AI creates opportunity for global economic change, yet requires local AI implementation
Aiolian, Your Partner for Global AI Deployed Locally
Aiolian International is a boutique consulting firm that helps developing nations, their local companies and investors create sovereign AI infrastructure while comprehending sovereign AI innovations, regulations and export controls.
The partners at Aiolian are an international team of data center veterans who have built data center & AI technology companies, created secure, multi-megawatt data centers and successfully engaged with the US government on policy and trade
Major companies served:
Will Trump’s AI Action Plan Create Chaos for Sovereign AI Initiatives?
The plan’s policies & official comments bear all the hallmarks of the Biden AI Diffusion Rule
July 28th, 2025 – On July 23rd, the Trump administration released their AI Action Plan that sets out 90 specific federal policy actions under three key pillars — increasing domestic innovation, building domestic AI infrastructure, and asserting international leadership. Our clients and partners are acutely aware of the now-rescinded Biden Framework for AI Diffusion interim-final-ruling (IFR) that was designed to mitigate risks of adversary’s use of U.S. AI technology. The international leadership pillar of the AI Action Plan is fundamentally different with policies designed to dominate the global demand for AI via the export of a “full AI technology stack” to U.S. “allies”. The policy’s goal is for sovereign nations and their AI cloud companies to procure this full AI stack with the help of U.S. financial tools.
While this might sound like good news to sovereign nations after the thrash of the Biden IFR, the action plan also includes policies to control the diversion to and use of U.S. AI technology to and by China/etc. In other words – like a Phoenix rising from the ashes – the spirit of the Biden IFR will re-emerge in what we believe will be an even messier, less transparent, incarnation as we’ll describe below.
From Defence to Offence – Until now, the Trump administration had prioritized America First without anthing amounting to a Marshall plan. But it has always cared about blunting BRIC, China’s Belt-and-Road or China’s Digital Silk Road Initiatives – and in fact, the Biden IFR had required importers to rid themselves of Chinese technology (Huawei), investors and employees. We believe that David Sacks – Trump’s AI and Crypto czar – has played a key role in shifting the philosophy from one of defense (Biden IFR) to one of offence (domination via a full AI stack) as he has consistently advocated making U.S. technologies the de facto global standard, warning that failure would concede ground to Chinese competitors (e.g. Huawei). The resulting AI Action Plan has an amibitious scope that some might suggest amounts to a new “Marshall Plan for American AI.”
What is the “Full AI Technology Stack” and Policy? – Within 90 days – as further described in a corresponding executive order - the policy requires the creation of a program to gather proposals from industry consortia for full-stack AI export packages, targeted regions and requested financial support. The table below shows the likely U.S. consortia, the Chinese vendors - that the policy is trying to preclude - and other non-U.S., Non-Chinese vendors that will be left-out. One likely example of a full stack is a version of HPE’s AI Factory (integration of Nvidia GPUs, networking and AI system software with HPE GPU servers, storage and AI lifecycle management) augmented with U.S. LLMs and other AI software. Selected AI stacks will “designated as priority AI export packages and will be supported through priority access to the U.S. Federal Financing tools” and multiple U.S. agencies will coordinate to “facilitate deals”.
The Trump administration could have taken an extreme position of limiting the permissible, exported AI stack to international Hyperscalers (Microsoft Azure, Google Cloud, AWS, Oracle Cloud) but that wouldn’t address nations’ goal for sovereign clouds (e.g. France’s Bleu, Saudi’s Humane, UAE’s G42, Qatar’s Ooredoo, Malaysia’s YTL), who will be among the the primary targets of the program’s full AI stack. Access to funding programs might offer appeal to countries pursuing a sovereign strategy, but challenges remain. Below we summarize seven key challenges for countries and companies under the new AI Action Plan and the full stack strategy.
1 - Full AI Stack Dependency - By definition, a company that deploys a full AI stack will incur full dependency on the U.S. vendor(s) for all layers of that stack resulting in reduced autonomy / sovereignty. As an analogy, sovereign nations remember when U.S. sanctions forced Google to stop supporting the Android operating system software for Huawei smartphones.
2 – Inclusion of U.S. LLMs – every country has their own values and culture which are being integrated into each country’s development of sovereign LLMs. The inclusion of U.S. LLMs in the full stack is an obvious non-starter, especially if exclusive.
3 – Likely Attempts at Influencing Local AI Regulations – While the plan’s policy for international AI regulation seems relatively general (“… vigorously advocate for international AI governance approaches that promote innovation, reflect American values, and counter authoritarian influence”), Trump’s speech to U.S. industry at the plan’s launch reflect the intention… “we also have to watch Europe, Asia, and all foreign countries so that they don't make rules and regulations that likewise make it impossible for you to do business…”. And as part of promoting the U.S. full stack, the executive order spells out that the U.S. State Department be responsible for “supporting partner countries in fostering pro-innovation regulatory, data, and infrastructure environments conducive to the deployment of American AI systems.”
