The Software-Defined Answer to Hardware Inflation
Presentation version, built as a backdrop rather than a document. Enter your credentials to continue.
The Software-Defined Answer to Hardware Inflation
Not a shortage. A repricing.
- Nothing broke. No flood, no fire, no backlog
- Supply is at record levels and still rising
- A new buyer took it. AI capex roughly doubled in a year
Capacity curves take three years to bend.
A ten-year assumption broke in four months
- Per terabyte, per year, for a decade
- Reversed January to April 2026
- Quote validity cut from 90 days to 30
Blue DRAM, orange NAND. Contract prices, TrendForce. Never negative.
Do you build out of a problem this profitable?
- Western Digital sold out all of 2026
- Kioxia declared 2026 NAND sold out
- Counterpoint: no correction in the second half of 2027
Plan for the constraint, not for its end.
The cloud was the escape hatch. Then it filled up.
- Azure UK South and UK West hit capacity limits
- Same DRAM. Same NAND. Same shortage
- Hyperscalers allocate to whatever earns most
A fine bridge for one dataset. Not an exit.
What is being squeezed
Every budget is absorbing it
- Next 12 to 18 months: 22% significant, 55% moderate
- Only 2% expect no impact at all
Cost is the headline. Not the story.
Consolidation stopped being optional
What they plan to buy next
Among non-users. Rank order only. Positions five through eight sit inside sampling error.
The market is raising its own failure rate
Premium prices for data nobody can name
- Delete it and lose whatever it was worth
- Keep it and pay 2026 prices to store it
What buyers said
- Non-users ranked Software-defined storage first of eight. Hyperconverged placed sixth
- 82% of current users increasing investment. Highest in the study
- 98% consolidating layers. Two percent are not
- 74% pulling protection into primary storage
Positions five through eight sit inside sampling error of one another.
UCI is what software-defined storage becomes once you stop stopping at storage
- SDS decouples storage software from storage hardware. One layer
- That 98 percent says the problem is four layers
- Each carries its own buffer, refresh, contract, spares
- Fix storage alone and three-quarters of the bill stays
The ranking and the 98 percent are the same finding measured twice.
We were surprised by the number of organizations that are considering software-defined storage as a potential solution. HCI is the epitome of software-defined, and VergeOS deserves serious consideration.
Simon Robinson, Chief Analyst, OmdiaThree questions for any vendor
Passing
Reduction across the whole pool. One metadata model, so a block stays reduced in the copy, the replica, and at DR.
Failing
Scoped per volume or per backup job. You pay for the same block four times.
Passing
Mixed drives, mixed capacities, mixed server generations. Standard enterprise hardware.
Failing
One vendor’s compatibility list decides what you are permitted to buy.
Passing
Snapshots, replication, and recovery where the workload already runs. The second array comes off the plan.
Failing
Ask Tesco. New platform, and the Veeam and Zerto stack did not follow.
Select a question to open it
Two you cannot lose. Or twenty you can.
- Two servers plus external shelves
- All I/O through the appliance
- Lose one, exposed. Lose two, down
- All four functions on every server
- Lose one, it adapts. Lose another, again
- No server worth naming
Buy five old servers instead of one new one
- Any vendor, any generation, any condition
- Five refurbished 32-core boxes beat one 200-core box
- More failure domains means more resilience
- The hyperscalers refreshing their fleets feed that channel
Telemetry migrates before it breaks. Ignore it and workloads restart in seconds.
Nobody makes flash cheaper. That is not the gap.
Every byte lands on media somebody bought at 2026 prices. What is not fixed is the gap between the capacity an enterprise buys and the workload it serves.
- Ask what is allocated. Then what was written in 90 days
- Cut the number of architectures
- Move cold data down. 22.6x against disk, up from 4.9x
- Stop buying a second set of hardware to protect the first
None of it is exotic. Most organizations have never measured any of it.
Tested on a live system
- Keeps serving through the failure
- No race against a rebuild window
- Running hot stays defensible
- Workload stays up during the move
- Between media types and generations
- No maintenance window