1300 166 138 Tell us what you need
News / Microsoft

Microsoft’s $25B Australia AI plan skips local startups

By Monika Tantau • Published 6 months ago
Microsoft commits $25 billion to AI and cloud infrastructure in Australia, but local startups face barriers to compute access

Key takeaways

• Microsoft has committed $25 billion to AI and cloud infrastructure in Australia, the largest single buildout of its kind.
• Without explicit policy measures, advocates argue local startups will struggle to access affordable compute or pilot customers.
• The Big LEAP's Jasmine Batra proposes guaranteed startup compute, structured enterprise pathways, and an innovation dividend.
• Australia's current data centre policy frames the buildout as a property and energy question, not an ecosystem one.

Most Read

Microsoft has committed $25 billion to AI and cloud infrastructure in Australia. The buildout is real: data centre announcements, energy deals, regional capacity. But a sharper question is taking shape underneath it: will any of that compute, capital or commercial pathway end up in the hands of local founders?

That is the argument Jasmine Batra, who leads The Big LEAP and advises the Australia India Chamber of Commerce, made this week in Startup Daily. She writes that Australia is treating its data centre wave like a property and infrastructure project rather than an ecosystem one. Approvals, land, energy, capital inflows: all in. Compute credits for local startups, structured procurement pathways, an innovation dividend tied to the deals: not yet.

What Microsoft has actually committed

The $25 billion figure covers AI and cloud infrastructure across Microsoft’s Australian footprint: sustained capex into data centres, training compute and regional capacity. It sits alongside the company’s broader push to anchor Azure AI services in-region, a play other hyperscalers have mirrored at smaller scale with new capacity in Sydney, Melbourne and South East Queensland.

The political framing is straightforward. Federal and state governments have welcomed the buildout as a sovereign capability story: AI capacity inside Australian borders, jobs during construction, energy contracts that double as grid commitments. The economic framing is also clear. The spend is being handled, mostly, as a category of foreign direct investment with construction milestones, planning approvals and energy procurement attached.

What is less developed is the third question. What does the buildout mean for the people who would actually build companies on top of it?

Australian startups risk being locked out of $25 billion in new AI compute as policy treats data centres as property, not ecosystem

The startup access argument

Batra proposes three measures. First, a portion of new compute capacity should be reserved or credited to local startups, either through structured access programs or guaranteed capacity at predictable prices. Second, the buildout should come with explicit pathways into customers: pilot programs with enterprises and government, structured procurement, and routes into adjacent growth markets like India. Third, infrastructure deals should carry an innovation dividend: founder programs, research partnerships and talent investment, embedded in the terms rather than left to corporate goodwill.

The argument is not anti-buildout. It is that compute access is being treated as a market problem when it functions more like a utility one. Without intentional design, the cheapest compute in Australia will flow to the largest enterprise buyers and global operators with existing volume agreements. Local startups, by contrast, end up paying retail rates for a resource the country is now hosting at scale. The piece points to India as a counter-example, where compute and customer pathways for local founders have been wired into infrastructure policy more deliberately.

The view from Australian businesses

For Australian businesses building on or near AI, this is not an abstract policy debate. The structure of compute access shapes which AI vendors are commercially viable to work with, which startups can be procured from at SMB-relevant prices, and which categories of AI work end up being feasible on-shore.

The energy and cost side of the buildout has already drawn attention. Our analysis of who pays for Australia’s AI data centre boom set out how household power bills are being asked to subsidise compute that flows largely to hyperscale tenants. The compute access side now adds a second layer to that question. Even if the electricity gets sorted, the customers and pricing terms remain skewed away from local players.

That matters for mid-market Australian operators because the practical alternative to hyperscaler compute often involves smaller local AI vendors, niche integrators, and increasingly running models on-device. If those vendors cannot reach scale, Australian buyers narrow their options across the board.

Where the buildout goes next

There is no single fix. The lever government has used so far (planning approvals, energy contracts, headline investment figures) has rewarded scale and speed, not local access. Adding access conditions to the next round of deals is the obvious step Batra is calling for.

For Australian businesses building on AI today, there is also a parallel track. Local model deployment, smaller specialist vendors, and running compute on-device are increasingly viable substitutes when hyperscaler economics do not work. Our OpenClaw guide walks through how Australian businesses can run open-source AI agents locally without leaning on hyperscale compute.

Combined with a more deliberate ecosystem strategy from policymakers, that is the version of the AI boom that actually reaches local operators. For mid-market businesses thinking through this, AI strategy work increasingly means asking which vendors will still be around in three years given how the buildout is currently structured.

Scroll to Top