.jpg?w=1024)
A new problem is emerging with the rapid deployment of AI within enterprise workflows, and Australian enterprises aren’t immune to the risks.
Gartner expects Australian organizations to spend more than A$33.6 billion on public cloud services in 2026. That’s a 17.9% jump from last year. That demand curve is now colliding with a supply problem even the biggest AI vendors can’t fully solve.
According to a Financial Times report citing people familiar with Google’s operations, Google has limited Meta’s access to its AI models because of computing constraints. The limits have delayed some of Meta’s AI projects and pushed the company to ration tokens or look at alternative models.
The detail that should get the attention of Australian IT leaders isn’t the Google-Meta dynamic itself. It’s what that dynamic confirms: paying for AI capacity doesn’t guarantee access to it when a provider runs short.
AI compute is now a supply chain risk & Australia carries extra exposure
Enterprises have spent years building resilience against cloud outages and internet disruptions. AI capacity constraints introduce a different kind of dependency: the service can remain online even as the capacity behind it quietly tightens.
Australian organizations face an additional layer of that risk: much of the AI infrastructure they use is hosted overseas. That leaves local enterprises exposed to global compute demand and capacity allocation decisions made well beyond their borders.
Despite hyperscalers now pouring billions into local data centers — AWS’s A$20 billion Sydney and Melbourne build-out and Microsoft’s expanded Australian capacity among them — a problem remains.
First, the chips and hardware that feed local data centers are themselves in globally constrained supply, so expansion timelines for Australian capacity compete with the same worldwide shortage everyone else is drawing against.
Second, access to the newest frontier models is often provisioned via the provider’s global infrastructure before it’s fully replicated in the region, meaning local businesses can still experience a capacity squeeze upstream, even when the data center they use day-to-day is in Sydney or Melbourne.
Business continuity and redundancy now go together
Many businesses rely on AI for customer support, fraud detection, software development, compliance checks, or internal operations. For those organizations, reduced model availability or provider-imposed capacity limits can slow critical functions, resulting in a business continuity gap.
As a result, resilience is becoming just as important as model quality. Many organizations already use multi-cloud strategies to avoid depending on a single vendor, and the same thinking is starting to apply to AI. Designing applications that can switch between models or providers should become the norm rather than merely an ad hoc implementation.
Apple’s approach to manufacturing offers a useful comparison: rather than relying heavily on a single supplier, the company is gradually expanding its manufacturing suppliers to reduce concentration risk. For Australian enterprises deploying AI, where practical alternatives exist, using dual providers is a sensible way to reduce operational risk.
That shift is also changing procurement priorities. Model performance and per-token pricing are no longer the only considerations. Organizations are increasingly evaluating guaranteed capacity, service-level commitments, fallback arrangements, and the speed at which workloads can be moved to another provider if capacity becomes constrained.
Owning infrastructure isn’t for everyone, but a real option for some
Some larger enterprises have products or operations that lean heavily on AI and simply can’t afford to fail. Financial and health institutions sit in this category.
For these businesses, running selected workloads on owned infrastructure with open-weight models could be a better alternative than simply switching. That is not about replacing frontier models outright. It’s about ensuring a critical service isn’t entirely dependent on a single external provider’s spare capacity.
For most businesses, especially the 98% of Australian small and medium enterprises, building and maintaining owned AI infrastructure won’t be practical or worth the overhead. The more realistic task is understanding the trade-offs of full reliance on external providers: capacity limits, pricing shifts, model deprecations, and service disruptions that sit outside the organization’s control.
What Australian IT leaders should do next
None of this is really about Google or Meta specifically. It’s a sign that AI infrastructure is becoming a finite resource with direct business consequences.
Australian CIOs and technology leaders have a narrowing window to treat AI capacity as another operational dependency — one that gets the same planning, redundancy, and risk management already applied to cloud infrastructure and other critical technology services.
Two starting points don’t require a large budget: mapping which business functions would stall first under a capacity squeeze, and building at least one tested fallback path — a second model provider or an open-weight option — for whichever function tops that list.