The New Map of AI Power

**Ariana Salvatore** (0:01)
Welcome to Thoughts on the Market. I'm Ariana Salvatore, Head of US. Public Policy Research at Morgan Stanley.

**Stephen Byrd** (0:06)
And I'm Stephen Byrd, Head of Global Thematic Research at Morgan Stanley.

**Ariana Salvatore** (0:10)
Today, we'll be talking about AI sovereignty, what it means, what countries around the world are doing to advance their own goals, and what a more fragmented AI ecosystem could mean for investors. It's Thursday, August 20th at 2 p.m. in New York.

**Stephen Byrd** (0:24)
And it's 9 p.m. in Helsinki.

**Ariana Salvatore** (0:27)
As AI becomes more powerful and therefore more important to the global economy, countries are asking a basic question. How much of it do we need to control ourselves?
That's at the heart of AI sovereignty, making sure governments around the world can access the computing power, data, energy, and technology they need, even as geopolitical tensions may rise.

**Stephen Byrd** (0:48)
And that seems to fit into a broader trend we've been talking about for some time, a more multipolar world where governments are increasingly willing to intervene in markets around strategically important technologies.

**Ariana Salvatore** (1:00)
Exactly. We describe this as a potential two-worlds dynamic. The US and China have been gradually de-risking from one another, particularly in advanced technology. We've already seen policy tools including export controls, tariffs, and incentives for domestic manufacturing. And as AI becomes more strategically important, our expectation is for policy intervention to increase rather than decrease.
But what's interesting is that the US and China aren't necessarily pursuing sovereignty in the same way.

**Stephen Byrd** (1:29)
So let's unpack that.
Can you start with the US.? What does the American approach look like?

**Ariana Salvatore** (1:35)
Yes, we think the US is trying to do two things at once, basically. On one hand, it wants to preserve national security guardrails around some of the most sensitive AI capabilities. But on the other hand, it has an incentive to make sure the American AI tech stack is broadly available to allies and partners.
So, there's an inherent tension there between those two objectives. Obviously, if you restrict access too much, you can encourage other countries to develop alternatives. But if you allow unrestricted access, policymakers may begin to worry about losing control over strategically important technology.
So the way that we chart this is through a middle path. We think the direction of travel looks less like complete technological separation and more like selective access, tighter controls around sensitive capabilities alongside an effort to maintain the global reach of the US. AI ecosystem.

**Stephen Byrd** (2:26)
Whereas China's approach is more focused on building out an indigenous ecosystem. Specifically, we see policymakers in China pursuing greater self-sufficiency across the AI stack, from chips and computing infrastructure to cloud and models.
Our China strategists argue that bifurcation could actually increase China's incentive to build a larger China-compatible AI ecosystem abroad, particularly across the global south and other markets that aren't firmly aligned with the US ecosystem. China's model emphasizes lower-cost models, open-weight ecosystems, subsidized compute, cloud partnerships, and infrastructure exports. So the competition could increasingly be about not only which country has the most advanced model, but which ecosystem can achieve the widest adoption.

**Ariana Salvatore** (3:17)
That's right. And that brings us back to this idea of two worlds. So, Stephen, is the implication here that we're going to be heading toward two completely separate AI systems?

**Stephen Byrd** (3:26)
Not necessarily, I'd say. You know, the supply chains are still deeply interconnected. So our research does not suggest a sudden decoupling, but we could see greater duplication and less globally fungible infrastructure.
Countries may increasingly want compute located domestically or regionally. Sensitive data may need to stay within particular jurisdictions. And companies may need different cloud, cybersecurity or distribution arrangements in different markets. And that means the same global level of AI demand could require more physical infrastructure than it would in a completely integrated world.

**Ariana Salvatore** (4:02)
So fragmentation, like other themes within multipolarity, are more economically inefficient, but potentially pretty important for the investment cycle. We think sovereign AI can make the system more redundant and more capital intensive as a result. Our research teams think there are potential beneficiaries from that across semiconductors, data centers, networking, power, cloud, cybersecurity, and infrastructure software.
Let's look at data centers specifically. If governments and enterprises increasingly require local hosting and greater control over sensitive data, you will inevitably need more geographically distributed infrastructure. Co-location operators we think can benefit because they provide the power, cooling, space, security, and interconnection that can allow customers to keep workloads in specific jurisdictions. So the fragmentation we're talking about may introduce inefficiency at a system level while simultaneously creating incremental infrastructure demand.