scx.ai logo

Strategy Blog

The EU Just Flipped the Switch on Sovereign AI. Australia Should Be Watching.

The EU AI Act's high-risk obligations became enforceable on August 2. At the same time, frontier model prices collapsed 80%. SCX.ai CEO David Keane explains why this convergence reshapes the entire AI infrastructure conversation.

By David Keane, Founder and CEO, SCX.ai7 min read

Two things happened in August that should reframe how Australia thinks about AI infrastructure.

The first was regulatory. On August 2, the EU AI Act's high-risk obligations became enforceable. In-jurisdiction compute. National model control. Sovereign cloud deployment. After years of drafts and consultations, the switch flipped.

The second was economic. OpenAI slashed GPT-5.6 Luna pricing by 80 percent. Google cut Gemini 3.6 Flash output costs for agentic workloads. Frontier models that cost serious money six months ago now cost pocket change.

Taken separately, each is interesting. Taken together, they tell you exactly where the AI industry is heading next.

Regulation just got real

The EU AI Act is not a white paper. It is binding law across 27 member states, and its most consequential provisions are now live.

What does it actually require? If you operate a high-risk AI system in the EU, your compute must run inside the jurisdiction. Your models must be under national control. Your cloud deployment must be sovereign. You cannot train on EU citizen data in a US data centre and call it compliant. You cannot run inference through a hyperscaler that routes your prompts through another continent.

This is not theoretical. The EU has already backed the regulation with hardware. Plans for up to seven AI Gigafactories, backed by over 30 billion euros in public and private funding. Nineteen AI Factories and 13 AI Factory Antennas already live through the EuroHPC network. Mistral AI raised 830 million euros to build a data centre near Paris. Deutsche Telekom stood up a 0.5 ExaFLOPS Industrial AI Cloud in Munich.

The EU is not debating sovereignty. It is building it.

Models are becoming a commodity

At the exact moment regulation is forcing compute into specific jurisdictions, the cost of the models themselves is collapsing.

OpenAI cut GPT-5.6 Luna to 20 cents per million input tokens. Google made Gemini 3.6 Flash dramatically cheaper for long agentic tasks. These are not minor adjustments. They are 80 percent price reductions on the frontier.

When models cost real money, the model was the moat. You competed on which model you had access to. Now the model is becoming a commodity. The differentiation shifts to something else entirely.

What is that something else? Where the model runs. How fast it responds. Whether the data stays in jurisdiction. Whether the infrastructure is designed for your workload, not the hyperscaler's margin.

This is the conversation nobody was having two years ago and everyone will be having twelve months from now.

The convergence that matters

Put these two trends together and the picture is clear.

Regulation is pulling compute into sovereign jurisdictions. Economics is pushing the competitive battleground from models to infrastructure. The companies that win the next phase will not be the ones with the best model access. They will be the ones with the best infrastructure in the right places.

The EU understands this. That is why they are not just writing rules. They are building factories.

Australia should be taking notes.

We do not have the EU's regulatory framework yet, but the direction of travel is unmistakable. South Australia launched the nation's first AI Royal Commission this month. The federal government is actively developing AI policy. The question is not whether sovereignty requirements will arrive. It is whether we will have the infrastructure ready when they do.

Australia's window is open but it will not stay that way

The global sovereign AI infrastructure market is projected to reach 50 billion dollars in 2026 and grow to nearly 250 billion by 2035. More sovereign AI initiatives were announced in Q1 2026 alone than in all of 2024. Sovereign wealth funds have committed an estimated 120 billion dollars to AI data centres and semiconductor fabrication between 2025 and 2026.

The Middle East and East Asia account for over 80 percent of publicly disclosed sovereign AI investment. Saudi Arabia's HUMAIN project is deploying over 100 billion dollars across 11 data centres. The UAE's Stargate initiative is targeting a gigawatt of AI data centre capacity.

These are not pilot programs. They are national bets on infrastructure as strategic capability.

Australia has advantages. A stable regulatory environment. World-class renewable energy. Deep expertise in critical infrastructure. A location that sits naturally between Asian and Western markets.

But advantages without infrastructure are just good intentions. If Australian enterprises and government agencies are forced into sovereign compute by regulation, regulation that is already taking shape, and the only infrastructure that meets the requirement is overseas, we will have traded one dependency for another.

What we are building at SCX.ai

This is not an abstract argument for us. It is what we are building.

Our first sovereign AI node is live at Equinix SY5 in Sydney, running on purpose-built SambaNova SN40L processors. We listed on the ASX on August 21 after a 40 million dollar IPO. A week later we announced our partnership with DDN to integrate their Infinia data platform into our inference cloud, delivering sub-millisecond latency for data-intensive AI workloads.

We have contracted annual recurring revenue of approximately 6.5 million dollars and more than 400 active users. Node 2 is on the roadmap for late 2026.

But the bigger picture is this: we are building Australia's sovereign AI infrastructure layer before the regulation makes it mandatory, not after.

That is the difference between leading and scrambling. The EU just showed us what the playbook looks like. The smartest thing Australia can do is read it now, not after someone else writes ours.

David Keane is the Founder and CEO of SCX.ai, Australia's sovereign AI infrastructure company.

Sources

Related Topics

EU AI Actsovereign AIAI regulationAI infrastructuremodel pricingAustraliaSCX.aidata sovereigntyenterprise AIcompliance
The EU Just Flipped the Switch on Sovereign AI. Australia Should Be Watching.