4 – Hurdle of Addressing Local Cyber-Security Requirements – Sovereign AI clouds will be subject to existing local security and cyber-security requirements. Every nation has different regulations (e.g. the extensive, strict cloud security certification of France’s SecNumCloud). Related to (3) above, sovereign nations will be adverse to unifying their cyber-regulations, and U.S. full-stack vendors will stumble with the heavy liftning required to customize their full-stacks for every nation.
5 - Inclusion of Location Verification Technology – To preclude transshipments and smuggling of U.S. AI hardware to China/etc, the action plan includes a policy to explore “location verification features”. While the Biden IFR relied on on-site biannual inventory compliance checks, the AI Action Plan policy leverages the concept of location verification technology - described an existing draft bill - the “Chip Security Act” proposed by Senator Cotton, et al - where a mechanism (presumably GPS-based) is included before the chip or system is exported, and for the exporter to notify the Secretary of Commerce if the exporter learns the chip or system “is in a location other than the location specified” or has been tampered with. One U.S. startup is already offering such a system-level solution. Such technologies are ostensibly passive “ankle bracelets”. But when integrated into sovereign AI infrastructure, many nations will be suspicious of these tamper-proof solutions as providing a backdoor especially with their online connections to the U.S. vendor, which furthermore breaks the air-gap required by sovereign infrastructure. While this policy is still in the exploration phase, its eventual implementation – whether integrated into a full AI stack or discretely – will be perceived as an even worse solution versus the Biden IFR’s manual verification approach.
6 - Inclusion of End-User Monitoring Technology – At a speech at the 2025 AI Honors Gala, Commerce Secretary Howard Lutnick said “…Deep Seek [LLM] and their peers were not trained for $5 million, okay… they found other players in other countries to redirect our assets or let them use those other…”. To preclude China/etc’s use of U.S. AI hardware installed in other nations’ AI clouds, the Biden IFR relied on know-your-customer (KYC) requirements and biannual compliance checks. Similar to location verification, the AI action plan’s policy also puts its hope on automation/technology for end-user monitoring. It also leverages the draft Chip Security Act which directs U.S. agencies to identify additional mechanisms to “prevent, hinder, and detect the unauthorized use” of U.S. AI hardware. The bill’s use of the word “hinder” goes beyond passive monitoring and implies a “remote kill switch,” which would be an obvious non-starter with sovereign infrastructure. As with (5) above, while this policy is still at the exploration phase, its implementation would be an even stronger dis-incentive for U.S AI technology.
7 – Cluster sizes will be limited or require operation by U.S. firms – Even with end-user monitoring (whether manual or automated), both the Biden and Trump administrations were and are aware of the national security risks of large GPU clusters outside the U.S. The Biden IFR had tried to address this risk via a three-tier system of nations with different allowable GPU cluster sizes in each tier (the concept of total processing performance / TPP quotas). How to address this risk is missing from the AI Action Plan. Despite recent pressure on Secretary Lutnick from China hawks in Congress (June 12th closed door House Republican meeting), the Trump administration has yet to release their promised replacement of the Biden IFR that addresses national security. However, Lutnick’s comments at the AI Action Plan’s launch reveal a philosophy similar to the Biden IFR… “we are comfortable with ... a large cluster provided that cluster is operated by a trusted American operator... As you go down from there… if they want a smaller cluster, would you expand the number of people who are trusted, right? And the answer would probably be yes, right? And then when you go down from there, to a smaller and smaller cluster, right? How do you deal with that?” As a sign of the ongoing back-room deliberations, on July 16th, the Wall Street Journal reported that “some Trump administration officials are holding up efforts to finalize a landmark agreement that would open the door to the UAE buying billions of dollars in Nvidia’s cutting-edge AI chips due to national security concerns”. These chips were promised by Trump during his visit to the Gulf Region.
