For most of the global AI race, the competition has been relatively easy to understand. Who has the best models? Who has the most advanced chips? Who can build enough data centres to keep feeding both of them?
Now something more complicated is happening. The United States and China aren't only competing to build better AI. They're trying to shape the wider technology ecosystems that countries build around it.
On 14 August, Reuters reported that the US State Department had prepared a draft letter warning partner countries that they could be excluded from the US-led Pax Silica initiative if they also participated in competing AI frameworks. The letter hadn't been sent when Reuters reported on it, and could still change. But the direction is difficult to ignore.
If countries really do start facing harder choices about which AI ecosystem they belong to, the consequences won't stop at diplomacy. They could eventually reach right into the technology multinational organisations are able to use.
The AI Race Is Moving Beyond Models
Pax Silica isn't simply an alliance around American AI models. Launched by the US in December 2025, the initiative is designed around AI supply chains, bringing together areas including critical minerals, semiconductor manufacturing, compute infrastructure and AI itself.
China now has a competing framework. President Xi Jinping launched the World Artificial Intelligence Cooperation Organization in July, alongside a wider push to promote Chinese AI technology and international cooperation. Kazakhstan has joined both initiatives, making it the clearest test so far of whether countries will be allowed to participate across competing ecosystems.
That changes the nature of the US-China AI race. Building the strongest model is still important, but so is influencing which technologies, supply chains and standards other countries build around. And once you start looking at what's included in those ecosystems, AI quickly stops looking like a standalone technology problem.
An AI Ecosystem Now Reaches Far Beyond AI
An AI model doesn't exist by itself. It needs compute. Compute needs chips. Those chips depend on specialised manufacturing, materials and minerals. The data centres running everything need networks, equipment and a frankly enormous amount of energy.
Pax Silica reflects that dependency chain. The US State Department describes it as an effort spanning energy, critical minerals, semiconductors, advanced manufacturing, connectivity, compute and foundation models.
China is building across many of the same layers, from domestic semiconductor development to increasingly capable open-weight models, where the model's trained parameters are made available for others to run or adapt. Beijing has also considered restricting overseas access to some advanced Chinese models and semiconductor technologies, showing how export controls are beginning to reach beyond physical hardware.
So AI geopolitics is increasingly becoming a question of who controls the dependencies needed to build and run AI, not just who wins the latest benchmark. We're already starting to see what happens when those dependencies meet different national rules.
Regional AI Stacks Are Already Starting To Look Different
Apple gives us probably the clearest example.The company has reportedly trained a large language model specifically for China with support from Alibaba. That's a departure from the approach Apple uses elsewhere, where technologies from companies including OpenAI and Anthropic can form part of its wider AI offering.
Those US models aren't available in mainland China, so Apple has had to build differently for that market. Same company. Broadly the same AI ambition. Different AI architecture because the jurisdiction changed. There are commercial reasons this fragmentation can keep growing too.
Chinese developers are producing increasingly capable models at considerably lower costs, while demand for open-weight alternatives is pushing American companies including Meta and Nvidia to invest more heavily in their own offerings. Major US frontier providers still hold important advantages for demanding workloads, but enterprises don't necessarily need one model to win every task.
Europe makes the picture messier again. The European Commission signed the Pax Silica declaration in June while continuing to pursue greater European technology sovereignty across cloud, semiconductors and AI. This probably isn't heading towards two perfectly separated stacks labelled "US" and "China".
A more realistic outcome is partially overlapping regional AI ecosystems, with different combinations of providers, infrastructure, regulation and dependencies. For multinational enterprises, that gets complicated quickly.
One Global AI Stack May Become Harder To Maintain
A global enterprise AI strategy usually works more easily when the underlying technology remains reasonably consistent. The same models are available. Data can move where it needs to. Cloud infrastructure behaves similarly across markets. Hardware remains accessible. Regional restrictions can start pulling those assumptions apart.
An organisation could face one set of approved model providers in one country and another somewhere else. Data sovereignty rules may dictate where information can be processed. Export controls can affect hardware or model access, while local regulation changes the governance controls surrounding the same workload.
IDC expects this problem to become significant. It predicts that by 2028, 60 per cent of multinational firms will split AI stacks across sovereign zones, tripling integration costs as regulatory fragmentation and supply-chain risks make global scaling harder. That doesn't mean enterprises need to start choosing geopolitical sides.
- It does mean they may need to understand their dependencies much better.
- Which workloads can move between models or providers?
- Where are local requirements already creating differences?
- Which parts of the AI technology stack would be difficult to replace if access changed?
Those are architecture questions today. Increasingly, they may also be resilience questions.
Final Thoughts: AI Strategy May Need To Become Regional
The global AI race isn't only expanding. It's moving further down the technology stack, into the chips, minerals, infrastructure, models and rules organisations depend on to make AI work. We don't yet have two clean technology blocs, and the commercial ties between these ecosystems remain far too tangled to suggest otherwise.
But geography is already influencing what technology can be deployed and how companies deploy it. For multinational organisations, resilience may depend less on predicting which ecosystem eventually wins and more on avoiding an enterprise AI strategy built around the assumption that today's access will always remain available everywhere.
The draft US letter brings that possibility into sharper focus. If governments begin asking countries to make harder choices between technology frameworks, some decisions enterprises currently treat as procurement or architecture choices could become geopolitical ones too.
The next stage of the AI race may not be decided only by who builds the best model. It may depend on which technologies can still work together when the rules around them no longer do. As that relationship between AI, infrastructure and geopolitics keeps changing, EM360Tech will continue connecting those shifts to the technology decisions enterprises have to make.
Comments ( 0 )