Topics Pioneers Insight Method Author
Back to Pioneers
马克汤
Investors 2 Curated Dialogues

马克汤

Rendering the long image…

Key Views & Dialogues

August Chatter: AI and Robots Are Changing the World—Are You Paying for It? — 马克汤|AI Finance Buddy

苔藓之火

Nvidia is joining BlackRock, Brookfield, Goldman and KKR to finance data centers as REIT-like assets, while guarantees remain uncertain. If demand is disproved, prices could halve; A100 prices, volumes and CDS offer indicators, while DeepSeek V4’s ten-minute switching tests AI economics.

View summary (8)
  • Nvidia has brought in BlackRock, Brookfield, Goldman and KKR to form a lending syndicate and package data centers into REIT-like assets for financing; the market is also debating whether Nvidia is providing guarantees, but Raymond says that claim is far from certain. 马克汤认为老黄的“卡有残值”叙事在单点违约时可以自圆其说,但真正的违约可能发生在全行业需求被证伪之时,届时“芯片残值本身也会下降得非常猛烈”,是系统性风险而非单点风险。Raymond 补充:供过于求只要 3-5 个百分点,就可能足以让市场价格腰斩;存储缺口 5-10 个百分点,则可能把价格推高两三倍。
  • Michael Burry’s fund was tiny, with roughly $150M in AUM before it was shut down; his Substack, Cassandra Unchained, has more than 300,000 paid subscribers paying over $300 a year and nearly $400, implying roughly $100M in annual revenue. 他的核心控诉是循环借贷会崩,以及科技公司折旧造假(芯片应像 iPhone 一样快速 write off)。两位主播的反驳很具体:8 年前的 A100 现在仍能租出 5 年长期租约,价格坚挺、流动性好,因为推理需求分层、客户多元化,本地部署、蛋白质研究、材料科学等场景都可能不需要顶级算力——“他是坐在办公室里做预测的”,应该去硅谷和数据中心转一转。
  • Early warning indicators for a bubble include the marginal moves in CDS premiums for Nvidia, CoreWeave and Oracle, along with A100 prices and trading volumes—“I’m certain GB300 is hard currency; I’ll buy as many as I can,” while A100 is a more marginal asset and potentially a better leading signal. 马克汤则强调看“链外”的真实用户需求:产业链内部互相刷单能把双方 ARR 都做上去,“我们赚的钱没有多,甚至还要额外交税”;自由现金流、存货和应收账款都值得看,但公开财务数据可能滞后且不透明。
  • China’s open-source-model “kill line” has become a life-or-death threshold for AI application companies: a model must either outperform DeepSeek V4 or cost less than V4, otherwise “most models have no reason to exist”—and switching software takes only 10 minutes, with memory and skills carrying over. 马克汤指出开源模型也开始分梯队:Kimi K3 是 2.8T 参数模型,部署成本较高,需要约 64 张卡才能达到较好的部署效果;DeepSeek 可能只需 4 张或 6 张,具体数字两人并不确定。K3 定价对齐 GPT-5.6 Terra 档,甚至比最便宜的 Luna 还贵;GLM-5.3 Flash 这类小模型才守在斩杀线附近。
  • Distillation is the episode’s biggest fault line: 张一鸣 defines it as negative—students who copy the teacher’s homework can never catch up, while organizations fill up with “hackers” instead of people who actually build data and pre-training; Raymond flatly disagrees: “If a frontier-model company is willing to distill others and does it well, I see that as a positive for the company and would be willing to buy it.” 他用拼多多类比:在二选一的环境下砍一刀“蒸馏”了微信社交关系,Temu 则被马克汤描述为抽取供应链精华做白牌,先模仿后超越;马克汤认为中美都在互相蒸馏,中国模型还会公开技术报告,DeepSeek 的 KV Cache、Kimi 的 KDA Attention 等架构优化也被海外模型参考。
  • The token-quota era has arrived: companies in China and abroad are limiting employees’ use of closed-source models—“closed-source models have become unaffordable, while open-source models are usable”—and more quota now requires a productivity case; the age of “token maxing” is over, and “real results have to be delivered.” 马克汤判断国产推理卡普及正在加快,但从他看到的报道来看,尚未到完全供应不上的阶段;Raymond 用科威特类比提出疑问:开采成本最低但 Q 不够大,所以 Anthropic 的 P 并没有被逼降一百倍。
  • On robots, 马克汤 sees the endgame as UBI leading to UHI and agrees that current use cases, capabilities, BOM costs and end demand still do not work—robots are staging sports meets, performing martial arts, cutting fruit and throwing out their backs while running, while factory deployments are mostly ASIC-like single-task machines rather than GPU-like general-purpose systems; half or more of China’s 1,000-plus robotics startups may die. Raymond 的爆帖直指要害:连洗碗机都不用的中国人,怎么会接受家务机器人进家门?两人都看好行业长期发展,但不等于看好单一企业;马克汤认为宇树估值高或短期目标难达成并不奇怪。
  • Macro-wise, a chart cited by 马克汤 is worth remembering: long-end yields could fall under either AI outcome—successful AI drives productivity and deflation, while a burst bubble sends money into Treasuries—possibly by the end of 2027, once the AI outcome becomes clearer. 眼下贝森特的 Treasury Twist(发短债、买长债)“第一天有反应,一个星期之后全部回来”,长端利率处于 2007 年以来最高;30 年期收益率从 5.3% 降到 4.7%-4.8% 时,股价可能不同,而利率走高会直接抬升 Raymond 所说的自由现金流已转负的大厂的融资成本、压低 ROIC。

