Podcast Atlas
entity Updated 2026-08-21 Topics: Technology, Economics

Nvidia

Dan Loeb: The Lost Art of Short Selling, and Why Stock Picking is Back adds Dan Loeb’s active-investor view of Nvidia. Loeb says Nvidia can look undervalued on earnings over the next two or three years despite unprecedented scale, and warns that some long-short investors may treat it as a psychologically comfortable “safe short” in the way earlier skeptics underestimated Google or Amazon.

Anthropic’s $2T IPO, Zuck’s AI Manifesto, Nvidia’s $500B AI Bet, Grok’s Comeback adds Nvidia as the proposed financing layer for AI compute. The hosts say Nvidia is working with Goldman Sachs, BlackRock, and others on a $500 billion compute-financing plan, with GPU Compute Asset-Backed Financing treating GPU clusters as cash-flowing collateral whose value depends on utilization, useful life, residual guarantees, and continued model-company demand.

Inside America’s AI Strategy: Infrastructure, Regulation, and Global Competition adds Nvidia to the broader U.S.-China AI-stack comparison. David Sacks frames U.S. advantage as deeper in chips and semiconductor equipment than in models, while also saying China may discourage or block Nvidia imports to strengthen Huawei and domestic AI-chip production. The source extends Nvidia’s branch from export licensing into American AI Stack Strategy and Domestic AI Chip Catch-Up.

Howard Lutnick: How America Can Hit 6% GDP Growth in 2026 adds Nvidia through Howard Lutnick’s inside-government account of chip export licensing. Lutnick says Jensen Huang argued that a complete cutoff from China would push demand toward domestic national champions, while the administration allowed less-than-best chips under testing, license rules, and government revenue sharing. The source adds Nvidia H20, Nvidia H200, AI Export Controls, and Taxpayer-Return Industrial Policy to Nvidia’s policy branch.

150. 对英伟达研究副总裁刘洺堉的4小时访谈：Cosmos 3、世界模型、武术、黄仁勋影响我的，和你不需要击败所有对手 adds an internal research-and-platform account through Liu Ming-Yu / 刘洺堉. Liu says Nvidia builds and opens models partly to understand future developer needs before chips and infrastructure can be redesigned, and partly to help the Physical AI ecosystem focus on domain pain points instead of rebuilding every base capability. In this source, Cosmos Lab and Cosmos 3 are therefore part of Nvidia’s market-creation strategy, not only a model benchmark.

算力狂想曲，我在AI工厂的奇遇 adds a satirical cultural branch rather than a factual company update. The episode’s fictional Jensen Huang cooks GPUs in an AI Factory Allegory, turning Nvidia into the compute supplier that benefits whether the imagined future runs through agents, Physical AI, generated worlds, digital companions, or AI Circular Infrastructure Financing loops.

贾扬清：我所经历的「人工智能已死」到「AI 颠覆世界」的数年巨变丨串台「声东击西」S10E24 adds Nvidia as both the acquirer of Lepton AI and an organizational counterexample in Jia Yangqing’s account. The source connects Nvidia’s AI position to GPU ecosystem building, academic support, CUDA, tightly integrated infrastructure, and an unusually aligned company culture.

AI debt is flooding the bond market adds Nvidia to the bond-market side of AI infrastructure finance. The episode cites Wall Street Journal reporting that Amazon, Nvidia, and SpaceX accounted for $75 billion of one month of large-tech bond issuance, making Nvidia relevant not only as a chip supplier and valuation-risk focal point but also as a direct participant in AI Infrastructure Debt Financing.

E227｜美国医疗市场AI争夺战：巨头押注，创业公司能赢吗？ adds Nvidia as a healthcare AI infrastructure partner through Eli Lilly. 张璐 / Zhang Lu says the companies announced a strategic cooperation with an initial budget around $1 billion, using it as evidence that large pharma and AI-infrastructure actors now treat medical AI integration as urgent.

