No Nvidia H20 China sales; AI chip capacity; Correct view of the history of World War II
Summary of today’s top items:
1. Nvidia reports no H20 sales to China - Nvidia released Q2 earnings earlier today and the company announced it made no H20 sales to China in Q2 and was not forecasting any in Q3:
There were no H20 sales to China-based customers in the second quarter. NVIDIA benefited from a $180 million release of previously reserved H20 inventory, from approximately $650 million in unrestricted H20 sales to a customer outside of China...
Revenue is expected to be $54.0 billion, plus or minus 2%. The company has not assumed any H20 shipments to China in the outlook.
They did not name the “customer outside of China”; anyone know who it is?
The CFO said “We have not included H20 in our Q3 outlook as we continue to work through geopolitical issues…If geopolitical issues resolve, we should ship $2B to 5B in H20 revenue in Q3”. Neither she nor Jensen Huang, who also referenced “geopolitical issues”, specified what those issues are or mentioned any of the reported H20 pushback from PRC officials.
Huang did say that they continue to advocate for the US government to approve sales of the modified Blackwell chip to China, that China could be a $50 billion opportunity for Nvidia this year and that he expects that to grow 50% per year. He also reiterated the importance of making the American tech stack the global standard, an argument that may help him with President Trump but is anathema to Xi Jinping.
As I wrote last week:
The Financial Times reports that Commerce Secretary Lutnick’s comments around the July announcement of the approval for Nvidia to sell H20s to China were “insulting”, and have given ammunition to various constituencies in China to push back on purchases of the Nvidia chips. Lutnick told CNBC at the time that “We don’t sell them our best stuff, not our second best stuff, not even our third best…You want to sell the Chinese enough that their developers get addicted to the American technology stack…That’s the thinking”.
I think Lutnick’s comments just reinforce and strengthen the forces - security, bureaucratic and corporates like Huawei and SMIC - that were already pushing for ripping the Nvidia band-aid off and forcing indigenization. Sure, lots of the PRC firms want access to the best technology and don’t want to be reliant on Huawei, but Jack Ma and Alibaba didn’t want to go into the regulatory meat grinder, the platform companies didn’t want to be rectified, the education companies didn’t want their industry destroyed…But guess what? Party and security forces matter more than market forces or corporate preferences. The “keep China addicted to Nvidia’s stack” seems like just another variant of the “China Fantasy”. Too many people are too quick to dismiss the signal Xi sent at the April Politburo study session:
Xi Jinping emphasized that to seize the initiative and gain the advantage in the AI field, breakthroughs must be achieved in fundamental theories, methods, tools, and more. We must continuously strengthen basic research, concentrate resources to overcome challenges in core technologies such as high-end chips and foundational software, and build an independent, controllable, and collaboratively functioning AI foundational hardware and software system [集中力量攻克高端芯片、基础软件等核心技术,构建自主可控、协同运行的人工智能基础软硬件系统]. We should use AI to lead a paradigm shift in scientific research and accelerate technological innovations and breakthroughs across various fields.
The obvious tell will be if the Trump Administration approves sales to China of the modified Blackwell chip and Chinese firms purchase them in bulk. If PRC firms do that, we will know that the pushback on the H20 was more of a ploy to get this better chip and my take is wrong. But if PRC officials still discourage PRC firms from buying that much more advanced chip, then it will be hard to see how Nvidia sales to China recover, as the top leadership has set the course for accelerated indigenization of the PRC AI tech stack.
2. AI chip capacity in China about to ramp up - The Financial Times reports that 7nm chip fabrication capacity should increase significantly in the next several months. 7nm is far behind Nvidia’s leading-edge chips, but will it be good enough for the PRC policymakers to have confidence to push harder to squeeze out Nvidia?
3. More on the AI Plus initiative - I discussed the ambitious “Opinions on Deepening the Implementation of the “Artificial Intelligence+” Initiative” in yesterday’s newsletter, and provided a full translation. It is worth revisiting, and this section fits with the Financial Times story about the coming surge in chip production and the accelerating development of the PRC AI ecosystem:
III. Strengthen Foundational Support Capabilities
三、强化基础支撑能力
(7) Enhance foundational model capabilities. Strengthen research on AI fundamentals, support multi-path technological exploration and innovation in foundational model architectures. Accelerate research into more efficient model training and inference methods, and actively promote coordinated development of theoretical, technological, and engineering innovation. Explore new forms of model applications, enhance complex-task handling capabilities, and optimize interaction experience. Establish and improve model capability evaluation systems to promote effective iterative improvement of model capabilities…
（七）提升模型基础能力。加强人工智能基础理论研究，支持多路径技术探索和模型基础架构创新。加快研究更加高效的模型训练和推理方法，积极推动理论创新、技术创新、工程创新协同发展。探索模型应用新形态，提升复杂任务处理能力，优化交互体验。建立健全模型能力评估体系，促进模型能力有效迭代提升。
(9) Reinforce coordination for AI computing power. Support breakthroughs and innovation in AI chips and the cultivation of enabling software ecosystems; accelerate technological breakthroughs and engineering deployment of ultra-large-scale AI-computing clusters. Optimize the national layout of AI-computing resources; improve the national integrated computing network; give full play to the role of the national “Eastern Data, Western Computing” hubs; and strengthen coordination among data, compute, electricity, and networks. Enhance interconnection and supply–demand matching for AI computing power; innovate operating models for AI-computing infrastructure; encourage development of standardized, scalable compute cloud services; and promote AI computing that is inclusive and easy to use, economical and efficient, and green and secure…
（九）强化智能算力统筹。支持人工智能芯片攻坚创新与使能软件生态培育，加快超大规模智算集群技术突破和工程落地。优化国家智算资源布局，完善全国一体化算力网，充分发挥“东数西算”国家枢纽作用，加大数、算、电、网等资源协同。加强智能算力互联互通和供需匹配，创新智能算力基础设施运营模式，鼓励发展标准化、可扩展的算力云服务，推动智能算力供给普惠易用、经济高效、绿色安全。
(11) Promote a thriving open-source ecosystem. Support the building of AI open-source communities; promote the open aggregation of models, tools, datasets, and more; and cultivate high-quality open-source projects. Establish and improve mechanisms for evaluating and incentivizing AI open-source contributions; encourage universities to include open-source contributions in student credit recognition and teacher performance assessment. Support enterprises, universities, and research institutes in exploring inclusive and efficient new models for open-source applications. Accelerate the formation of an open-source technology system and community ecosystem oriented toward global openness and develop open-source projects and developer tools with international influence.
（十一）促进开源生态繁荣。支持人工智能开源社区建设，促进模型、工具、数据集等汇聚开放，培育优质开源项目。建立健全人工智能开源贡献评价和激励机制，鼓励高校将开源贡献纳入学生学分认证和教师成果认定。支持企业、高校、科研机构等探索普惠高效的开源应用新模式。加快构建面向全球开放的开源技术体系和社区生态，发展具有国际影响力的开源项目和开发工具等。