Read the deep dive →

Listen to full conversation →

40. Can AI Applications Raise the Next Round? Acquisitions, Gray-Zone Plays, Intermediaries, and Big-Tech Hunting — Conversation | 马克汤

苔藓之火

AI applications are being judged by ARR, profitability, and ecosystem position rather than DAU, while capital remains concentrated in foundation models and hardware. Cursor’s roughly 75% Claude API gross-margin cost pressures its model, while Seedance’s account pools and intermediaries highlight supply-chain leverage; watch model vendors’ entry and AI retrofits of traditional companies.

View summary (7)
  • 马克汤 was categorical that Anthropic would never, ever acquire Cursor: “It doesn’t need to acquire it; you’ll die on your own anyway.” The logic: Cursor is essentially an advanced, highly engineered wrapper, but it has to pay Anthropic the cost implied by Claude API’s roughly 75% gross margin; over time, it cannot beat Claude Code. On the same $40/$100 plan, Cursor offers far less Opus usage—“it’s not that it doesn’t want to give you more; it can’t afford to.” Raymond said Musk bought Cursor into SpaceX for $60B; 马克汤 believes the asset is the “intersection” for high-value tasks, and that connecting it to X’s data centers would make the gross-margin model work. Whether the deal uses SpaceX stock is only Raymond’s speculation; the $10B breakup fee is 马克汤’s description of an option-like structure.
  • AI applications have three possible moats—data advantage, data-cleaning “dirty work,” and inference efficiency—and 马克汤 believes efficiency is what survives. Model labs “haven’t eaten them yet only because they haven’t had the bandwidth”: enterprise services are labor-intensive, while model companies pursuing AGI do not want the distraction. Vertical small models are essentially out of the conversation by 2026: “They’ll all be steamrolled by the emergent intelligence that comes with a sufficiently large parameter count.” Raymond’s analogy is that hiring generalists from Peking University or Tsinghua will beat specialists from lower-tier universities even with the same training time. He also cited OpenEvidence buying access to major U.S. journals as a reminder that data advantages can be weakened if model vendors simply buy the rights.
  • Account pools and intermediaries form an AI-application supply chain, and they were also part of Seedance’s success. Users resell unused Codex 20x keys; unused enterprise token allocations may likewise be converted into pocket money. Cloud providers’ trial and credit balances flow into account pools, allowing intermediaries to undercut direct purchases from Amazon and Azure—the competition is supply-chain management. Volcano Engine generates RMB1B a month from Seedance alone and raised its 2026 target from RMB10B to RMB15B, with wrappers such as LibTV making an important contribution.
  • AI-app gross margins structurally cannot match traditional software: application companies effectively pay model-company gross margins as costs. Traditional software routinely carries 80%-90% gross margins; for an AI application, getting above 50% or 60% is already exceptional. Many to C/prosumer products are negative-margin businesses, including Manus in 马克汤’s view; positive margins are concentrated among niche vertical users willing to pay a steep premium. The fate of companion products is even more explicit: “Borderline content is the easiest way to make money; without it, honestly, making money is pretty hard.”
  • ByteDance’s Seed lineup is underappreciated, and Feishu is due for an AI-era re-rating. Doubao feels good but benchmarks poorly because its free product assigns users to different model tiers—“the only thing that can respond quickly is a small model.” The newly released Seed 2.1 is already benchmarking in the GPT-5.5 and Opus 4.7 tier, with particularly strong multimodal capabilities; its audio performance benefits from Douyin data. Feishu had already built comprehensive interfaces before the AI era and is “native MCP”; in 马克汤’s Codex/Claude Code workflow, its skills rank second in usage frequency. ByteDance rates itself “at least an 8” after 2 years, with its focus on Doubao as the winning choice.
  • The most valuable AI application may not be an app at all, but the acquisition and retrofit of a traditional company. The Thrive Holdings-style roll-up and AI insurers that write their own policies and underwrite them follow the same logic: licenses, SOPs, and customer relationships already prove the business model; what is missing is a scaling engine. OpenAI and other model companies are also forming JVs with PE firms, potentially covering 500 portfolio companies across 5 PE firms. Raymond’s analogy is that the winner in real-estate internet was Beike, with physical stores—not a pure lead-generation app.
  • Asked whether to spend RMB100 on an AI application or on storage, optical components, or PCBs, 马克汤 chose the application: “If applications don’t take off, all the hardware is a bubble.” He invoked Huang’s five-layer cake: “Consumers don’t pay for the packaging technology inside HBM… They pay for products that improve the experience, optimize efficiency, and make life better.” But pure-play AI-app names remain scarce in public markets; Meitu, BlueFocus, and Chinese Online do not qualify—“the product function has to be sufficiently AI-native, and the organization has to be sufficiently AI-native.”

Read the deep dive →

Listen to full conversation →