppliers come to them, and they can provide ambient AI via on-device processing without leading-edge models.
  2. The smartphone remains the perfect device and Apple's ecosystem and physical manufacturing are hard to disrupt.
  3. He prefers Apple to stick to what it does best rather than chase probabilistic AI development.
• 14:00GOOGLLongBerkshire being the symbol of that equity issuance... Are they actually not just an investor in Google, but a model for Google and where they're going? — Speaker 0+%2,1 bölümden beri
  Bu işlemin gerekçesi

  “Berkshire being the symbol of that equity issuance... Are they actually not just an investor in Google, but a model for Google and where they're going?” — Speaker 0

  1. Ben says Google's search is a perfect high-margin aggregator like See's Candies, and AI is the BNSF-like low-margin but huge absolute-dollar opportunity.
  2. Google's ability to issue equity and tap debt markets gives it a lower cost of capital than Nvidia's customers.
  3. The trade is long GOOGL as the company best positioned to convert search profits into AI infrastructure dominance.
• 40:26TSMKaçınBen warns TSMC's conservatism transferred overcapacity risk to big tech, and its dominant position is now threatened by customers actively seeking alternatives like Intel and Samsung due to acute scarcity. — Speaker 0+%2,7 bölümden beri
  Bu işlemin gerekçesi

  “TSMC is arguably worse because there's only one... My concern for the memory makers is they might have done the same thing.” — Speaker 0

  1. Ben argues TSMC offloaded risk onto big tech by not investing enough, leading to compute shortages and foregone revenue for customers.
  2. This shortage is forcing big tech to build alternatives, which could erode TSMC's long-term dominance.
  3. The view is that TSMC's near-monopoly is more fragile than it appears due to geopolitical and supply chain dynamics.
• 1:00:33MSFTLongBen says Microsoft's middleware and enterprise-platform strategy is rational and a repeat of IBM's 1990s playbook, collecting tolls on AI adoption without competing on the frontier. — Speaker 0+%2,2 bölümden beri
  Bu işlemin gerekçesi

  “Microsoft will help you figure out AI... We're gonna build this platform, this harness, this sort of middle layer where you can... we're dependable, we're stable.” — Speaker 0

  1. Ben compares Microsoft to IBM under Lou Gerstner, building a consulting and middleware layer to help enterprises adopt AI.
  2. Microsoft provides a dependable, backwards-compatible platform and manages model changes for customers averse to risk.
  3. This allows Microsoft to generate massive free cash flow and dividends while others fight on the frontier.
• 54:57OPENAILongBen sees OpenAI as one of the two frontier labs with the biggest upside, fueled by religious-like belief, though also the riskiest given the need to make the business work. — Speaker 0+%26,6 bölümden beri
  Bu işlemin gerekçesi

  “Opening eye and anthropic obviously are the riskiest, but also have the biggest upside... the power of belief.” — Speaker 0

  1. Ben says OpenAI and Anthropic are clearly on the frontier and their power of belief goes a long way.
  2. The biggest upside is offset by the risk that they need to build a viable business to survive.
  3. This is a venture-style bet on a private company, not a directly tradable stock.
• 54:57ANTHROPICLongBen groups Anthropic with OpenAI as a frontier lab with the biggest upside and religious-like belief, but also high risk, noting it is 'kicking' OpenAI in the enterprise market. — Speaker 0+%19,3 bölümden beri
  Bu işlemin gerekçesi

  “Opening eye and anthropic obviously are the riskiest, but also have the biggest upside.” — Speaker 0

  1. Ben says OpenAI and Anthropic are the two clearly on the frontier with the biggest upside and biggest risk.
  2. He notes Anthropic's strong enterprise performance versus OpenAI.
  3. This is a private company bet, not a listed instrument.
• 1:09:03METALongBen argues Meta's advertising business is uniquely positioned to benefit from AI via improved ad matching and content generation, and that being on the frontier is worth it for this alone. — Speaker 0+%19,1 bölümden beri
  Bu işlemin gerekçesi

  “This alone is worth them investing in being on the leading edge in having these amazing models.” — Speaker 0

  1. Meta has a massive liquid ad marketplace that acts as a verification machine for AI models.
  2. AI-driven ad matching improvements of even a few percentage points could generate billions in incremental revenue.
  3. The trade is long META based on its underappreciated ad+AI flywheel.
• 1:07:10METALongBen suggests Meta is well placed in a world where AI interactions increase the desire for human connection, potentially making social networking important again. — Speaker 0+%19,1 bölümden beri
  Bu işlemin gerekçesi

  “It's possible in AI, actually social network is important again, because like we actually want humans. We want to have some sort of connection to them.” — Speaker 0

  1. Ben argues if people interact with AI all the time, the desire for human connection becomes greater.
  2. This could benefit Meta's social networking roots and its network effects.
  3. The view is that Meta's social graph becomes more valuable in an AI-saturated world.