Nvidia is discussed in EP39 风满楼下集：全球衰退慢慢逼近，严防死守步步为营！漫聊下半年美股、美债、汇率 as the central example of a company that can be operationally excellent while still carrying high public-market expectation risk. 大雄 describes the company as strong in GPUs and AI infrastructure, but argues that investors still have to ask whether the market is pricing the moat, earnings growth, and future AI demand too perfectly.

EP76 穿越1940：我与股票大作手利弗莫尔的最后对话 uses Nvidia as a modern chart example for Trend Following. In that frame, the point is not to judge Nvidia’s business quality directly, but to show why buying throughout a downtrend differs from waiting for right-side confirmation and then adding only if the position works.

EP57 美股动荡，东升西降？这回是走是留 adds Nvidia to the mega-cap concentration and DeepSeek repricing discussion. The episode again avoids saying Nvidia is simply bad; instead it argues that investors may start asking whether AI capex across the ecosystem produces enough return, and whether a strong company can remain a good stock after high expectations are already priced in.

EP86 面子、底子、日子：财报只讲这三件事 adds Nvidia as a Financial Statement Analysis case rather than mainly a valuation case. The episode uses Nvidia’s revenue growth, high gross and net margins, large cash balance, light fixed-asset base, operating cash flow, free cash flow, and buybacks to show how Asset-Light Vs Heavy-Asset Models and Profit And Cash Flow Quality can make a chip designer’s statements look very different from a heavy-asset foundry such as SMIC.

把 AI 吹成核武器的人，亲手拉下了新冷战铁幕 adds Nvidia as the hardware comparison for AI Export Controls. The hosts argue that physical GPU export restrictions are at least legible through manufacturing and shipping chains, while model APIs, code, and weights are harder to regulate with the same tools.

Vol. 164 从苹果聊到软件未来：Agentic Software 真的要来了？ adds Nvidia as an ecosystem-building contrast to slower platform cycles. The hosts frame Jensen Huang as having treated GPU naming, academic support, and AI infrastructure as long-term ecosystem work, making Nvidia a case of informed FOMO rather than surface-level trend chasing.

E155.似乎没什么人再提「AI 泡沫论」了 adds Nvidia through the “five-layer cake” AI infrastructure frame. The episode uses chips, data centers, energy, cooling, and power demand to argue that AI token growth can turn Human Resource Deflation Compute Infrastructure Inflation into demand for hard infrastructure and Holo Assets, while still leaving stock valuation subject to AI Equity Valuation Risk.

商业小样43 | AI时代，谁在给服务器“降温” adds Nvidia as the power-density reference behind AI data-center cooling pressure. The episode says next-generation AI-system roadmaps could push rack density toward 600 kW, making Data Center Thermal Management a limiting condition for the GPU-based compute that supports MaaS Infrastructure.

134. 【数据的综述】和谢晨聊，新时代的石油、历史、版图、数据金字塔、定价与Recipe adds Nvidia as a robotics and physical-AI context. 谢晨 worked on autonomous-driving simulation around Nvidia-related infrastructure, and the source treats Nvidia’s physical-AI emphasis as one signal that Robotics Simulation Evaluation and embodied data infrastructure are becoming strategically important.

170: 【具身季报 26Q2】世界模型大风不停，和不想被贴标签的人 adds Cosmos 3 as Nvidia’s productized World Models marker for embodied AI. Chen Zhe Peter treats Cosmos 3 as a more open omni-world-model stack and uses Nvidia’s taxonomy of Video World Model, Action-Conditioned World Model, and World Action Models to explain why World Model VLA Fusion may matter for robot policies.

How convergence will define the tech sector in 2026 adds Nvidia as a public signal for Physical AI momentum. Amy Webb says Nvidia is openly betting on more robots in many shapes and sizes, making the company relevant to robotics infrastructure and AI Convergence, not only data-center accelerators.

