Geopolitics
2030 outlook: European firms must navigate two separate tech stacks
Read more
Export controls used to look straightforward on paper: determine whether sensitive technology can be shipped to a given country, then approve or deny. But in practice, global supply chains and cloud infrastructure have made "where it goes" a weak proxy for "who benefits."
That is the core theme of this episode. Host Theresa Terzer frames the trend succinctly: Washington is increasingly trying to control not only where strategic technology is shipped, but also who ultimately owns and controls and accesses it. Kevin Allison argues that this shift reflects the reality of modern AI infrastructure, where compute can sit in one jurisdiction, be operated by a trusted firm, and still be accessible to users elsewhere.
Kevin places today's chip controls in a nearly decade-long arc. As AI became a driver of economic growth and military capability, governments (especially the U.S. and China) started treating AI leadership as a determinant of national power.
Semiconductors are the physical foundation of that competition. From Washington's perspective, U.S. influence over key nodes of the global semiconductor industry can function as a potential "choke point" to preserve “U.S. leadership”. The latest BIS guidance is another step in a larger geopolitical contest over who sets the pace of AI progress.
Kevin describes the diffusion framework introduced late in the Biden administration as a plan to control the flow of the most advanced chips tied to frontier AI capabilities. The approach divided countries into buckets with varying levels of access. Unsurprisingly, it drew controversy, both because of the geopolitics and because global tech ecosystems do not map cleanly onto simple country tiers.
Under the next administration, enforcement was effectively paused and the rule was withdrawn in practice, reflecting a strategic pivot favored by parts of Silicon Valley: rather than strictly limiting overseas compute, the U.S. would prefer that AI infrastructure abroad be built on American compute.
The new BIS guidance (May 31, 2026) signals a different emphasis: advanced compute exports may require licenses not just based on the destination country, but based on whether the end user is owned by or tied to a restricted group of countries (including China and Russia). In other words, geography matters, but so does corporate control, beneficial ownership, and access.
Kevin's answer is nuanced. The concepts of end user and end use have always been part of export controls, and governments have long worried about indirect pathways (for example, sensitive chips shipped to a third country but ultimately used by restricted actors).
What feels new is the way cloud services and global data centers stretch the model. Even if chips sit in a "trusted" location or are operated by a trusted company, a crucial question remains: can restricted end users still access that compute remotely? That is why ownership, control, and access are becoming central to the compliance logic.
Kevin outlines multiple U.S. objectives:
But he also highlights the friction points. Export controls are blunt instruments. They can slow access to cutting-edge chips, but they also incentivize rivals to:
He notes that China has continued to innovate in hardware (e.g., Huawei’s Ascend line) and in AI labs (e.g., DeepSeek), despite multiple rounds of U.S. restrictions, suggesting the policy outcomes may be less decisive than intended.
A key warning for companies is that policy risk is now bidirectional. Kevin argues that China has expanded its own statecraft toolkit, mirroring parts of the US approach. That includes supply-chain rules and export restrictions that focus on critical inputs further down the AI stack—especially critical minerals and rare-earth magnets.
These inputs matter not only for chips and data centers, but also for automobiles, renewables, and industrial production. That means geopolitical competition around AI can surface as availability shocks or licensing delays in places procurement teams might not traditionally flag as “AI-related.”
What companies should do next
The episode’s practical message is that this is no longer a niche semiconductor issue. Any global company that depends on AI infrastructure—or on supply chains that support AI—needs a more granular view of exposure.
The compliance question is evolving from "Where are we exporting?" to "Who ultimately controls and can access the capability?" For companies, that shift raises the bar: the risk is no longer confined to a shipment or a destination country. It lives in organizational structures, cloud access models, and the increasingly intertwined policy levers of Washington and Beijing.
Read further analysis on this and other topics on Geolytics.Hub with a free demo account.
Watch the full episode of the podcast on YouTube.
Subscribe to the Geopolitics and Business Briefing in your favorite podcast app:
