Business · · Researched and written by TissoAI
Gartner warns of 3.5 trillion dollar AI infrastructure write-down
Gartner forecasts that up to 80 per cent of specialist AI cloud providers could fold by 2030 as soaring infrastructure costs outstrip projected software revenue.
A report from technology research firm Gartner warns that the technology sector faces up to 3.5 trillion US dollars in asset write-downs by 2030, with that exposure potentially climbing past 4 trillion US dollars by 2032.
The research outlines a widening divide between the capital being spent on data centres and the income generated by enterprise and consumer software. Infrastructure operators are on track to invest nearly 1 trillion US dollars during 2026 alone, with another 6 trillion US dollars in capital spending projected between 2027 and 2030.
According to Gartner senior director analyst Will Sommer, data centre owners would need to generate 8.5 trillion US dollars in cumulative revenue by 2030 to cover their capital spending, borrowing costs and an assumed 12 per cent return on invested capital. Gartner projects actual revenue attributable to those data centres will reach no more than 3.6 trillion US dollars over the same period, leaving less than half of the required return.
This shortfall creates direct pressure on specialised cloud providers known as neoclouds, which rent out processing capacity for machine learning workloads. Gartner projects that 80 per cent of these standalone operators will cease independent operations by 2030 through acquisitions, reorganisations or closures, because many lack diversified revenue streams to cushion heavy borrowing costs.
What this means for Australian businesses
For Australian small and medium business owners investing in automation, the forecast is a prompt to review where tools and data are hosted.
- Companies renting compute directly from smaller, dedicated machine learning cloud providers should ensure they have contingency arrangements in place for pricing spikes or platform closures.
- Avoid long multi-year lock-ins, as commercial terms and consumption pricing will shift when infrastructure providers face margin squeezes. Favour agreements with clear exit terms and portable workloads.
- Australian owners should deploy tools only where they solve measurable operational problems, such as administrative backlogs or client response times, rather than spending money on speculative compute projects.