Invest Like the Best

Ben Thompson on Big Tech, China, and the AI Boom Running Out of Money - [Invest Like the Best, EP.487]

18 Ağu 2026803h

Ben Thompson discusses the AI boom, emphasizing the risk of an overbuild and the potential for a capital crunch as tech companies exhaust free cash flow and debt markets, pointing to equity issuance and NVIDIA's $500B fund as signs. He highlights Amazon as the most compelling setup due to its 'first best customer' strategy, is bullish on Meta's advertising opportunity with AI, and sees Microsoft as a rational IBM-style middleware play. He warns that NVIDIA's margins are effectively a price cut via risk-taking, and that compute is becoming a commodity, favoring hyperscalers like Google and Amaz

Fikirler · 14

• 45:03AMZNLongAmazon's approach of building for itself first and then selling to others, as evidenced by AWS, Graviton, and Trainium, makes it the most compelling tech setup and deeply insulated from AI disruption. — Ben Thompson-%2,8 bölümden beri
  Bu işlemin gerekçesi

  “The answer is always Amazon. The reason Amazon is so compelling is the extent to which they build for them. They are their first best customer.” — Ben Thompson

  1. Ben states 'The answer is always Amazon' and details how Amazon's scale as its own first customer lets it iterate on products like Trainium and then offer them externally.
  2. Amazon's core retail and logistics businesses are impervious to model-based AI, while AWS benefits from AI demand, creating a durable moat.
  3. This positions Amazon to win in AI infrastructure without the existential threat faced by pure digital companies.
• 1:03:05METALongMeta's advertising business stands to gain billions from AI's ability to improve ad targeting and content generation, making it a top AI beneficiary, though its story is undercommunicated. — Ben Thompson+%18,5 bölümden beri
  Bu işlemin gerekçesi

  “The potential upside in terms of showing people better ads that are more relevant to them. They only need to increase a few percentage points for the returns to be billions and billions of dollars.” — Ben Thompson

  1. Ben argues that AI will enable Meta to show better ads and that even a few percentage points of improvement in ad relevance yields billions in revenue.
  2. Meta's real-world validation loop via its ad marketplace gives it a unique advantage in AI-generated content.
  3. This upside justifies Meta's frontier AI investments, making it a compelling long despite the company's failure to articulate its ad business's value.
• 56:06MSFTLongMicrosoft's rational IBM-style middleware play for AI, leveraging its enterprise relationships and avoiding frontier risk, sustains its business model, though it faces a long-term existential threat from AI. — Ben Thompson+%3,1 bölümden beri
  Bu işlemin gerekçesi

  “That's Microsoft's playbook. Microsoft will help you figure out AI. It will help you figure out in a way where you're not giving away the crown jewels of these companies.” — Ben Thompson

  1. Ben compares Microsoft's AI strategy to IBM's middleware approach in the 1990s, where they sell a dependable platform rather than frontier models.
  2. This strategy generates massive free cash flow and dividends while allowing Microsoft to avoid the risk of frontier AI research.
  3. The trade is to hold Microsoft for its cash generation, but recognize the potential long-term disruption to its software business from AI.
• 1:08:08NVDAShortNVIDIA's margins are illusory because the company is taking on risk through circular financing deals, effectively a price cut, and faces rising competition from hyperscaler chips that will commoditize compute. — Ben Thompson-%0,3 bölümden beri
  Bu işlemin gerekçesi

  “If you actually ascribe a value to that, to NVIDIA's taking equity in the neo clouds or whatever, they guarantee they're gonna buy all their compute to 2030... that is a diminution of Nvidia's profitability.” — Ben Thompson

  1. Ben argues that NVIDIA's equity investments in neo-clouds and backstops are risk-taking with a cost, reducing true profitability.
  2. He sees hyperscalers like Google and Amazon as NVIDIA's biggest problem, as they build and sell their own chips as commodities, eroding NVIDIA's differentiation.
  3. The trade is to be short NVIDIA, as its current valuation does not reflect the true economics and competitive threats.
• 13:14GOOGLLongGoogle's high-margin search business is akin to See's Candies, funding an AI future with astronomical absolute profits, and its TPU sales to Anthropic show a commodity chip strategy that threatens NVIDIA. — Ben Thompson+%2,2 bölümden beri
  Bu işlemin gerekçesi

  “And the purest aggregator of them all, like scales in every direction... Meanwhile, there's this AI opportunity which requires just astronomical it's just incinerating cash.” — Ben Thompson

  1. Ben draws a parallel between Google's search margins and See's Candies, suggesting AI is a lower-margin but much larger absolute profit opportunity.
  2. Google's equity issuance and TPU sales to Anthropic indicate a strategy to monetize AI infrastructure as a commodity, benefiting from the AI buildout.
  3. This positions Google as a long-term winner in AI, with the scale to compete on cost and a diversified business model.
• 40:24TSMKaçınTSMC's conservative capacity expansion has offloaded risk to its customers, but the resulting scarcity will incentivize them to fund competitors like Intel, threatening TSMC's dominance. — Ben Thompson+%2,2 bölümden beri
  Bu işlemin gerekçesi

  “TSMC's like, we don't wanna take the risk. Risk doesn't disappear. It just moves. ... The scarcity is what ultimately saved Intel.” — Ben Thompson

  1. Ben argues TSMC's underinvestment in recent years has forced big tech to seek alternatives, with scarcity saving Intel.
  2. The risk TSMC avoided by not building capacity has transferred to its customers, who are now motivated to diversify supply.
  3. This creates a long-term structural challenge for TSMC, making it an avoid despite its current strength.
• 44:15INTCLongIntel is poised to benefit as a geopolitical insurance alternative to TSMC, and the scarcity of compute will force big tech to bring Intel's foundry up to speed. — Ben Thompson+%9,8 bölümden beri
  Bu işlemin gerekçesi

  “I expect at some point that they're gonna announce some major partner for the first time. It's gonna be a big deal. But ultimately, TSMC brought it on themselves.” — Ben Thompson

  1. Ben notes that TSMC's underinvestment and the concentration risk will push big tech to use Intel despite the costs.
  2. He expects Intel to announce a major foundry customer, which would be a significant catalyst.
  3. The trade is to go long Intel on this turnaround thesis, though it is a long-shot with high risk.
• 51:23OPENAILongOpenAI has the biggest upside as a frontier AI lab, driven by belief and existential urgency, but its financial path is risky with high capital needs, making it a speculative long. — Ben Thompson+%24,9 bölümden beri
  Bu işlemin gerekçesi