The Vision of an American AI Alliance - The AI Action Plan refers to exporting the full AI technology stack to “countries willing to join America’s AI alliance”. This definition of this alliance is not defined in the plan. But the vision is captured in what Secretary Lutnick reportedly calls an “AI Economic Zone” and what Vice President Vance – in his launch speech - referred to as a “NATO alternative outside of Europe … built on the same technology stack”. Vis-à-vis the challenges above, nations in this envisioned alliance would embrace dependency on full U.S. AI stacks, align their AI and cybersecurity regulations, agree to implement any U.S. AI national security measures and align their AI export controls. Reality won’t be that simple…
A Dynamic and Unpredictable Next Six to Eighteen months - The policies of the AI Action Plan and the pending IFR replacement (AI export controls that address national security) try to accomplish three objectives that are at best in tension with each other, and at worst, contradictory: (1) Dominate the global demand for AI (pre-empt China’s footprint), (2) Address U.S. National Security concerns, (3) Maximize export of U.S. AI technology (beyond GPUs).
Given these tensions, we expect a dynamic, unstable environment for sovereign AI nations and their companies as they try to anticipate and navigate the impact of U.S. policies and regulations and the impact on their own sovereign AI strategies, policies, regulations and investments. Below, we describe two possible scenarios that we believe might unfold over the next six to eighteen months.
Scenario-1: Countries compelled to reach separate, bilateral agreements under unique, tough terms -– This will be familiar to readers seeing how tariff negotiations have unfolded. In fact, the AI Action plan includes a policy to “Expand new initiatives for promoting plurilateral controls for the AI tech stack…”. We believe the knobs available in these negotiations will be: (i) the degree to which the full AI stack is a “hard bundle” versus a “soft bundle” – i.e. whether GPUs are still available separately and/or whether sensitive layers (e.g. LLMs) can be excluded from the stack , (2) financing, (3) alignment of AI regulations, (4) automated versus manual location verification and end-user-monitoring, (5) the size of non-U.S.-operated clusters, etc. As a stick during these negotations, U.S. agencies will threaten action against use of Chinese AI tech per a May 13th guidance issued by the BIS (U.S. Bureau of Industry and Security) where use of Huawei AI chips are presumptively subject to the U.S. FDPR (Foreign Direct Products Rule). The actions include ominous “substantial criminal and administrative penalties”. As a further threat, the AI Action Plan makes reference to “secondary tariffs to achieve greater international alignment”.
Scenario-2: Countries give up on the US AI stack and turn to China/etc instead. In this scenario the world will bifurcate into (i) nations that are willing to embrace U.S. dependency/etc ala the vision of the American AI Alliance as described above, and (ii) nations that will eventually adopt China AI tech on the assumption it will be more friendly to achieving AI sovereinty. Case in point: Malaysia has been a flash-point given the suspicion of GPU transshipments to China and that Chinese LLMs have been trained in Malaysian AI data centers. While the Malaysian government is collaborating with U.S. agencies to modify its export controls and has warned its AI companies not to train Chinese LLMs, Bloomberg reported on July 5th that the U.S. is planning on curbing shipments of GPUs to Malaysia and Thailand (similar concerns) with tight controls over their imports of GPUs. While the extent of the curbs have yet to be disclosed, it’s inevitable that they will not be friendly to Malaysia’s goals for sovereign AI, and might create a long-term motivation for Malaysia to adopt Chinese AI tech (note that U.S. hyperscalers operating in Malaysia will likely be unimpacted). We could see this dynamic playing out in other nations where if the U.S. maintains extreme regulations for their national security, it does so knowing very well that it creates an opening for China.
We reiterate that the next six to eighteen months will provide a dynamic environment for sovereign nations and their companies who are creating sovereign AI infrastructure. Our mission at Aiolian International is to help developing these nations and their local companies create sovereign AI infrastructure while navigating the evolving U.S. regulations for AI technology exports. Please don’t hesitate to contact us if you have questions.
Aiolian International LLC helps developing nations, their local companies and investors create sovereign AI infrastructure while comprehending sovereign AI innovations, regulations and export controls. . In the fast moving AI ecosystem, the AI stack is evolving with new sovereign technologies, and is intertwined with dynamic U.S. AI export regulations. Aiolian addresses this complexity with a seamless combination of experience and capabilities: AI full stack technology planning, U.S. regulatory consulting and compliance solutions.
Questions from our Clients
What Happened to the Biden "AI Diffusion" Interim Final Ruling?