EP90 从美加墨世界杯看懂期权—华尔街的终极武器 adds Nvidia as an option-selling example through Duan Yongping. The episode says selling calls on an existing Nvidia position can be coherent when the holder accepts sale above a chosen price and treats premium as cost reduction rather than free income.

Bytes: Week in Review - Micron’’s big earnings, Oracle’’s data center woes and “slop” is Merriam-Webster’’s word of the year adds Nvidia as the AI memory-intensity reference behind High Bandwidth Memory demand. Anita Ramaswamy cites the GB200 as having 192 gigabytes of memory per chip, making Nvidia systems a driver of AI Hardware Supply Chain Pressure for suppliers such as Micron Technology, SK Hynix, and Samsung.

AI is eating up the world’s computing memory adds a broader consumer-spillover version of the same memory point. The episode says high-bandwidth memory is paired with Nvidia chips and that AI data-center demand for memory can tighten supply for PCs, smartphones, gaming rigs, and AI PCs.

Bytes: Week in Review - New chip exports for China, Microsoft to pay electricity for AI data centers, and Gemini will power Apple’s AI adds Nvidia through the H200 export arrangement with China. Anita Ramaswamy says H200 chips can again be sold to China under new security rules and a 25% U.S. government sales cut, while Jensen Huang’s strategic argument is framed as keeping Chinese AI builders dependent on American infrastructure rather than accelerating Domestic AI Chip Catch-Up around firms such as Huawei.

存储三巨头破万亿市值，存储超级周期何时能见顶？｜ S10E13 adds a more detailed memory-hierarchy view of Nvidia. The source says each accelerator generation raises High Bandwidth Memory capacity, while Nvidia also explores NAND+DPU designs to prefetch PB-scale data and ease the Memory Wall without replacing HBM.

Bytes: Week in Review - SpaceX and xAI merge, Nvidia and OpenAI’s funding relationship and U.S. TikTok’s rough start adds Nvidia as a strategic counterparty to OpenAI. The episode says Nvidia’s planned investment of up to $100 billion in OpenAI was reported to be stalled, while Jensen Huang publicly rejected that framing and said Nvidia still believed in OpenAI. The source treats the relationship as continuing but more cautious: Nvidia wants OpenAI’s future data-center spending, but does not want to put all its chips on one company.

7000 亿美元砸向 AI：这是下一代互联网，还是泡沫重演？ | S10E12 adds Nvidia to the AI Circular Infrastructure Financing frame. Aaron uses Nvidia, OpenAI, and CoreWeave to ask whether AI infrastructure demand is independently grounded or partly self-reinforcing through investments, compute leases, and GPU purchases. The same source treats Nvidia chips and TSMC capacity as supply-constrained assets that can push customers into early capex, while warning that fast hardware iteration can make older GPUs depreciate faster if utilization or rental prices weaken.

TPU? GPU? What’s the difference between these two chips used for AI? adds Nvidia as the incumbent GPU ecosystem that Google TPUs may challenge for some AI workloads. Christopher Miller says Nvidia’s general-purpose chips remain the most commonly used across the AI ecosystem, and that Nvidia’s decade of software ecosystem work makes it difficult for specialized-chip challengers to displace the company even when they offer speed or power advantages in narrower workloads.

EP270 一枚芯片的漫长征途：我们离“算力自由”还有多远？ adds a public-explainer version of the same moat. 张从志 explains why GPUs fit deep learning’s parallel matrix work, then argues that domestic AI chips must compete with Nvidia’s hardware and CUDA-style software ecosystem together, not only with peak chip specifications.

E230｜1万亿收入预期背后：英伟达的巅峰与软肋 adds Nvidia at the peak of its own AI-infrastructure narrative. Jensen Huang’s GTC claim about at least $1 trillion in cumulative Blackwell and Vera Rubin orders becomes a way to test whether demand, Inference as Cash Flow, Token per Watt, Advanced Packaging, High Bandwidth Memory, data-center power, GPU Cloud Operations, and NeMo Cloud-style software can all align. The episode’s view is that Nvidia’s moat is now full stack, but that same breadth exposes the company to power, memory, interconnect, cloud-operations, and custom-chip pressure.