  “OpenEye and Anthropic obviously are the riskiest, but also have the biggest upside. Never discount, number one, the power of belief. They think they're creating God.” — Ben Thompson

  1. Ben describes OpenAI and Anthropic as the riskiest but with the biggest upside due to their evangelical approach to AI.
  2. He notes that OpenAI is 'down bad' financially, having to pivot to advertising and enterprise sales to fund compute costs.
  3. The trade is to have exposure to OpenAI through potential public listings or partners, given its frontier position.
• 51:23ANTHROPICLongAnthropic is a frontier AI lab with similarly high upside and risk, benefiting from Amazon's Trainium partnership and its own model frontier, but lacks public market access. — Ben Thompson+%18,4 bölümden beri
  Bu işlemin gerekçesi

  “OpenEye and Anthropic obviously are the riskiest, but also have the biggest upside. Never discount, number one, the power of belief. They think they're creating God.” — Ben Thompson

  1. Ben groups Anthropic with OpenAI as riskiest but with biggest upside due to belief and urgency.
  2. Anthropic's use of Amazon's Trainium and its competitive pressure on OpenAI highlight its strategic importance.
  3. The trade is to gain exposure through Amazon's partnership, as Anthropic itself is private.
• 52:46SpaceX AIKaçınSpaceX AI's data-center-in-space play is highly differentiated but the weakest among frontier players because owning a model may be unnecessary and capital could be wasted. — Ben Thompson–
  Bu işlemin gerekçesi

  “Probably the case for SpaceX AI is probably the weakest because the data center and space play is so highly differentiated. If that plays out, I'm not sure to what extent they need to even have their own model.” — Ben Thompson

  1. Ben argues that SpaceX AI's differentiation in space data centers is so strong that they may not need their own model.
  2. He questions the allocation of billions to model development when sold capacity yields returns.
  3. This makes SpaceX AI's setup the least compelling among the five frontier contenders.
• 48:09AAPLLongApple is insulated from AI disruption due to its ecosystem, physical products, and access to customers, but risks falling into a Microsoft-like trap by assuming the phone remains central. — Ben Thompson+%9,4 bölümden beri
  Bu işlemin gerekçesi

  “If you own access to customers, suppliers come to you, not the other way around it. So they can get suppliers for their AI as needed... You can see a future where this absolutely can work on device.” — Ben Thompson

  1. Ben notes Apple's ownership of customers and physical goods creates a strong moat against AI disruption.
  2. He suggests Apple could eventually run on-device AI using customer electricity, saving on inference costs.
  3. However, he warns of the risk of being left behind in ambient AI, making the long thesis a balanced one.
• 12:12BRKBLongBerkshire's model of using high-margin cash flows like See's Candies to fund lower-margin, large-absolute-profit businesses like BNSF is analogous to Google's AI pivot, making it a symbolic and actual investor in Google. — Ben Thompson+%2,0 bölümden beri
  Bu işlemin gerekçesi

  “The brilliance of the BNSF railway thing was basically they took the Seas Candy profits and said, here's another industry whose margins are way worse, but the absolute dollar amounts are so large...” — Ben Thompson

  1. Ben draws a direct parallel between Berkshire's capital allocation and Google's decision to issue equity for AI.
  2. Berkshire's investment in Google is seen as symbolic of Google's own future, where high-margin search funds lower-margin AI.
  3. This reinforces Berkshire's positioning as a long-term beneficiary of the AI buildout through its Google stake.
• 38:31Memory makers (Samsung etc.)KaçınMemory makers face a strategic risk as their leverage over scarcity will be routed around, similar to Iran closing the Strait of Hormuz, leading to long-term erosion of their market power. — Ben Thompson–
  Bu işlemin gerekçesi

  “My concern for the memory makers is they might have done the same thing. No one's gonna let themselves get in this situation again, as far as memory goes.” — Ben Thompson

  1. Ben compares memory makers to Iran, noting that successful use of leverage forces others to build alternatives.
  2. He warns that customers will seek alternatives to avoid dependency, undermining the memory oligopoly's power.
  3. This makes the memory sector an avoid despite current boom conditions.
• 1:14:24Power/energy companiesLongPower is the ultimate constraint and lasting output of the AI bubble; even if AI blows up, energy abundance could be transformative, making power investments a durable long. — Ben Thompson–
  Bu işlemin gerekçesi

  “The railroads, BNSF is throwing off money that's going to Google from Northern Pacific and Jay Cook selling bonds to retail investors. ... What is it gonna be? Power. It has to be power.” — Ben Thompson

  1. Ben argues that the lasting legacy of AI will be power infrastructure, like fiber from the dot-com era.
  2. He highlights the US's rapid power buildout, including nuclear restarts and natural gas, as a positive signal.
  3. Investing in power generation and infrastructure is a bet on long-term energy abundance, regardless of AI outcomes.

Invest Like the Best

Everyone Is Still Undersizing the AI Market | Eric Vishria

11 Ağu 2026967h

Pozisyonlar · 7

• 1:12:10NVDALongVishria says every layer of the AI stack works, explicitly answering 'Will NVIDIA do well? Yes,' while adding that differentiation still matters most. — Eric Vishria+%0,5 bölümden beri
  Bu işlemin gerekçesi

  “Will NVIDIA do well? Yes. Will these chip startups do well? Or some set of them? Yes.” — Eric Vishria

  1. Vishria argues there is massive zero-sum thinking about AI value accrual and that the pie is far bigger than people assume.
  2. He lists semis and Nvidia as among the players that will do well as compute demand keeps rising.
  3. Buy/own Nvidia as the core accelerator beneficiary of the AI buildout.
• 38:49CBRSLongVishria's firm backed Cerebras since 2016 and took it public in May; he calls it a winner in the specialized AI chip generation and 'astonishingly bullish' on the effort. — Eric Vishria-%14,5 bölümden beri
  Bu işlemin gerekçesi