May 13th, 2025 – As we reported on May 9th, the Trump administration initiated the rescission of the Biden AI export regulation (Framework for Diffusion of AI). As of today, the Under Secretary of Commerce for Industry and Security, Jeffery Kessler has instructed BIS (Bureau of Industry and Security) enforcement officials not to enforce the Biden Administration’s AI Diffusion Rule. In addition, BIS has issued three new guidelines to further restrict China’s access to U.S. AI technology and to prevent the diffusion of China’s AI technology (Huawei’s GPUs). While not regulatory, these guidelines and their stated “penalties” are intended to act as a stopgap while the BIS continues to work on a replacement regulation that will be issued “in the future”. The first guideline states that all companies in the supply chain and end-customers must conduct rigorous due diligence to identify end users and uses of GPUs, and to detect potential diversion to China, etc. Firms must also follow new “Know Your Customer”, “red flag” guidance and may not “self-blind”, particularly for cloud and infrastructure providers, to prevent unauthorized LLM training, exports or transfers. Penalties for non-compliance include criminal actions for U.S. entities, triggering of (presumably onerous) license requirements for foreign entities, and a blanket threat to be added to the U.S. Entity List (which precludes any U.S. firm from doing business with). In addition, the second guideline specifically puts exporters on the hot seat by requiring them to fully vet their customers (that they’re legitimate, not transshipping to China and not allowing China use of IaaS for LLM training). The exporter will also obtain certifications, attestations, etc to this effect. The above guidelines will add non-trivial operational burden to exporters, importers, end-customers and every company in the supply chain. Nevertheless, we advise our clients and all companies to fully comprehend and operationalize compliance controls to preclude the BIS of one day claiming, “you should’ve known, you're not trustworthy and now we need to talk consequences…”. The final guideline is designed to blunt China’s Digital Silk Road initiative by limiting purchases of Huawei’s GPUs. As we reported in a prior Aiolian update, while Huawei’s Ascend GPUs and their Cloudmatrix 384 system are sub-optimal on data center footprint, power and system cost (by 3-4X), they do demonstrate that Huawei is fully capable of delivering AI capabilities at scale. The BIS makes a sweeping claim that “…such chips likely are either designed with certain U.S. software or technology or produced with semiconductor manufacturing equipment that is the direct product of certain U.S.-origin software or technology, or both” and therefore in violation of U.S. export regulations. Consequently, the BIS threatens that if any U.S. or foreign firms procure and/or use Huawei GPUs, they will be subject to “criminal and administrative penalties, up to and including imprisonment, fines, loss of export privileges, or other restrictions”. Assuming compliance, this policy damages China’s Digital Silk Road initiative since any China AI cloud company expansion (e.g. Alibaba Cloud) into developing nations will now be dependent on U.S. AI chips, and therefore subject to possible “use-case” restrictions in the future. Our mission at Aiolian International is to help developing nations and their local companies create sovereign AI infrastructure while navigating the evolving US regulations for AI technology exports. Please don’t hesitate to contact us if you have questions.
Why are the Hyperscalers Not Only Ignoring DeepSeek but Growing their AI Spending?
On January 27th, 2025, as awareness of China’s DeepSeek grew with investors, the stock prices of AI infrastructure companies worldwide plunged as DeepSeek’s LLM (large language model) challenged the prevailing thesis that creating breakthrough AI solutions requires exorbitant investments in AI infrastructure (GPUs, XPUs, Memory chips, data centers, power, real-estate, etc). While the LLM efficiencies claimed by DeepSeek are groundbreaking, this report will explain why it will have little impact on the growing demand for AI infrastructure. In fact, shortly after the DeepSeek news, neither Microsoft, Meta, Apple, Alphabet nor Amazon indicated any softness in their AI capital spending plans. Please contact us for the detailed report.
What is AI Sovereignty?
Artificial Intelligence (AI) Sovereignty is a hot topic, and many vendors, operators and nations alike are waving the flag enthusiastically. But what exactly does it mean? And what are implications for national policies and vendor and operator strategies? A set of countries we might describe as “emerging datacenter economies” – in south east Asia, India, south America, the middle east, etc – are hard at work defining answers to these critical questions. In this article, we outline recommendations for these emerging datacenter economies to strive for a mix of multi-national hyperscalers and sovereign-owned AI clouds that will help them optimize the balance between technology leadership and autonomy, and to offset the significant national risks of becoming largely dependent on multi-national corporations (MNC).
How are Multinational AI and Cloud Companies Evolving to Address AI Sovereignty?
While developing nations recognize the need to establish AI Sovereignty, creating an entirely home-grown capability and infrastructure is unrealistic. Nations are pursuing various models, from state-owned AI clouds to public-private partnerships. These local enterprises are adept at building AI data centers (land, power, data center buildings) and equipping them with racks of AI computing hardware (notwithstanding the US export regulations described separately). But the other critical layers of the “tech stack” are substantial (infrastructure management, data management, workload/tenant management, AI development and deployment tools, training, support, etc). Naturally, Global AI and cloud companies have invested and continue to innovate in these capabilities, but their current offerings are not “sovereign AI friendly”. This report describes how GPU vendors (e.g. Nvidia) and global AI / cloud companies are evolving to address this trend with the goal of partnering with local AI enterprises. Please contact us for the detailed report.