In AI, control is the real advantage

Anthropic's suspension of foreign access to its most advanced AI models is a reminder of how quickly access to technologies can be restricted when core AI infrastructure is controlled offshore. It has also resurfaced debate around what sovereign AI really means in practice. 

The timing is particularly significant for Australia, as the country ramps up its own AI build-out. In late April, Microsoft announced plans to invest $25 billion over the next three years in Australian data centres, the largest ever investment by a global technology company in the country. It's a big number that conveys confidence in Australia's digital future. 

But the more important question isn't how much is being invested, it's who controls the systems that will run AI, shape decisions and capture the value it creates.

For years, Australian businesses have relied on global cloud services because they offer speed, scale and flexibility. These are still important considerations, but in AI, speed without control can quickly become a supply chain risk.

Sovereignty now means more than data location

In the past, sovereignty was mostly about where data was stored, who could access it and how it was protected. AI makes that definition much broader.

When we refer to sovereignty today, that includes the model itself, how it is trained, the rules that shape its behaviour, the prices attached to usage, and the provider's ability to change features or remove access.

This matters because an organisation can keep data onshore and still not control the most important part of the system. The servers may be local, but the intelligence layer can be owned and governed elsewhere.

That's why sovereignty is best understood as a spectrum, with some companies controlling only their data, and others controlling data, models, workflows and commercial terms. 

The more of the stack an organisation controls, the more flexibility and resilience it has.

Control improves resilience

Control is more than a governance issue, it's also an operating issue. Today, our global digital infrastructure still depends heavily on assets outside Australia. About 99 percent of international internet traffic travels through submarine cables, for example.

Clearly, organisations can't avoid engaging with global infrastructure altogether. But it does mean leaders should be realistic about concentration risk, network dependency and the limits of local hosting on its own.

This is about advantage, not just compliance

Conversations on sovereign AI are often framed around risk, privacy or regulatory issues. What is overlooked is the strategic edge it can give organisations. With sovereign AI, businesses can draw on their domain knowledge, operating model, decision logic and the way they serve customers, and build them into AI systems that can be shaped and improved. 

In this way, AI stops being a generic tool and becomes an expert source, leading to faster decisions, more consistent execution and better economics over time. It can also reduce the risk of building critical capability on top of terms, pricing or product roadmaps set entirely by someone else.

The economics are getting clearer

As AI moves from experimentation to scaled use, the cost model becomes harder to ignore. Usage-based pricing is making the economics of external models much more visible. This isn't necessarily a negative, particularly as it can force more deliberate decision-making about where and how AI is used. Which workloads should stay on external platforms? Which ones are important enough, frequent enough or sensitive enough to justify more local control?

The point is not to replace global models everywhere but to understand where organisations can derive the most value. Those that understand when to rent and when to own will be in a much stronger position.

The market is moving toward greater control

This shift is already visible. Gartner forecasts that worldwide sovereign cloud infrastructure-as-a-service spending will reach US$80 billion (AU$116 billion) in 2026. It  also predicts that by 2027, 35 percent of countries will be locked into region-specific AI platforms using proprietary contextual data.

Together, these predictions point to a bigger change in strategy. Organisations are no longer optimising only for global scale, they are starting to balance scale with control, resilience and long-term flexibility.

Public trust also matters. People are more aware of how data is used, where it moves and who is accountable when something goes wrong. That makes control not just a technical question, but a trust question too.

Local AI environments can help teams move faster

Another benefit that's easy to miss is sovereign AI can accelerate adoption.

When organisations are able to test and deploy AI in environments that align with local legal and governance requirements, they reduce friction around sensitive use cases, shortening the path from pilot to production.

While that doesn't remove the need for strong governance, it can make innovation easier to manage, especially in the public sector, critical infrastructure and regulated industries, where trust, auditability and accountability matter most.

Australian organisations will continue to use a mix of global and local capability which is realistic and in many cases desirable. But the next phase of AI will reward more than speed. 

The organisations that get the most value from AI will be the ones that know what they must control, what they can share, and how to build advantage on top of both.