  “You know, you've had Grok and Cerebras etch. Like, it'll keep getting fought out. But I think that will end up being big.” — Eric Vishria

  1. Vishria claims AI is a giant new workload with a novel communication bottleneck that specialized wafer-scale chips solve.
  2. He was an investor in Cerebras from a 2016 seed pitch, it 'works' by 2020, and the firm took it public in May 2026.
  3. Own Cerebras as the specialized AI compute winner of the fifth compute generation.
• 48:15Sunday RoboticsLongVishria says his firm is an investor in Sunday Robotics, which is solving the hard part of robotics — a pre-training-plus-post-training data pipeline using custom gloves — for household robots. — Eric Vishria–
  Bu işlemin gerekçesi

  “We're investors in Sunday Robotics, which is going after this exact pipeline.” — Eric Vishria

  1. Vishria claims the key robotics problem is bootstrapping a model without internet-scale data, solved by prioritizing high-value data and vertical integration.
  2. Sunday Robotics, backed by his firm, uses gloves matched to robot hands for high-quality data transfer and a pretraining pipeline.
  3. Own Sunday Robotics as a robotics winner before the task (laundry) matters.
• 1:14:15New LanternLongVishria cites his firm's investment in New Lantern, which is approaching the radiology AI opportunity Hinton identified — contingent on aggregating data and solving reimbursement/liability hurdles. — Eric Vishria–
  Bu işlemin gerekçesi

  “And actually I think the studies and areas have shown that to be true. And we have an investment in a company called New Lantern which is approaching this.” — Eric Vishria

  1. Vishria claims AI can read radiology imaging better than humans, but the training data doesn't exist in aggregated form and reimbursement/liability are hurdles.
  2. His firm invested in New Lantern, which is working on exactly that radiology AI application.
  3. Own New Lantern as the AI-radiology beneficiary that solves the data and workflow bottlenecks.
• 21:06Public SaaS companies (basket)KaçınVishria argues legacy public SaaS names face a shifted competitive frontier — multiple compression from ~30x to ~6x means 'every single day that you are hitting your plan, you are destroying equity value'. — Eric Vishria–
  Bu işlemin gerekçesi

  “Get to AI or be worth three times revenue.” — Eric Vishria

  1. Vishria claims the SaaS competitive frontier completely shifted: criteria to win (cost, zero-to-infinity scaling, portability) changed with AI.
  2. Public SaaS traded ~30x in 2021 and ~6x now, so even 4x revenue growth left them worth less.
  3. Avoid/short legacy public SaaS that keeps executing its old plan instead of reinventing.
• 1:09:48Private SaaS companies ($100M–$500M ARR)KaçınVishria says 500+ mid-size private SaaS companies are 'stuck' — AI natives sucked the oxygen out of the room, the IPO window was missed, and employees/investors have no exit. — Eric Vishria–
  Bu işlemin gerekçesi

  “There's 500 something, probably SaaS companies that are between $100,000,000 and $500,000,000 that are private... They're stuck.” — Eric Vishria

  1. Vishria claims AI-native growth rates absorbed all investor interest, closing the IPO window for traditional SaaS.
  2. Hundreds of private SaaS companies between $100M and $500M ARR have no tenders or exit path.
  3. Avoid investing in that cohort — the liquidity window is gone.
• 1:12:04Hyperscale cloud providers (CSPs)LongVishria explicitly answers 'will the CSPs do well? Yes,' as part of his 'everything works' view of AI value accrual across the stack. — Eric Vishria–
  Bu işlemin gerekçesi

  “Will the CSPs do well? Yes. Yes. Will some of these Neo Clouds do well? Yes.” — Eric Vishria

  1. Vishria claims AI demand is enormous and value will accrue broadly, not to one winner.
  2. He explicitly says CSPs (AWS, Azure, GCP) will do well alongside Neoclouds, Fireworks, Nvidia and chip startups.
  3. Own the CSPs as AI infrastructure beneficiaries.

Fikirler · 9

• 6:34NETLongVishria cites Cloudflare as an AWS-era example of a sub-scale cloud provider that emerged as a $100B company outside the big three, supporting his 'oligopoly plus smaller $100B winners' thesis. — Eric Vishria+%5,5 bölümden beri
  Bu işlemin gerekçesi

  “You have Cloudflare, which is a cloud provider in a different sort, which is another $100,000,000,000 company.” — Eric Vishria

  1. Vishria argues AWS did not eat everything — Azure, GCP, and others all prospered alongside it.
  2. He points to Cloudflare as proof a differentiated non-hyperscaler cloud can reach $100B.
  3. Own Cloudflare as a differentiated cloud/infrastructure beneficiary of the same AI-adjacent dynamic.
• 6:34AMZNLongVishria uses AWS as the model: investors were wrong in 2007 that it was a commodity, and the hyperscalers all became 'unbelievable businesses' that will do well. — Eric Vishria-%8,9 bölümden beri
  Bu işlemin gerekçesi

  “So now, is AWS the biggest? I think it's like a fortythirtytwenty split. You ended up with an oligopoly of that.” — Eric Vishria

  1. Vishria claims the 2007 consensus that AWS was a low-margin commodity was 'massively wrong'.
  2. AWS became the anchor of a cloud oligopoly (~40/30/20 split) and its parent Amazon captured that value.
  3. Own Amazon as the parent of the dominant cloud infrastructure franchise in the AI era.
• 6:34MSFTLongVishria says Azure and GCP were irrelevant in 2014 but are now 'unbelievable businesses' and will do well, refuting the AWS-eats-everything thesis. — Eric Vishria-%2,1 bölümden beri
  Bu işlemin gerekçesi

