Final Tuesday, Satya Nadella stood beside the Prime Minister and dedicated$25 billion to develop Microsoft’s Australian cloud capability by 140% over 4 years. Huge numbers, huge stage. The identical speech he had just lately delivered in India, Canada, Japan and the UK.
Just a few days later the books opened on what flows the opposite means. Newly filed accounts present Google and Meta transferred near $11 billion of Australian income offshore final yr in service charges and reseller funds, paying simply over $140 million in tax between them. Amazon booked nearly $12 billion in native income, with its information centre arm up 50%.
The Australian Bureau of Statistics noted earlier this year that the surge in information centre funding has been largely offset by a corresponding rise in imports, with restricted internet affect on GDP. The e61 Institute, an unbiased assume tank, factors out that only around eleven and a half thousand workers are employed in the relevant local industry. None of those figures sit comfortably with the language of nationwide funding.
The press conferences cowl the influx. The accounts cowl the outflow.
The race they want you to run
Many of the dialog about AI in Australia is framed as a race. Transfer quicker. Undertake sooner. Don’t fall behind. The visiting CEOs reinforce it as a result of urgency is sweet for procurement. The consultants reinforce it as a result of urgency is sweet for charges. The commerce press reinforces it as a result of urgency is sweet for headlines.
A technology of Australian executives is being advised, with rising confidence, that the worst factor they’ll do just isn’t run. There’s a race. The reframe value holding onto this week is that it isn’t a race to deploy.
What the hyperscalers are promoting is pret-a-porter. The identical swimsuit, made to straightforward sizing, hanging in each store window from Sydney to Sao Paulo. Purchase it shortly sufficient and the salesperson will let you know it matches.
Working the identical Microsoft stack as everybody else doesn’t generate aggressive benefit. It generates parity, which is the alternative.
A technology of Australian executives is being advised, with rising confidence, that the worst factor they’ll do just isn’t run. There’s a race. The reframe value holding onto this week is that it isn’t a race to deploy.
The Pioneer markets on the high of the EY adoption index aren’t successful a dash. They’ve embedded AI into infrastructure they already personal. India layered AI onto UPI, the general public cost rail it had spent a decade constructing, and turned its fintech sector into one thing no person else can replicate with out first proudly owning the rail.
The benefit isn’t velocity; it’s what sits beneath.
The actual race isn’t who deploys AI quickest. It’s who owns what they deploy.
The high quality print
The AFR reported one native instance of that alternative:
Telstra’s board now not has to recollect each element of each venture, as a result of Telstra constructed its personal AI bot inside its personal perimeter.
In early 2025, earlier than the hyperscalers had any sovereign-flavoured merchandise to promote, Telstra had entry to each main vendor within the nation and will have signed with any of them. As an alternative, they consolidated 18 distributors down to 2, dedicated $700 million over seven years to a three way partnership with Accenture, and began constructing.
They didn’t dash. They consolidated, managed, and constructed one thing no person else has.
JPMorgan did the identical at a lot bigger scale with its LLM Suite, now serving 230,000 workers on vendor fashions wrapped contained in the financial institution’s personal governance layer.
These aren’t outliers. They’re early indicators of the place the economics are heading. Arvind Krishna advised IBM’s January earnings name that inside three to 5 years, half of all enterprise AI usage will move into private cloud or on-premises environments, pushed by each management and price.
The hyperscalers learn the identical numbers, which is why each one in all them now sells a “sovereign” model of their product, racing to be the inside-the-firewall supplier earlier than prospects construct their very own.
What the press releases don’t point out is how a lot their model of sovereignty prices.
Google Distributed Cloud, for instance, begins at US$415 per node per month on a 3 or 5 yr dedication, earlier than you add Gemini utilization, storage, networking, integration or skilled companies. The air-gapped variations, constructed for essentially the most delicate deployments, are quote-only. These are Telstra-scale options with Telstra-scale value tags and lock-in intervals.
For many Australian organisations, this isn’t the form of sovereignty they’ll afford.
However the precept is the suitable one, and it doesn’t take $700 million to use.
A bespoke swimsuit isn’t quicker than off-the-rack. It’s slower, prices extra upfront, and lasts significantly longer. Deal with frontier intelligence as a commodity. Personal the layer that sits on high: the information, the governance, the workflows, the institutional data. The mannequin you may hire.
An rising cohort of startups, together with the one I’m constructing at New Dialogue, is now making precisely this sort of functionality accessible to organisations that aren’t Telstra.
The query value asking isn’t whether or not you may afford hyperscaler scale sovereignty. It’s what a part of your AI deployment you truly must personal.
Sitting on the backside
Australia sits within the bottom-left of the EY adoption index: low adoption, low sentiment. The sincere studying isn’t that we’re being virtuously cautious. It’s that we haven’t but determined what an Australian-context AI deployment truly seems like.
Whereas we determine it out, each hyperscaler on the earth is queuing on the door with the identical off-the-shelf answer, and the meter is operating on the best way out in addition to in.
The work of determining what to personal isn’t gradual. It’s simply deliberate, which is one thing the running-shoe sellers can’t invoice for. One of the best response to a uniform product isn’t to purchase it quicker than the subsequent nation. It’s to construct one thing the subsequent nation can’t.
Yet another factor
A few many years in the past, Google’s first try at enterprise {hardware} was a shiny yellow field referred to as the Search Equipment.
You purchased it, plugged it in, and bought Google search inside your firewall.
It ran for seventeen years earlier than Google quietly shut it down, at which level the field turned a really costly paperweight.
Distributed Cloud is similar form, three mannequin generations later. Price remembering, because the packing containers begin arriving.
- Matt Vitale is cofounder of New Dialogue, an Australian sovereign AI platform that helps enterprises deploy AI privately, securely, and with out vendor lock-in. He’s a lawyer by background and Australian Fintech Chief of the Yr 2024.
