Owning servers used to make financial sense. The maths just changed.
When was the last time an SMB built its own office tower, generated its own electricity, or ran its own telephone exchange? Most Victorian businesses stopped doing any of those things decades ago. Not because the buildings, generators or PBXs weren't useful, but because the unit economics flipped. Owning the asset stopped being cheaper than renting access to it. Compute is the last utility-shaped expense that SMBs still buy on CapEx terms. The numbers held for about thirty years. They no longer do.
Three assumptions, all broken in twenty-four months
Three quiet assumptions sat under the CapEx case for owning servers, and all three have moved in the last twenty-four months. The first was that memory and storage got cheaper every refresh cycle, so the replacement cost trended down even as performance trended up. The second was that the supported life of the operating system roughly matched the physical life of the hardware, so the depreciation schedule made sense. The third was that capital was effectively free — borrowing or tied-up cash carried a near-zero opportunity cost. None of those are currently true.
The cost of capital came back
Take the capital cost first. The RBA hiked the cash rate to 4.35 per cent in May 2026, the third hike this year, fully unwinding last year's easing cycle. For an SMB with secured business lending, that translates to a borrowing rate around 6.5 to 7 per cent. That is the structural setting now, not a transitory spike. Forty thousand dollars tied up in a depreciating server in 2026 has a real opportunity cost that the same forty thousand dollars in 2020 did not. Even if the business isn't borrowing for the purchase, the cash isn't free. It could be earning a return somewhere else, or absorbing a working-capital shock you didn't plan for.
The depreciation maths cuts the wrong way
Now the depreciation side. The ATO treats server hardware as a four-to-five-year effective life asset under prime-cost or diminishing-value methods. A $36,000 server becomes roughly $9,000 a year on the P&L. That is a deductible expense, but only $9,000 a year — not the full $36,000 in the year of purchase. Operating expense, by contrast, is fully deductible in the year it is incurred. For a cash-flow-sensitive business, the difference between deducting $36,000 now and $9,000 a year for four years is material.
Then the asset itself. The CapEx model assumes a stable depreciating asset whose replacement cost trends downward. Server prices have moved sharply in the opposite direction. Dell's entry-level rack server now lists at nearly $37,000, up from around $12,000 two years ago, driven by AI demand pulling DRAM supply out of the conventional server market. Five-year-old equipment doesn't retain meaningful residual value, because the replacement assumption baked into the original depreciation schedule has broken.
Five-year-old equipment doesn't retain meaningful residual value, because the replacement assumption baked into the original depreciation schedule has broken.
The numbers, side by side
The pattern has already played out elsewhere
The mechanism behind this is the same one behind every other CapEx-to-OpEx transition that has already happened. Office space went OpEx because building tenancy became the right unit of consumption for businesses that needed flexibility. Electricity went OpEx because grid economics beat on-site generation by a wide margin. Telephony went OpEx because PBX maintenance was always overhead, not revenue. Compute resisted longer than any of them because the unit economics of a single tin box in a comms cupboard genuinely worked at SMB scale. Two recent trends closed that window: AI-driven hardware inflation, and OS lifecycle pressure that no longer maps to physical hardware life.
What this means at board level
For an SMB CFO running the numbers in 2026, the CapEx case for owning servers no longer survives a hard look. The depreciation schedule, the cost of capital, and the unstable asset value combine into an argument that doesn't add up. The transition is going to happen anyway. The only real question is whether it happens proactively, on the business's own timing and with a planned migration, or reactively, after a forced refresh hits at the worst possible moment.
Private cloud is the obvious landing point at SMB scale. The consumption shape matches how the business actually uses compute, and the per-VM pricing absorbs the underlying hardware market across the whole estate, rather than concentrating the price shock into one quote you have to approve. For a business that has historically refreshed every five years and is now staring at the next refresh, the OpEx conversation isn't about cloud as a buzzword. It's about whether owning a depreciating asset still passes a board-level financial test, and on the full TCO once power, cooling, HA and DR are counted, the answer is increasingly that it doesn't.
Communicat runs this analysis for Victorian SMBs as part of private cloud planning. If the next refresh is on the horizon and the maths is starting to look uncomfortable, that conversation is best had before the quote arrives, not after.
Frequently asked questions
Why doesn't owning servers make financial sense for SMBs in 2026?
Three assumptions that underpinned the CapEx case have moved in the last two years. The cost of capital is back at structural highs after the RBA's hikes. Server prices have roughly tripled at the entry level because AI demand pulled DRAM out of the conventional server market. And the supported life of the operating system no longer maps cleanly to the physical life of the hardware. Together they leave the CapEx case unable to survive a hard financial look.
Isn't depreciation a deductible expense that makes CapEx attractive?
It is deductible, just slowly. A $36,000 server depreciates at roughly $9,000 a year over four years under ATO prime-cost or diminishing-value rules. Operating expense, by contrast, is fully deductible in the year it is incurred. For a cash-flow-sensitive business, deducting $36,000 now beats deducting $9,000 a year for four years.
When does owning servers on-prem still make sense?
Specific sovereignty or compliance scenarios where regulation dictates the physical location of data — certain healthcare, government, and defence workloads, or contracts that prohibit cloud storage. Outside that narrow band the financial case has effectively closed for SMB-scale workloads.
Why is private cloud the natural landing point rather than public cloud?
At SMB scale, public-cloud per-resource pricing often runs hotter than expected when the workload is steady-state rather than bursty. Private cloud, sized to actual usage, gives you OpEx accounting, predictable monthly cost, and consumption that matches how an SMB actually uses compute — without the per-egress and per-API charges of the hyperscalers.
Written by
Managing Director
John Zammit is Managing Director at Communicat IT, a Melbourne MSP serving Victorian SMBs since 1987. He writes about cloud economics, infrastructure strategy, and the gap between sales narratives and operational reality.