What Hunterbrook actually found

Start with what was downloaded and compared, because that part of the report is concrete. AMD writes the startup firmware that prepares its server processors to run, and server makers build that code into their own BIOS updates. Hunterbrook downloaded IEIT's posted updates and found AMD's code inside them, in versions issued after March 2025, when the Commerce Department placed IEIT on the Entity List.

The count is at least six newer versions across updates for three server models. Where AMD's release dates were public, five of IEIT's builds followed them by 25 to 61 days. The most recent case is an update IEIT posted on October 19, 2026, in a package dated August 25, 2026, containing AMD's TurinPI 1.0.0.B code. One earlier update, posted in October 2025, included a fix for a flaw that AMD did not publicly disclose until February 24, 2026. Hunterbrook presents that timing as part of its case. It does not, on its own, show where IEIT's copy came from.

The report then widens to three server sellers that Hunterbrook notes are not themselves blacklisted: Maginfra in China, Kaytus in Singapore and Aivres in California. Hunterbrook compared their posted firmware with IEIT's and found it 99.94% identical, apparently for the same AMD server design, which Aivres and Kaytus sell as the KR2280-E3. The corporate links are individual, and the report states them separately: IEIT's 2022 annual report listed Aivres, then called Inspur Systems, as a subsidiary; Kaytus's audited accounts name IEIT as its ultimate parent company. AMD's own website lists Aivres among the makers of servers built around its Instinct AI chips. The dates and the 99.94% match show shared code. They do not identify a supplier, a transfer route, or a license status.

The three things the report does not show

Hunterbrook is unusually plain about its limits, and those limits are the story's center. It does not know how IEIT obtained AMD's code. It does not know whether AMD, a distributor, or a firmware partner held a license covering any supply. And its findings, in its own framing, do not establish an export violation.

That restraint matters because the legal question turns on exactly those missing pieces. A license is required to send covered items to an Entity List company, and Commerce's stated policy is to deny such licenses. But a former undersecretary of commerce for export administration, William Reinsch, told Hunterbrook that a license cannot be ruled out, that licenses are not automatically denied in practice, and that the Commerce Department does not publish who holds one. AMD was asked directly whether it or any partner holds licenses for the companies named. According to Hunterbrook, AMD asked for more time, was given until the following day, and did not provide a substantive answer before publication. After the report was published, AMD gave a statement to Bloomberg, reported by The Motley Fool, saying the claims misrepresent its business practices and relationships and that it complies with U.S. export laws and global compliance procedures. That statement did not answer whether a license covered this code. Both facts belong in the record: no substantive answer before publication, a general denial after it.

There is a precedent for treating software updates as exports, and Hunterbrook cites it. In October, the Commerce Department settled with Lambda Research, a Massachusetts software company that kept sending automatic updates to China's SiCarrier after SiCarrier was listed in December 2024. Commerce counted each of seven updates as a separate violation, and Lambda admitted the conduct. That settlement shows the category is enforced. It also differs from the AMD case in the ways that matter: there, Commerce identified the sender and the transfers, and the company admitted what it had done. None of that is present in the AMD reporting so far.

The wider exposure is arithmetic, not allegation

Separate the firmware finding from the report's second argument, which rests on AMD's own filings. Hunterbrook cites $7.75 billion billed to customers in China and Hong Kong in fiscal 2025, 22.4% of AMD's revenue, against 9.1% for Nvidia in the same markets, and $4.28 billion billed to Singapore. Those are billing-geography figures from filings; they say where customers were billed, not where chips ended up or how they were used. The report also points to two rule changes, which are separate from each other and from the firmware question. A suspension of the rule extending restrictions to foreign companies at least 50% owned by listed entities expires on January 10, 2027; whether it then reaches Kaytus or other IEIT-linked firms is a question that rule change would decide, not one the firmware finding answers. A separate interim final rule, effective on the report's publication day, requires licenses for advanced chips shipped to Singapore and four other countries; it governs chip shipments, not firmware already inside server updates.

Those are dated, checkable facts about exposure. They describe what could become harder for AMD if the rules tighten and if regulators take an interest. They are not evidence that anything improper has already happened, and the report's own language keeps them apart, even when its headline does not.

The market reaction, kept in its place

AMD shares fell about 9% during November 19 trading, per The Motley Fool, and closed about 8% lower, per investinglive's market coverage. This article records the move because it happened on the report's publication day; it does not establish how much of the move the report caused, and a price move is not evidence either way. What would move the story is narrower: a license record, a Commerce statement, or a company disclosure that answers how post-listing code reached IEIT's updates.

Sources

  1. [1] Hunterbrook Media, "AMD Has Its Own China Problem," November 19, 2026 (the primary report: short-position disclosure, IEIT firmware findings, 99.94% match, Lambda Research settlement, Reinsch comments, fiscal 2025 billing figures, January 10, 2027 rule date, and the statement that Hunterbrook does not know how IEIT obtained the code or whether a license existed)Read source
  2. [2] The Motley Fool, "First Nvidia, Now AMD: A Short Seller Just Put Its China Business Under the Microscope," November 19, 2026 (market coverage reporting AMD's post-publication statement to Bloomberg and the intraday share fall)Read source
  3. [3] investinglive, "ICYMI: AMD slides 8% as short seller Hunterbrook ties its firmware to a blacklisted Chinese firm," November 20, 2026 (market coverage for the closing fall of about 8% on November 19)Read source