  “Azure and GCP were irrelevant then. And fast forward to 2026 and they're unbelievable businesses.” — Eric Vishria

  1. Vishria claims the 2014 view that AWS would eat everything was wrong partly because Azure/GCP were dismissed.
  2. By 2026 Azure is part of the ~40/30/20 cloud oligopoly and a major AI compute provider.
  3. Own Microsoft as an oligopoly cloud/AI infrastructure winner.
• 38:49GOOGLLongVishria frames GCP as one of the 'unbelievable businesses' in the cloud oligopoly that survived the AWS-eats-everything scare, and notes its TPU as a winner. — Eric Vishria-%2,0 bölümden beri
  Bu işlemin gerekçesi

  “Obviously, the TBU itself is a winner, Tranium is a winner.” — Eric Vishria

  1. Vishria claims GCP was dismissed in 2014 but is now a major cloud business in the AI era.
  2. He explicitly cites the TPU as a winner among specialized AI chips.
  3. Own Alphabet as the parent of GCP and TPU, both AI-era winners.
• 1:12:10Fireworks AILongVishria's firm is an investor and he says Fireworks gets 5x speed and multiple-x throughput over cloud providers running the same open-source models on the same Nvidia hardware, and that it 'will do well'. — Eric Vishria–
  Bu işlemin gerekçesi

  “Will the fireworks of the world do well? Yes.” — Eric Vishria

  1. Vishria claims running large models is very hard and specialization drives 5x performance and multi-x throughput vs hyperscalers.
  2. His firm invested in Fireworks, which sells that inference-efficiency advantage while paying cloud margins.
  3. Own Fireworks as an AI inference/infrastructure winner.
• 13:17SierraLongVishria's firm backs Sierra; he calls it an application company doing real AI work, close to the models' 'jagged edge', acting as enterprises' AI sherpa starting with customer service and expanding via Horizon. — Eric Vishria–
  Bu işlemin gerekçesi

  “And one of the things that I think makes Brett so special is, and the team, is they're technologists, but they actually have lived in enterprise world for a long time.” — Eric Vishria

  1. Vishria claims winning AI companies understand the 'jagged edge' of model capabilities and build around it.
  2. Sierra, backed by his firm, does exactly that with agents, expanding from customer service to long-running agents via Horizon.
  3. Own Sierra as an AI application winner.
• 14:28CursorLongVishria repeatedly praises Cursor as the exemplar of building around the jagged edge of AI capability, evolving from IDE to autocomplete to agents, obsoleting its own work every six months. — Eric Vishria–
  Bu işlemin gerekçesi

  “You saw the evolution of Cursor from an IDE to like tab autocomplete to agentic work over and over and over again.” — Eric Vishria

  1. Vishria claims AI-era winners constantly evolve their product as model capabilities shift ('sandcastles, not castles').
  2. Cursor is cited as the prime example, repeatedly obsoleting its own prior work and capturing developer demand.
  3. Own Cursor as an AI-native developer tooling winner.
• 29:32Energy / power producersLongVishria says energy is the key bottleneck for AI — China is bringing on ~10x more energy than the US — making energy the area he's most worried about and most focused on. — Eric Vishria–
  Bu işlemin gerekçesi

  “I would say this. I am worried about energy.” — Eric Vishria

  1. Vishria claims models translate compute into intelligence, compute requires energy, and intelligence demand is effectively unlimited.
  2. China is adding ~10x more energy capacity than the US, potentially capping US AI capacity or raising token costs.
  3. Own/gain exposure to energy (solar, nuclear, gas, turbines) as the critical AI input.
• 54:57BenchlingLongVishria recounts Benchling's journey from no churn to seven years of churn in twelve months, but says the company found a path by applying AI for biotech/pharma customers and is 'crushing it'. — Eric Vishria–
  Bu işlemin gerekçesi

  “Benchling is you know, life sciences SaaS, the company's absolutely crushed, done really, really well.” — Eric Vishria

  1. Vishria describes Benchling facing a biotech crash with seven years of churn in twelve months.
  2. The team kept iterating, applying AI models to biotech/pharma workflows, and found a path.
  3. Hold/own Benchling as a life-sciences SaaS name that reoriented toward AI for its customers.

Invest Like the Best

Eric Vishria - A Decade of Lessons Investing in Software & Hardware - [Invest Like the Best, EP.486]

11 Ağu 2026971h

Eric Vishria shares lessons from investing in AI infrastructure and applications, arguing that AI demand is massive and will spawn an oligopoly of winners rather than a single dominant player. He highlights the difficulty of running large models efficiently, the strategic importance of energy, and the shift in competitive dynamics for SaaS companies. He expresses bullish views on AI compute, particularly Cerebras, and discusses emerging opportunities in robotics and CPUs, while warning about market timing and the need for differentiated execution.

Pozisyonlar · 7

• 35:33CBRSLongEric's experience with Cerebras shows it's a leading AI chip company with a wafer-scale architecture that solves communication bottlenecks, and he expects it to be a major winner in AI compute. — Eric Vishria-%13,7 bölümden beri
  Bu işlemin gerekçesi

  “Obviously, the TPU itself is a winner, training is a winner. You've had Grok and Cerebras etch. It'll keep getting fought out, but I think that will end up being big.” — Eric Vishria

  1. Eric describes Cerebras as a key AI hardware investment with a unique wafer-scale chip that maximizes cores, memory, and communication.
  2. He notes that each prior compute generation produced a $100B company (e.g., NVIDIA for GPUs), and AI is a huge new workload with specialized needs.
  3. Cerebras recently went public, and Eric's bullishness suggests investors should consider owning it for exposure to AI compute.
• 1:01:27NVDALongEric believes NVIDIA will likely do well as AI compute demand grows, despite intense competition, because of its dominant position in training and inference. — Eric Vishria+%1,1 bölümden beri
  Bu işlemin gerekçesi

