Two stories broke this week that, on the surface, look unrelated. One is about export controls. The other is about memory chips. Together, they describe the same underlying problem: U.S. semiconductor policy keeps discovering gaps after the fact, and the industry keeps absorbing costs that policy didn't anticipate.
BIS Spent 18 Months Not Noticing Its Own Rule
On May 31, the Bureau of Industry and Security issued a weekend guidance notice — the regulatory equivalent of a fire alarm at 2 a.m. — confirming that export licenses are required for advanced AI chips shipped to any entity whose ultimate parent company is headquartered in China, Russia, or another Country Group D:5 nation, regardless of where that entity is physically located. Reuters reported that the new guidance does not change anything for Nvidia, since the company says it could not ship the chips anyway.
That last sentence is doing a lot of work. Because according to TechTimes, a Bloomberg report revealed that Trump administration officials spent the days after the guidance in an internal dispute over whether their own China tech policy had inadvertently left the loophole open for nearly 18 months — and whether companies used it to ship Nvidia Blackwell chips to Chinese-linked firms in Malaysia and Singapore without government approval.
The mechanics of the workaround were straightforward: Chinese technology companies established wholly-owned or majority-owned subsidiaries in Singapore, Malaysia, Taiwan, and other Asia-Pacific jurisdictions, then purchased advanced AI accelerators through those entities. The argument, implicit or explicit, was that neither the seller nor the immediate buyer was located in China. TechTimes notes that the ultimate-parent-company test has been embedded in the Export Administration Regulations since November 2023 — meaning the rule was always there. BIS just wasn't enforcing it.
I wrote in April about BIS drowning in its own enforcement backlog. This is what that looks like in practice: a rule exists on paper, enforcement is inconsistent, and the gap becomes commercially significant before anyone in Washington officially acknowledges it. The weekend guidance closes the stated loophole. It does not answer the harder question — what happens to the chips that already moved through it.
The compliance exposure here is real. Any distributor, cloud provider, or contract manufacturer in the Asia-Pacific region that transacted with subsidiaries of Chinese technology companies now faces retroactive scrutiny stretching back to November 2023. That's not a warning shot. That's a liability.
The Memory Market Is Breaking in a Different Direction
While BIS was quietly patching its AI chip controls, a coalition of nine U.S. trade associations sent a letter to Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick on June 3, warning that the AI-fueled rush for memory chips is creating a supply crisis with consequences well beyond the data center. Cryptobriefing reports that the groups represent industries spanning automotive, consumer electronics, medical devices, and telecommunications — and that DRAM prices surged over 60% in 2025.
The structural shift is not subtle. Major producers including Micron, Samsung, and SK Hynix have redirected manufacturing capacity toward high-bandwidth memory and high-performance DRAM for AI infrastructure. Cryptobriefing cites projections that data centers will consume roughly 70% of total global memory chip output by the end of 2026. The Synopsys CEO is quoted predicting the shortage persists into 2027.
The CHIPS Act funded domestic fab construction — Micron received significant support for new U.S. facilities — but as the trade groups' letter implicitly acknowledges, meaningful capacity from those investments is still years away. Policy moved; physics didn't.
This is the accountability gap that industrial policy tends to paper over. Announcing a fab investment and cutting a ribbon are measurable events. Waiting three to five years for that fab to produce chips at volume, while the market moves in a direction the policy didn't anticipate, is the part that doesn't fit on a press release.
The Common Thread
Export controls and memory supply look like separate files. They're not. Both reflect the same pattern: policy designed around one threat model (China acquiring leading-edge logic chips) while a second problem (AI infrastructure consuming the entire memory supply chain) develops in the background without a coherent response.
The trade associations are now asking the government to intervene — accelerate capacity, adjust trade policy, do something. Watch whether the Commerce Department's response, if any, comes before or after the next earnings season reveals how badly input costs have hit automotive and consumer electronics margins. That timing will tell you whether this is being treated as a supply chain emergency or a letter to file.
The BIS loophole guidance, meanwhile, sets up a different accountability moment: whether any enforcement actions actually follow, or whether the weekend notice was itself a form of paperwork — the regulatory equivalent of marking something resolved by writing it down.
