Vertiv Orders Double. Is Power and Cooling AI's Next Bottleneck?
Vertiv's third-quarter results point to strong demand for data center power and cooling gear. They do not show that the bottleneck has fully moved from chips to electricity.
Vertiv's third-quarter 2026 results, released Oct. 28, are about more than one revenue growth rate.
Orders more than doubled, and profit grew faster
| Key metric | Q3 result | Change vs. a year earlier |
|---|---|---|
| Revenue | $4.313 billion | +61% |
| Organic revenue growth | 55% | Excludes acquisitions |
| Organic orders | - | Up more than 100% |
| Adjusted operating profit | - | +83% |
| Adjusted operating margin | 25.2% | +2.9 points |
| Adjusted EPS | $2.21 | +78% |
| Operating cash flow | $794 million | +64% |
| Adjusted free cash flow | $652 million | +50% |
Adjusted figures are non-GAAP, which exclude certain items.
Adjusted operating profit grew faster than revenue, so profitability improved as the company grew. The rise in operating cash flow and adjusted free cash flow is also a positive. Earnings are turning into cash, not staying as accounting profit.
Vertiv raised the midpoint of its full-year revenue outlook by $650 million, to $14.65 billion. For the fourth quarter, it expects revenue of $4.413 billion at the midpoint and organic growth of 50%. These are company projections, not reported results.
The AI data center race is also fought outside the chip
An AI data center does not work just because GPUs are installed.
GPUs do the computing, and HBM feeds them data quickly. But if power does not arrive reliably, or the heat cannot be removed, even expensive chips struggle to perform fully.
Three layers of AI infrastructure
- Compute (GPUs): the ability to train and run AI models.
- Memory (HBM): the ability to deliver data to the processors fast enough.
- Physical infrastructure (power, distribution, cooling): the ability to supply electricity and remove heat so chips run reliably. This is where Vertiv mainly operates.
The three layers do not replace one another. They are complementary, and all three have to be in place.
One unit investors may need to watch more closely is the gigawatt (GW). GPU purchases show how much compute has been secured. A data center's power capacity helps show how much compute infrastructure can actually be operated.
Contracted power capacity, power actually used and power delivered to GPUs are different things. GW alone cannot be used to calculate AI performance or revenue.
Vertiv's results are evidence of strong demand for the physical equipment behind AI. They are not grounds to conclude that the industry-wide bottleneck has already moved fully from GPUs to power.
Do doubled orders mean doubled future revenue?
That is hard to say yet.
Customers may have ordered early out of fear of shortages, or ordered from several suppliers at once. If orders are canceled or deliveries rescheduled, revenue may not grow as fast as orders.
Even when equipment is ready, a data center cannot run if grid connections, construction permits or actual facility completion are delayed. That can affect delivery timing and customers' investment schedules.
If the stock already prices in high growth, valuation could adjust the moment results fall short of expectations, even while earnings keep rising.
Five things to watch
- Backlog: how fast new orders accumulate and convert into revenue.
- Order cancellation rate: whether early or duplicate orders show up as actual cancellations. If disclosure is thin, check order adjustments and delivery changes too.
- Customer concentration: how a spending cut by a few large customers would affect results.
- Liquid cooling share: how much the cooling technology needed for high-density AI servers contributes to revenue and profit.
- Pace of capacity expansion: whether added production shortens delivery times or raises the risk of oversupply.
Conclusion
Vertiv's results do not say AI chip demand is over. They show that growth in AI compute is pushing up demand for power and thermal management equipment too.
For long-term investors, the question is not the next fashionable stock. It is the whole industrial structure needed before AI compute actually runs and earns a return.
Insight Times Editorial Desk
