Chinese President Xi Jinping delivers a speech during a reception at the Great Hall of the People to mark Martyrs Day, September 30, 2025 (Adek Berry/Getty Images)

China steps up customs crackdown on Nvidia chips, launches antitrust investigation into Qualcomm, and plans special port fees on US ships

Beijing is doubling down on protectionism ahead of a planned Xi Jinping and Donald Trump meeting set for later this month.

David Crowther

10/10/25 5:56AM

For months now, China has been getting increasingly defensive over its domestic industries, particularly the all-important AI hardware space. This morning, we got the latest measures from those continued efforts.

First, the Financial Times reported that China has mobilized teams of agents at major ports across the country to “carry out stringent checks on semiconductor shipments.” The initial goal is reportedly to stop local tech companies from buying Nvidia chips, most notably the tech giant’s H20 and RTX Pro 6000D models, which Beijing has become particularly focused on stopping from entering the country. According to the FT, one person familiar with the matter also said that the more rigorous enforcement had been widened to all advanced semiconductor products.

Separately this morning, news broke that chip giant Qualcomm was the subject of a new antitrust investigation from China’s State Administration for Market Regulation over its acquisition of Israel’s Autotalks. Qualcomm fell 3% in early trading. In September, Nvidia itself fell foul of the same Chinese regulator over a 2020 acquisition.

Outside of AI, China is also planning to impose special import fees on vessels owned by US individuals, companies, or organizations, in a retaliatory move to a similar policy the US revealed back in April.

Per The Wall Street Journal, vessels docking at Chinese ports will be charged 400 yuan per net ton from October 14. That’s equivalent to ~$56. That fee is also set to rise over time, hitting 640 yuan per net ton in April 2026, 880 yuan the year after, and 1,120 yuan from April 2028.

The escalation of trade tensions between the world’s two most important economies comes ahead of a planned meeting between Chinese President Xi Jinping and US President Donald Trump, slated to take place at the end of the month at the APEC summit. Yesterday, CNN reported that Beijing had ramped up sweeping restrictions on rare earth exports.

Luke Kawa

The threat of AI disruption has dawned on the bond market

22V Research economist Peter Williams noted that medium-term rate expectations have disconnected from recent encouraging labor market data.

Hyunsoo Rim3h

Michael Burry flags “troubling” jump in Nvidia’s supply commitments

“The Big Short” investor Michael Burry — famous for betting against the 2008 housing bubble — just warned of a major risk in Nvidia’s latest annual report, pointing to a sixfold surge in purchase obligations over the past year.

In a Substack post Thursday, Burry called the increase from $16.1 billion to $95.2 billion in just 12 months “troubling,” noting that Nvidia has been forced to place noncancelable purchase orders well before knowing the final demand for its AI chips. The surge is partly tied to supplier TSMC requiring longer-term contracts, he added.

'Big Short' Michael Burry warns Nvidia is taking on risk that could be 'catastrophic'

Matt Phillips5h

Vistra beats Q4 earnings expectations for adjusted EBITDA, but dips on income decline

Power provider Vistra, a key player in the AI energy trade, reported better-than-expected adjusted earnings results early Thursday, but shares dipped in early trading as Q4 net income dropped.

The Texas-based company, which supplies nuclear- and natural gas-fueled power to wholesale and retail markets, reported:

Net income of $233 million, a decline of 52% from Q4 2024.
Adjusted EBITDA from ongoing operations of $1.74 billion vs. the $1.71 billion expected by Wall Street analysts.
Vistra maintained previously issued guidance for full-year EBITDA from ongoing operations and adjusted free cash flow from ongoing operations.

Vistra shares soared 258% in 2024 amid a flurry of excitement over the AI energy boom. Last year was more muted, with the stock rising 17%. So far in 2026, shares were up roughly 9% before the report.

Vistra beats fourth-quarter core profit estimates on AI-driven power demand

Net income of $233 million, a decline of 52% from Q4 2024.
Adjusted EBITDA from ongoing operations of $1.74 billion vs. the $1.71 billion expected by Wall Street analysts.
Vistra maintained previously issued guidance for full-year EBITDA from ongoing operations and adjusted free cash flow from ongoing operations.

Luke Kawa

Why Nvidia’s terrific quarter is getting a terrible reaction

Nvidia’s CEO gave a deeply unsatisfying answer about his biggest customers’ ability to generate cash. But they might be investing more in AI GPUs in 2027 anyway.

Jensen Huang in front of Vera Rubin at CES 2026

Claire Yubin Oh7h

Sandisk rises on partnership with SK Hynix to standardize memory chip architecture tailored for AI data centers

Sandisk is up 3% in premarket trading on Thursday after it began its global standardization strategy of high-bandwidth flash (HBF) memory solutions with SK Hynix.

SK Hynix commented in a press release on Thursday that “by making HBF an industry standard, together with Sandisk, we will lay the foundation for the entire AI ecosystem to grow together,” adding that the companies will set up a dedicated workstream to work on the standardization under the Open Compute Project, the world’s largest organization dealing with data center technologies.

First debuted last February, Sandisk’s HBF technology lies in between ultrafast high-bandwidth memory (HBM) and high-capacity SSDs. That is, these have more storage capacity than HBMs, but are still fast enough to be utilized in AI inferencing (albeit not as quick as HBM).

Sandisk has previously argued that this hybrid architecture is central to AI services that need user applications but require a significant amount of fast interconnect between GPUs. The latest announcement also notes that HBF technology is expected to be more cost-efficient compared to alternatives of similar scale.

The launch, which was shared in an kickoff event on Thursday evening, starts SK Hynix and Sandisk’s workflow, which was announced when the two companies signed a memorandum of understanding “to standardize the specification, define technology requirements and explore the creation of a technology ecosystem” last August, per Sandisk’s press release at the time. Ultimately, by collaborating with SK Hynix, one of the three key HBM suppliers, to standardize and commercialize the technology, Sandisk is manufacturing somewhat of a first-mover advantage to offer the system-level “AI-optimized memory architecture” required for AI inference markets, rather than focusing on the performance of a single chip element.