The US-China Battle Over NVIDIA: AI Chip Regulations and the Global Tech Capital Game
This article organizes the current intersection of tech, geopolitics, and capital markets. The discussion covers six main points: (1) China's request for a 'halt in purchases' of NVIDIA chips, (2) the AI infrastructure investment rush and politics in the US and UK, (3) the framework uncertainty surrounding TikTok's US operations, (4) StubHub's IPO, (5) Groq's large-scale funding and inference demand, and (6) autonomous driving, data centers, and Meta's hardware-AI strategy.
1. The Impact of China's 'NVIDIA Chip Order Cancellation' Request
It has been reported that Chinese internet regulators have asked companies like Alibaba to cancel orders for NVIDIA's downgraded AI chips for the Chinese market. The program also provided commentary that 'China is reluctant to accept the latest performance-suppressed versions,' suggesting NVIDIA's desire to obtain sales permits from the US government for higher-performance chips (e.g., beyond the H20).
The market reacted immediately, with NVIDIA stock falling for three consecutive trading days, and it was pointed out that the total market capitalization had decreased by over $200 billion. This new phase of geopolitical risk makes (1) China's negotiation pressure (buying time for domestic suppliers to catch up), (2) the redesign of US export controls, and (3) the feasibility of providing the Blackwell generation to China even more uncertain.
1-1. The Negotiation Arena is the 'Top-Level Summit'
The commentary emphasized that 'the US-China summit on Friday is the real climax.' It mentioned the possibility of handling TikTok and export control reviews as a 'package.' For NVIDIA, defining the 'sellable specifications' is the most important agenda.
2. The Link Between US-UK AI Infrastructure Investment and Politics
The program reported that Microsoft, OpenAI, and others are investing 'billions to tens of billions of dollars' in the UK, highlighting a broad capital commitment that includes quantum and cloud computing. While this is a political tailwind for the British Prime Minister, it is suggested that it could also serve as a bargaining chip for the US President (Trump, in the context of the program) regarding digital services tax (DST) negotiations.
Conclusion: The location, power, and skilled talent for AI data centers are directly linked to the practical benefits of inter-state negotiations. Attracting capital is also a redistribution game of regulations, tax systems, and standardization.
2-1. The Visible 'Construction Bottleneck'
Citing a statement from the CEO of a construction platform company, it was noted that DC construction investment in the US is at a scale of $40 billion per year, with rising material costs (+44% compared to February 2020) and a shortage of electricians (on the scale of 500,000). The trend of halving construction periods through pre-fab/modular methods was also cited as a concrete example.
3. The TikTok 'US Version' Concept: The Core is the 'Algorithm'
While the framework for acquiring the US business by a consortium including Oracle is progressing, experts emphasized that 'the key is the handling of the algorithm.' The degree of ByteDance's involvement and legal compliance are the biggest uncertainties, and the White House's stance that 'it is in the realm of speculation until details are confirmed' was also cited in the program.
If the 'limbo' of users and advertisers continues for a long time, it could cause temporary misallocation of advertising demand for competitors (e.g., Meta).
4. StubHub's IPO: A Three-Part Story of Regulation, Transparency, and AI
Ticket secondary market StubHub went public at $23.50. The CEO stated that 'the passage of All-in Pricing at the federal level is an improvement in the consumer experience,' and welcomed the DOJ's market opening pressure, saying, 'We are on the side of the fans.'
In terms of performance, growth is cautious at +3% early this year, but they claim that 'excluding one-time factors (AIP introduction adjustments, Taylor Swift tour reaction), the underlying trend is over 20%.' They stated that AI enables the 'simultaneous achievement of revenue expansion and cost reduction' in CS, search, and feature development, and also demonstrated the superiority of their proprietary data.
5. Groq: The 'LPU' Design Philosophy of the Inference Era
Groqcompleted a funding round with a valuation of $6.9 billion. The CEO stated that 'inference demand is insatiable,' and emphasized the economies of scale where models are distributed across many chips using the LPU architecture, making it 'faster and cheaper as the number of chips increases.'
Key point: For learning, 'throughput' is key, while for inference, the balance of 'sequentiality and parallelism' is key. The 'division of roles' between GPUs and LPUs is becoming a realistic solution, and examples of avoiding facility constraints, such as utilizing existing DCs that do not require liquid cooling, were specifically shown.
6. Autonomous Driving, Fleet Operations, and Meta's 'Hardware x AI'
Lyft and Waymo have partnered in Nashville. Lyft is focusing on its "asset-light" functions of demand generation and fleet management, taking on "24/7 operational optimization similar to aircraft maintenance." They are also factoring in city-by-city coexistence and competition with Uber and self-operated fleets, while keeping an eye on potential partnerships with Chinese players and new entrants in Europe.
Meanwhile, at Meta Connect, the core focus was the "hardware-software integration" of Ray-Ban smart glasses and Meta AI. RBC analysts pointed out that "differentiation lies in designing continuous usage value through software." While the experiential value of features like simultaneous interpretation was not as highly valued by the stock market as Apple's announcements, the search for a "killer use case" for daily adoption is now at a critical juncture.
Summary
The export control and negotiation game centered on NVIDIA, the avalanche of AI infrastructure capital into the UK, TikTok's algorithmic sovereignty, StubHub's regulatory compliance and data-AI integration, Groq's inference-specialized architecture, the operational division of labor between Lyft and Waymo, and Meta's hardware integration—the common thread is the acquisition and allocation of "Compute Capital." Who can generate the most inference with the least power, materials, and time? The keys to the next quarter lie in politics, supply chains, and design philosophy.
