Innovators
2 Curated Dialogues
姚欣
Key Views & Dialogues
29. Inside 2025 GTC (Part II) | The Weight of the Crown: How 黄仁勋 Answered Every Challenge
Nvidia’s 2025 GTC shifted the growth narrative from training to inference, where reasoning may drive much higher token and compute demand. Dynamo’s improvements to cluster deployment, scheduling, and utilization support an AI-factory model, but post-GTC weakness and possible AMD pressure or foundation-model convergence leave CUDA’s moat unresolved.
View summary (8)
- At 2025 GTC, Nvidia shifted its core growth narrative from training to inference. 姚欣 said the industry had recognized by June-July 2024 that the marginal returns from training-side scaling laws were weakening. DeepSeek turned that view into mainstream consensus: the next battleground would be cost-performance, applications and token costs. 黄仁勋 barely mentioned training, instead repeating “token token token” and arguing that inference still requires massive compute—roughly 20x the tokens for a single answer, and potentially 100x the compute.
- Behind Blackwell Ultra, Rubin and the 3-year roadmap was the capital market’s demand that Nvidia prove shipments and growth immediately. Early GB200 products had reportedly raised thermal and system-stability concerns, while DeepSeek triggered questions about whether the architecture itself was changing. Robots therefore gave way to nearer-term delivery commitments: major cloud providers bought roughly 1.2 million Hopper GPUs last year and have now ordered about 3.6 million Blackwell GPUs. “Tell me how many units you can ship this year, and how many next year” is the pressure point. 姚欣’s summary: “Nvidia is in a rush”（英伟达急了）.
- The key impact of DeepSeek is not the disappearance of compute demand, but whether Nvidia can replicate its training-era monopoly in inference. 卫诗婕 cited 袁金辉’s metaphor: “The compute required for training is a swimming pool; the compute required for inference may be a river.” Compute in that river is more distributed and scheduling-intensive, creating room for cost-effective challengers such as AMD. The post-GTC share-price decline shows the market has yet to accept the full “buy more, save more” argument.
- Dynamo is Nvidia’s most concrete software answer to the inference era—and may also be a rapid catch-up effort prompted by China’s open-source practice. It packages the data center as an “AI factory” that produces tokens, with its introduction featuring terms such as PD disaggregation and expert parallelism, while managing clusters and optimizing deployment and utilization. 姚欣 sees its approach as broadly parallel to the components DeepSeek presented during its February open-source week. 卫诗婕 also speculated that parts of the architecture resemble Kimi’s Mooncake; 姚欣 said open-source projects may borrow from one another. NVL72 sells a cluster of 72 GPUs, while Dynamo adds deployment, scheduling and utilization optimization.
- What DeepSeek demonstrated was the ability to extract gains through systems engineering, not a way around Nvidia. High-Flyer’s quant background led the team to focus on low-level efficiency for years. By calling PTX, an assembly-like language inside CUDA, it pushed Nvidia hardware harder—but may also have made the code more “Nvidia-only.” 姚欣 rejected the idea that only a handful of people globally can do this, arguing that many teams with more than 10 years of infrastructure experience have similar capabilities.
- CUDA’s ecosystem moat remains formidable, but the consolidation of foundation models is lowering the cost of porting across hardware. Nvidia once cultivated developers by giving away cards, placing engineers on-site and building out cuDNN, creating a default path of “get it working on CUDA first, then port it elsewhere.” If foundation models converge around a handful of systems, vendors such as AMD will find it much easier to support them one by one. 姚欣 therefore expects the software moat to erode over a 10-year horizon; 80-90% gross margins and 25-30x P/E multiples will be difficult to treat as permanent.
- The four stages—perception, generative, Agentic and physical AI—keep pushing Nvidia’s long-term TAM into the physical world. 姚欣 compares them with biological evolution: first senses, then the brain, then digital limbs and finally physical limbs. 卫诗婕 sees a complete reflex chain of perception, thought, agency and action. If physical AI is to simulate friction, bouncing and every physical law, the ultimate constraint is energy rather than imagination: “Nvidia’s ultimate adversary is the laws of physics.”