  “Will the CSPs do well? Yes. Not all of them, but will some of these Neo Clouds do well? Yes. Will the fireworks of the world do well? Yes. Will NVIDIA do well? Yes.” — Eric Vishria

  1. In discussing AI infrastructure winners, Eric explicitly says 'Will NVIDIA do well? Yes.'
  2. He emphasizes that the AI market is so large that multiple players can prosper, and NVIDIA is a core beneficiary of the buildout.
  3. This is a stated view, not a detailed analysis, but it's a direct bullish call on NVDA.
• 3:35Fireworks AILongFireworks is a portfolio company of Eric's that achieves 5x performance and multiple times throughput on the same open-source models and NVIDIA hardware, showing it can undercut cloud providers and capture value. — Eric Vishria–
  Bu işlemin gerekçesi

  “The performance difference for a Fireworks versus a cloud provider is like 5x. And that is just the speed performance. Then you add on top of that throughput...” — Eric Vishria

  1. Eric highlights Fireworks' 5x speed advantage and multiple-x throughput advantage over cloud providers on identical models and hardware.
  2. He explains this efficiency allows Fireworks to profit while paying cloud provider margins, indicating a strong business model.
  3. He lists Fireworks as one of the companies that will do well in the AI infrastructure ecosystem.
• 13:02Sierra AILongSierra is an application company with deep AI expertise, acting as an 'AI Sherpa' for enterprises, and Eric believes it's poised to win by bridging model capabilities and customer needs. — Eric Vishria–
  Bu işlemin gerekçesi

  “Yes, it's an application company on the surface but they're doing real AI work and they're experimenting with models and they're building agents...” — Eric Vishria

  1. Eric describes Sierra as one of his 'singular investments' that is working well, with an enterprise-focused approach starting with customer service.
  2. He praises Brett Taylor and the team for understanding the 'jagged edge' of AI capabilities and building solutions around it.
  3. This indicates a strong bullish view on Sierra's business prospects.
• 42:07Sundae RoboticsLongEric is very bullish on Sundae Robotics, which is developing household robots with a focus on pre-training data pipelines, and he sees it as a leading robotics company. — Eric Vishria–
  Bu işlemin gerekçesi

  “We're investors in Sundae Robotics, which is going after this exact pipeline. It's a cool company, and they're doing household robots...” — Eric Vishria

  1. Eric describes Sundae as an investment and highlights their approach of gathering high-quality data via gloves designed for their robot hands.
  2. He sees their training pipeline as the key to unlocking robotics, and he's witnessed demos of robots folding laundry, which impressed him.
  3. This is a direct endorsement from a major investor.
• 1:03:17New LanternLongEric mentions New Lantern as an investment in AI for radiology, which aims to improve radiologists' efficiency, and he believes the AI copilot market will grow even if full automation takes time. — Eric Vishria–
  Bu işlemin gerekçesi

  “And actually, I think the studies and areas have shown that to be true, and we have an investment in a company called New Lantern which is approaching this...” — Eric Vishria

  1. Eric cites New Lantern as an example of a company tackling medical imaging AI.
  2. He argues that AI will initially augment radiologists, creating a copilot situation, and New Lantern is positioned in this space.
  3. His positive tone suggests he expects the company to benefit from this trend.
• 28:21Energy SectorLongEric is worried about energy as the key bottleneck for AI compute, and he advocates for investing in all energy sources, implying a bullish view on energy broadly. — Eric Vishria–
  Bu işlemin gerekçesi

  “If energy is what you need for compute and is the bottleneck and there's unlimited demand for intelligence, then it stands to reason that...” — Eric Vishria

  1. He states energy is the bottleneck for compute and that China is bringing on 10x more energy.
  2. He lists gas turbines, natural gas, solar, and rare earths as areas that will see activity.
  3. His view that 'we should do it all' suggests he sees opportunities across the energy sector.

Fikirler · 5

• 14:56CursorLongEric praises Cursor as an example of a company that understands the 'jagged edge' of AI capabilities and has built a successful product, suggesting it is a winner. — Eric Vishria–
  Bu işlemin gerekçesi

  “I think that's one of the things that Michael and team at Cursor did so well from the very beginning is they really understood the jagged capability...” — Eric Vishria

  1. Eric mentions Cursor repeatedly as a model of effective AI application development.
  2. He notes that Cursor's team understood the jagged capability and iterated on it, leading to its success.
  3. This is an endorsement of Cursor's strategy, though not a direct investment recommendation.
• 1:01:41ANTHROPICLongEric dismisses the idea that Anthropic will 'eat everything,' but clearly views it as a major AI lab that will be one of the oligopoly winners, likely to go public eventually. — Eric Vishria+%41,8 bölümden beri
  Bu işlemin gerekçesi

  “It's like, Okay, how do we cut up this pie and they're gonna eat this much and like, Oh, no, no, no, no, anthropic or whomever is gonna eat 98% of the value...” — Eric Vishria

  1. Eric references Anthropic in discussions about labs and the market structure, implying they are a key player.
  2. He discusses that labs like Anthropic will ultimately go public, citing the ability to raise capital and public trust.
  3. His view that 'it all works' suggests he expects Anthropic to do well.
• 47:39BenchlingLongEric discusses Benchling as a life sciences SaaS company that has done extremely well, and he worked with them through a tough period, implying a positive view. — Eric Vishria–
  Bu işlemin gerekçesi

  “If I take Saji and Benchling, Benchling is life sciences SaaS companies absolutely crushed, done really, really well.” — Eric Vishria