国产 AI 算力能凭「超节点」弯道超车吗？｜WAIC 深度观察 S10E23 adds GB200 NVL72 as the benchmark for Chinese supernode comparison. The source says Huawei CM384 may exceed NVL72 in aggregate compute through a much larger system, but keeps Nvidia’s advantage at lower cited power, CUDA, product stability, and customer familiarity.

真正改变世界的技术，为什么一开始都不被看好？| S10E16 adds Nvidia as the organizational countercase to Intel. 汪波 says Jensen Huang studied The Innovator’s Dilemma and tried to reduce incumbent complacency through flatter management and yearly GPU roadmap renewal, even when the prior generation still sold well.

So are we in an AI bubble? Here are clues to look for. adds Nvidia as the named focal point of a possible AI bubble. The Planet Money episode says Nvidia’s valuation depends on beliefs about whether AI chips will transform the world, making it hard to separate business strength from bubble risk. Robin Greenwood treats Nvidia’s valuation and volatility as warning signs while noting that other Statistical Bubble Indicators, especially issuance and acceleration, are weaker.

A recycling startup joins the AI boom adds Nvidia as an investor in Redwood Materials. The Marketplace Tech episode uses that investment to connect Nvidia’s chip-centered AI infrastructure role to the power-storage layer: data centers need energy access and Second-Life EV Battery Storage as well as accelerators.

Vol.115 全球宏观和资本市场2025展望：短期问题不解决，就没有中期和长期了 adds Nvidia as the high-end-chip-demand marker inside U.S. Mega-Cap Tech Right-Side Trade. Ricky argues that he had not yet seen clear evidence of reduced demand from large buyers such as Microsoft and Meta, so the source treats Nvidia and M7-style technology as a mature right-side trade rather than an immediately broken thesis.

E228｜谷歌TPU能撼动英伟达吗？前TPU工程师首次揭秘 adds the strongest TPU-specific challenge to Nvidia so far. Henry says Google’s TPU can compete in large, stable, high-volume training and batched inference workloads because XLA, TPU Pods, High Bandwidth Memory, and data-center deployment can lower TCO. The source still protects Nvidia’s core position: GPU flexibility, CUDA ecosystem depth, rapid workload adaptation, and full-stack execution remain valuable when models change faster than ASIC-like chip cycles.

没有方向盘的出行，走到哪一步了？ NVIDIA × 小马智行一次聊透智能驾驶 adds Nvidia’s automotive platform branch through 卓瑞 / Zhuo Rui. The source explains Nvidia in Robotaxi as more than a chip vendor: SoCs, CUDA, CUDA-X, training computers, simulation computers, vehicle-side inference, sensor support, redundancy, OTA, and safety processes all become part of Car-Grade Autonomous Compute and AI Infrastructure Full-Stack Moat when L4 vehicles must operate without human fallback.