- China is both a major market Nvidia has lost and a source of technological pressure forcing Silicon Valley to iterate faster. Based on his reading of financial filings, 姚欣 said China once accounted for roughly 24% of Nvidia’s global chip sales; 卫诗婕 added that the share fell below 10% after the sales restrictions, which 姚欣 attributed to the large number of chips barred from sale. Apart from Lenovo, there were almost no Chinese booths at this GTC, and China AI Day moved online. Silicon Valley is nevertheless scrambling for Manus invitations and beginning to acknowledge China’s advantages in cost-performance, applications and data: “Although nobody talks about China, China has shaken Silicon Valley”（“虽然大家不讲中国，但是中国对于整个硅谷产生了冲击”）.
28. Inside GTC 2025 (Part I) | Nvidia’s History: How One Company Kept Betting on the Future for 30 Years
Nvidia’s moat was built by turning graphics GPUs, GPGPU, CUDA and clusters into platforms over 20-30 years, not by one AI bet; training may still exceed 50%-60% share. Inference is the structural pivot toward one leader and many specialists: revenue could meet or exceed expectations over 2-3 years, while margins or monopoly power decline, with 288GB memory and delivery timing as near-term tests.
View summary (7)
- Nvidia’s dominance was not the result of a single winning AI bet, but of turning future applications into platforms over 20-30 years—from graphics GPUs and GPGPU to CUDA and supercomputing clusters. 姚欣 believes the hardest thing to replicate is Jensen Huang’s “patience with the vision”: a decade ago, he was still personally promoting Tegra from company to company; in 2024, about 80% of Nvidia’s workforce was working on software, while demos were seeding future demand for compute in the metaverse, quantum computing, humanoid robots and other areas.
- Nvidia remains in a league of its own in training, but inference will likely evolve from a monopoly into a market with one dominant player and many specialists—the central structural turning point for its valuation. 姚欣’s framework: Nvidia could hold more than 50%-60% of training, while a 20% inference share would already make it a strong leader; as specialized chips develop, it could also fall back to being an ordinary participant with 5%-10% share, because the “impossible triangle” of cost, performance and power consumption will generate a large number of specialized solutions.
- Nvidia’s near- and medium-term revenue and its long-term share price can move in opposite directions: revenue could still meet or exceed investor expectations over the next 2-3 years, but at least one of its high gross margins and monopoly position will decline. 姚欣 describes the current setup as “profits are good, growth is good, and the market is monopolized,” and says that this three-way combination “will never last long in any period of technology history”; AMD, in-house XPUs and Chinese supply are unlikely to create a more meaningful challenge until 3-5 years from now.
- DeepSeek did not make compute unnecessary; it shifted incremental demand from training toward inference, forcing the market to recalculate the cost, memory requirements and delivery cadence of each inference run. 姚欣 is watching to see whether the new card reaches 288GB of memory, whether it also has IP4 capabilities (not elaborated in the original), and whether one machine can run the “full-strength version of DeepSeek”; if deliveries begin by year-end, Nvidia’s inference share could return to 60%-70%, but a delay until June next year would be “below expectations.”
- The campaign against Nvidia is not being waged by one rival, but by giants that are simultaneously buying GPUs, backing AMD and ROCm, developing specialized chips such as TPU, and investing in startups. AMD, the main traditional competitor, “has all the capabilities, but not the ecosystem”; Google has built a small closed loop around TensorFlow, TPU, the cloud and internal demand, Broadcom is taking the custom-chip business, and 姚欣 believes only Huawei can currently discuss competition with Nvidia in selected domestic markets.
- Jensen Huang is actively reducing dependence on major customers by combining chips, capital and financing leverage into a second demand network. From Oracle Cloud to CoreWeave as a “white glove” intermediary and then to “sovereign AI” for governments around the world, the logic is to prevent a handful of hyperscalers from absorbing most of the demand; when 卫诗婕 asked whether this had become a financial-capital game, 姚欣 compared it directly with “the next-generation LeEco,” suggesting that the strategy combines customer-mix optimization with short-term excess.
- Higher inference efficiency will not automatically end compute growth, because cheaper inference will expand usage; images may consume roughly 1000x as much compute as text, and video could add another 1000x. 姚欣 estimates that inference could rise from about 40% of compute in 2024 to more than 95%, while multimodality could expand the total market by “1000x or 10,000x”; Nvidia will still help “build the infrastructure for the next generation of human civilization,” but “it will not be the only one.”