  1. He says 'Benchling is life sciences SaaS companies absolutely crushed, done really, really well.'
  2. He describes the company's struggles and how they adapted with AI, showing his ongoing support.
  3. While he doesn't explicitly say to invest, his description reflects a strong track record.
• 7:34NETLongEric uses Cloudflare as an example of a smaller cloud provider that became a $100B company, illustrating that the cloud market was large enough for multiple winners beyond the big three. — Eric Vishria+%5,4 bölümden beri
  Bu işlemin gerekçesi

  “And even outside of those big three, you have Cloudflare, which is a cloud provider in a different sort, which is another $100,000,000,000 company.” — Eric Vishria

  1. He mentions Cloudflare as a 'cloud provider in a different sort, which is another $100B company'.
  2. He uses it to argue that AI will similarly create oligopoly winners.
  3. This is an illustrative reference, not a direct recommendation, but reflects a positive view.
• 32:46TSMLongEric implies TSMC is a critical bottleneck in hardware supply chains, and its dominance likely makes it a winner as AI demand grows. — Eric Vishria+%2,6 bölümden beri
  Bu işlemin gerekçesi

  “There's of course TSMC, which everyone knows, it's not just TSMC. 30 other vendors that matter in putting all this stuff together...” — Eric Vishria

  1. He mentions TSMC as a key supplier and notes the supply chain risks, but also implies TSMC is essential.
  2. In the context of his bullishness on AI compute, TSMC is a necessary enabler.
  3. This is an inferred positive, not a direct call.

Invest Like the Best

Why the Markets Are Pricing AI Wrong | Gavin Baker

4 Ağu 20261135h

Pozisyonlar · 10

• 21:20NVDALongNvidia's dominance is underpriced given it is the most financeable chip, its new credit-wrapper/revenue-share model raises revenue per gigawatt, and it sits at its lowest forward P/E in a decade. — Gavin Baker+%5,3 bölümden beri
  Bu işlemin gerekçesi

  “Nvidia is actually, as we record this, at its lowest forward PE of the last ten years. The market, a 100%, thinks they are significantly over [earn]” — Gavin Baker

  1. Speaker claims every AI demand metric is accelerating, compute is in acute shortage, and installed GPUs are repricing higher off contract.
  2. Nvidia is the dominant GPU supplier, most financeable asset for lenders, and now layers a credit wrapper plus revenue share on GPU deployments.
  3. Resulting trade: buy NVDA because it is under-earning and trading at a decade-low forward multiple versus improving fundamentals.
• 1:15:12SPCXLongSpaceX is mispriced by public markets because its compute build-out capability (8 GW potential), Grok 4.5, Cursor acquisition, and lowest-cost cluster deployment are not in consensus estimates. — Gavin Baker+%31,4 bölümden beri
  Bu işlemin gerekçesi

  “I just think very little is built in, from my perspective, to that stock for the amount of compute that they might be able to bring on.” — Gavin Baker

  1. Speaker claims SpaceX has brought on more compute faster and cheaper than anyone and is monetizing at ~$50B per gigawatt.
  2. Grok 4.5 plus Cursor could add ~$10B ARR, and a potential 8 GW build would swamp the $73B consensus revenue estimate.
  3. Resulting trade: long SpaceX stock as very little compute upside appears baked into the price.
• 4:16MSFTLongMicrosoft's operating cash flow accelerated from 28% to 32-35% (adjusted), and it has a huge slug of capacity that hasn't even shown up yet, making it a beneficiary of compute repricing. — Gavin Baker+%3,5 bölümden beri
  Bu işlemin gerekçesi

  “Operating cash flow from Microsoft, Meta, and Amazon has reported accelerated from '28 to '32... if you adjust for that, we went from 28 to 35.” — Gavin Baker

  1. Speaker claims Microsoft's operating cash flow accelerated materially and that June capacity additions did not show up in Q2.
  2. Microsoft is a major hyperscaler whose economics improve as installed compute reprices higher.
  3. Resulting trade: long MSFT as under-earning hyperscaler assets reprice.
• 5:51METALongMeta did not cut Capex, released its best model in years (Use 1.1), and is renting out compute at a premium, yet the market misinterpreted the news negatively. — Gavin Baker+%15,7 bölümden beri
  Bu işlemin gerekçesi

  “They didn't cut CapEx. What it was is they saw SpaceX have a big installed base of compute and sell some big trading optimized clusters... at a truly massive premium.” — Gavin Baker

  1. Speaker claims Meta did not cut CapEx and that its compute rental was an opportunistic IRR play, not a bearish signal.
  2. Meta remains a top-4 hyperscaler whose operating cash flow is accelerating.
  3. Resulting trade: long META as the market's negative interpretation was wrong.
• 4:16AMZNLongAmazon's operating cash flow accelerated with the rest of the hyperscalers, and its Trainium chips plus AWS position it to benefit from compute repricing. — Gavin Baker-%8,6 bölümden beri
  Bu işlemin gerekçesi

  “Operating cash flow from Microsoft, Meta, and Amazon has reported accelerated from '28 to '32.” — Gavin Baker

  1. Speaker cites accelerating operating cash flow for Amazon alongside Microsoft and Meta.
  2. Amazon is a major hyperscaler with custom silicon and cloud demand leverage.
  3. Resulting trade: long AMZN as under-earning cloud assets reprice.
• 38:39MULongMemory is the single most important axis for tokens per unit of compute, and the LTA game theory means memory suppliers have durable pricing power and multi-year allocations. — Gavin Baker+%16,3 bölümden beri
  Bu işlemin gerekçesi

  “Memory is the more memory you put with FLOP for a given unit of compute, the more tokens you get out... It's the single most important thing you could do.” — Gavin Baker

