SMCI · Forward model · OEM Appliance & Large Data Center · Bear case
What has to happen in OEM Appliance & Large Data Center
Model as of
This page changes OEM Appliance & Large Data Center inside the complete SMCI model. Every other vertical, corporate programme and valuation assumption stays in the selected portfolio case.
Shares this vertical and portfolio case. Slider and horizon edits stay in your browser.
OEM Appliance & Large Data Center
Basis quarter$5.50B
Final quarter$8.35B
Implied CAGR+9%
Final revenue mix58%
The AI factory business: hyperscalers, neoclouds, sovereign programmes and the very large private AI builds, buying liquid-cooled GPU racks by the row. It is where the growth is and where the concentration risk is — a single customer was 28% of FY2026 revenue, and nine customers each spent over $1 billion. What limits it is not demand but how many racks can be allocated, built and energised in a quarter.
Last four quarters
2026 Q1
$3.85B
Estimated
2026 Q2
$9.74B
Estimated
2026 Q3
$7.86B
Estimated
2026 Q4
$5.50B
Reported
Liquid-cooled GPU SuperClusters sold as complete racksAir-cooled GPU servers for large data centresDCBBS: cooling, power shelves, networking and deployment services sold with the computeOEM appliance business
Units
2000/qtr
growing +21.0% per quarter
About 2,000 rack-equivalents in the basis quarter — $5.5B at an assumed $2.75M a rack. Volumes are not disclosed.
Price per unit
$3M
drifting +1.0% per quarter
$2.75M a rack, an NVL-class liquid-cooled rack. This is the assumption doing the most work; argue with it first.
OEM Appliance & Large Data Center
Latest: $8.35B (2031Q4E)
| Period | Value |
|---|---|
| 2026Q1 | $3.85B |
| 2026Q2 | $9.74B |
| 2026Q3 | $7.86B |
| 2026Q4 | $5.50B |
| 2027Q1E | $6.52B |
| 2027Q2E | $7.71B |
| 2027Q3E | $9.08B |
| 2027Q4E | $10.69B |
| 2028Q1E | $12.55B |
| 2028Q2E | $14.06B |
| 2028Q3E | $13.61B |
| 2028Q4E | $13.15B |
| 2029Q1E | $12.70B |
| 2029Q2E | $12.25B |
| 2029Q3E | $11.81B |
| 2029Q4E | $11.38B |
| 2030Q1E | $10.96B |
| 2030Q2E | $10.55B |
| 2030Q3E | $10.15B |
| 2030Q4E | $9.77B |
| 2031Q1E | $9.39B |
| 2031Q2E | $9.03B |
| 2031Q3E | $8.68B |
| 2031Q4E | $8.35B |
Assumptions & reasoning
- Modelled on units rather than growth because the binding constraint is physical and publicly stated: Super Micro puts global manufacturing capacity at 5,000 racks a month, of which 2,000 are direct-liquid-cooled. The ceiling here is the DLC figure — 6,000 a quarter — not the headline one, because liquid cooling is what an AI factory buys and it is the narrower constraint. Even that is generous: what actually rations output is GPU allocation and the customer's power and networking readiness, which is the reason management gave for the fiscal Q3 2026 revenue miss — customers not ready, not factories full.
- The rack count is NOT disclosed. It is backed out of the reported $5.5B at an assumed $2.75M per rack, so treat the level as a calibration and argue with the price instead. If a rack is really $2.0M the volume is 2,750 and the ceiling binds sooner; if it is $3.5M the volume is 1,570 and it binds later. The revenue path barely moves either way, which is the point of splitting price from volume rather than hiding both in a growth rate.
- Margin falls rather than rises in the base case, and that is a deliberate disagreement with management. The basis quarter carried a 17.5% gross margin against 10.8% for FY2026 as a whole, on what the company itself described as a favourable customer and product mix. Taking the peak quarter as the new normal would flatter every projected year. The DCBBS thesis — that selling cooling, power, networking and services with the compute lifts margin structurally — is real and is what the Bull case pays for; it is not what the base case assumes.
- This vertical carries the concentration risk for the whole model. One customer was 28% of FY2026 revenue and is widely linked to the xAI/SpaceX Colossus build. Nine customers were each above $1 billion. There is no diversification inside this line to fall back on: if the largest programme pauses, roughly a fifth of company revenue pauses with it, which is what the Concentration case tests.