Source Position

  • The Loeb source adds the opposite side of the usual valuation-risk warning: extreme market capitalization alone is not a short thesis if earnings power and AI demand still justify the price.
  • The episode treats Nvidia as a strong company, not as a fraud or failed business.
  • The risk frame is AI Equity Valuation Risk: if growth or guidance falls short of very high expectations, the valuation multiple can reset sharply.
  • Nvidia is described as a B2B supplier to large customers such as Microsoft, Google, and Amazon, so its revenue path depends partly on hyperscaler AI capex decisions.
  • Jensen Huang selling stock is used as a sentiment and valuation question, not as standalone proof that the business is deteriorating.
  • EP57 adds that DeepSeek can pressure the AI trade by changing expected return-on-capex narratives, not only by reducing demand for Nvidia chips directly.
  • Nvidia also becomes part of Mega-Cap Concentration Risk because broad U.S. index exposure can depend heavily on a few AI-related leaders.
  • EP86 uses Nvidia as a financially strong statement-analysis example, while leaving the market-price question to AI Equity Valuation Risk and Investment Risk Management.
  • The Keji Luandun export-control episode treats Nvidia as the physical-goods contrast to API and model-weight restrictions.
  • Vol. 164 treats Nvidia as the token-production and ecosystem-infrastructure contrast to consumer-platform release cadence.
  • Episode 134 treats Nvidia as a physical-AI and robot-simulation reference point, not only a chip or stock-market case.
  • EP90 treats Nvidia as a position-management example under Option Selling Discipline, not as a fresh valuation call.
  • The LateTalk source treats Nvidia as an embodied-model infrastructure company, where open world-model releases can support robotics even if Nvidia’s core business still monetizes compute and platform infrastructure.
  • The 商业就是这样 cooling episode treats Nvidia as a driver of rack-density pressure, not as the supplier of the cooling system itself.
  • The What’s Next storage-cycle source treats Nvidia as both HBM demand driver and memory-hierarchy optimizer: higher HBM capacity, Memory Capacity Lock-In, and NAND+DPU prefetching all matter.
  • The January 13 Marketplace Tech source treats Nvidia as the accelerator platform whose HBM demand helps explain consumer memory spillovers.
  • The 2026 Marketplace Tech Bytes episode treats Nvidia as strategically dependent on OpenAI demand but still incentivized to keep other model-company customers and cloud partners close.
  • The What’s Next S10E12 source treats Nvidia as both supplier and financing-loop participant: scarce GPUs support capex urgency, but circular demand and faster chip turnover make utilization, rental prices, and depreciation important risk signals.
  • The 2026 Marketplace Tech TPU/GPU episode treats Nvidia’s moat as both hardware and software: general-purpose accelerator flexibility and ecosystem depth remain valuable even as AI Chip Specialization grows.
  • EP270 treats Nvidia as the benchmark for Domestic AI Chip Catch-Up because the substitution target includes performance, software tools, developer habits, application adaptation, and cost-effective availability.
  • E230 treats Nvidia as an AI-infrastructure company whose strength is no longer reducible to CUDA or one GPU generation, while its risk is that orders must pass through packaging, HBM, interconnect, power, cloud operations, and customer deployment before becoming usable token capacity.
  • The WAIC supernode source treats Nvidia as the standard domestic vendors want to beat, while warning that larger system specs are not equal to surpassing Nvidia if the system uses far more chips, power, or migration effort.
  • S10E16 treats Nvidia as a company that actively tries to self-replace before customers force the shift, contrasting it with Intel’s missed GPU-era transition.
  • The Planet Money AI-bubble source treats Nvidia as the clearest example of valuation uncertainty around real AI infrastructure demand, not as evidence that the company itself is unsound.
  • The January 29 Marketplace Tech source treats Nvidia’s Redwood investment as a power-infrastructure signal, not as evidence that Nvidia itself is becoming a battery company.
  • Vol.115 treats Nvidia as part of Fact/Future Asset Pricing: current demand still supports the trade, but valuation depends on continued future belief in AI capex and technology adoption.
  • E228 treats TPU as a real pricing and workload-share pressure point on Nvidia, while keeping the displacement claim conditional on model stability, Google customer support, and HBM/packaging supply.
  • The 科技乱炖 Robotaxi source treats Nvidia’s moat as extending into automotive AI: production L4 deployment depends on car-grade compute, software compatibility, simulation, redundancy, and field support rather than only raw accelerator performance.
  • Episode 150 adds the builder-side world-model branch: Nvidia can open world foundation models because helping Physical AI developers may grow the whole compute and platform market.
  • The Lutnick source treats Nvidia as a strategic export-control counterparty: U.S. policy can preserve American platform influence in China while extracting taxpayer upside and testing chip profiles.
  • The August 14 All-In source treats Nvidia less as only a chip vendor and more as an asset-class designer: the company can reduce financing constraints if lenders accept GPU fleets as standardized, rentable, residual-value-backed assets.

Connections