  1. Speaker argues more memory per FLOP yields more tokens, making memory the dominant axis and demand inelastic.
  2. Micron is a major HBM/DRAM supplier with LTAs that customers cannot break without risking future allocations.
  3. Resulting trade: long Micron as LTA durability and memory intensity benefit suppliers.
• 53:20ASMLKaçınThe China DUV news was likely an overreaction for ASML in the near term, though the risk shouldn't be dismissed since China is 25 years behind and learning by doing. — Gavin Baker-%1,0 bölümden beri
  Bu işlemin gerekçesi

  “If this ever hits ASML's orders, maybe it hits it in five years. And, like, the market has forgotten about it, got worried about it... multiple times.” — Gavin Baker

  1. Speaker claims the market overreacted to China having a DUV machine and that any ASML order impact would be years out.
  2. ASML is the dominant EUV supplier, but China's DUV progress is a long-term competitive threat.
  3. Resulting trade: cautious on ASML on China DUV headlines, but the near-term impact is likely overdone.
• 1:09:00EtchedLongEtched's SRAM-based accelerators are not HBM-constrained and are made on older nodes, making them a positive for AI ROI by disaggregating inference into prefill, attention, and feed-forward network. — Gavin Baker–
  Bu işlemin gerekçesi

  “You could do better. You just can't beat SRAM in particular for that feed forward network... being able to disaggregate it to these three parts... is gonna be really, really positive for the ROI on AI.” — Gavin Baker

  1. Speaker claims SRAM-based accelerators are the ultimate holy grail for feed-forward network inference and don't compete with latest GPUs.
  2. Etched is a private startup (Baker is an investor) making architectural choices outside HBM DRAM constraints.
  3. Resulting trade: long (private) Etched as disaggregated inference drives AI ROI.
• 56:30Fireworks AILongFireworks AI is making it easy to customize open-source models with RL and routing, generating better-than-frontier performance at lower cost for AI natives. — Gavin Baker–
  Bu işlemin gerekçesi

  “Fireworks, they did come out with a really cool product called Nexus... it is literally three lines of code, like 20 words, and, Fireworks ingests your data...” — Gavin Baker

  1. Speaker claims Fireworks' Nexus product lets companies use proprietary data to fine-tune models with three lines of code.
  2. Fireworks is a private inference cloud enabling the open-source shift that increases aggregate compute demand.
  3. Resulting trade: long (private) Fireworks as the open-source inference layer grows.
• 1:10:33CognitionLongCognition's Scott Wu is a killer founder and the Cognition Index shows companies that spend the most on AI grow meaningfully faster. — Gavin Baker–
  Bu işlemin gerekçesi

  “Alfred Scott Wu, you know, cognition is kind of like You're here to that one. Yes. I think those are the most obvious names.” — Gavin Baker

  1. Speaker calls Scott Wu a dark horse to watch and references the Cognition Index showing AI spend correlates with faster growth.
  2. Cognition is a private AI coding company whose tools drive token demand.
  3. Resulting trade: monitor Cognition as a private AI-native beneficiary.

Fikirler · 4

• 3:01OPENAILongOpenAI is back in the game and massively accelerating, evidenced by its compute build-out and the Pareto frontier position it regained after Anthropic's earlier lead. — Gavin Baker+%34,3 bölümden beri
  Bu işlemin gerekçesi

  “OpenAI has accelerated. Anthropic continues to grow really strongly and is almost certainly pumping out significant amounts of free cash flow.” — Gavin Baker

  1. Speaker claims OpenAI is massively accelerating and got back into the game on compute.
  2. Private company has no listed ticker, but its growth drives the hyperscaler operating cash flow thesis.
  3. Resulting trade: monitor OpenAI as a key demand driver for the public AI infrastructure complex.
• 3:01ANTHROPICLongAnthropic is in pole position, growing strongly and almost certainly generating significant free cash flow, though third-party data suggests its curve may be flattening slightly. — Gavin Baker+%43,1 bölümden beri
  Bu işlemin gerekçesi

  “Anthropic continues to grow really strongly and is almost certainly pumping out significant amounts of free cash flow.” — Gavin Baker

  1. Speaker claims Anthropic is clearly in pole position and is likely pumping out significant free cash flow.
  2. Private company with no listed ticker; its growth is a key input to hyperscaler demand.
  3. Resulting trade: watch Anthropic's trajectory as a driver of AI infrastructure demand.
• 43:47SK HynixLongSK Hynix should copy Nvidia's credit-wrapper/revenue-share model because its memory LTAs trade upside for durability and provide a royalty on recurring revenue. — Gavin Baker–
  Bu işlemin gerekçesi

  “What would you do if you were... the CEO of Hynix? I'd do the exact same thing NVIDIA is doing right now... Which is?” — Gavin Baker

  1. Speaker says if he were Hynix CEO he would do exactly what Nvidia is doing now with credit wrappers.
  2. Hynix is a leading HBM/DRAM supplier whose LTA pricing power is underappreciated.
  3. Resulting trade: long Hynix as memory allocation game theory favors incumbents.
• 38:39AMDLongAMD is one of only four players at scale in the compute Game of Thrones, and the LTA game theory means it benefits from durable supply-chain positions. — Gavin Baker+%10,5 bölümden beri
  Bu işlemin gerekçesi

  “There's four companies that, like, matter at scale. There's Amazon with their tradiums. There's Google with their TPUs. There's AMD, and then there's Nvidia...” — Gavin Baker

  1. Speaker names Amazon, Google, AMD, and Nvidia as the four companies that matter at scale.
  2. AMD's position in the compute supply chain is reinforced by pre-purchase allocations.
  3. Resulting trade: monitor AMD as a structural player in the compute build-out.