Share with your CTO
The physical layer of AI is generating real revenue, not just narrative, and three infrastructure companies illustrate the spectrum of risk and reward. Digital Realty Trust (DLR) is sitting on a $1.9 billion signed-but-not-started backlog across 310 data centers, though profit margins compressed from 23.6% to 11.2%. Forgent Power Solutions (FPS) posted 88% year-over-year earnings growth supplying switchgear and power systems to AI data centers, while Vertiv (VRT), at a $103.5 billion market cap, is embedded in next-generation GPU deployments with a $15 billion backlog and a 14.4% net margin. The full analysis of all three AI infrastructure plays surfaces how each converts AI demand into cash flow rather than just headlines.
What this means for your business
CTOs still treating infrastructure as a procurement afterthought need to reckon with what these backlogs actually signal. A $15 billion backlog at Vertiv and near $2 billion in bookings at Forgent mean that capacity is already spoken for by hyperscalers and large enterprise customers. If your organization hasn’t locked in power capacity, colocation commitments, or cooling infrastructure, you’re not competing for the same queue as hyperscalers, you’re behind it. The companies on the winning side of this story moved 18 to 24 months ago.
The margin compression at Digital Realty deserves more attention than the headline backlog numbers. Profit margins dropping by more than half, from 23.6% to 11.2%, at the largest player in the space isn’t a rounding error. It suggests that building AI-grade data center capacity is expensive in ways that aren’t fully captured in the demand story. For enterprise buyers, that pressure eventually flows downstream as higher colocation pricing or stricter contract terms. Vendor relationships that looked stable under 2023 pricing assumptions are worth revisiting now, before renewal cycles force the conversation.
Vertiv’s concentration risk is the clearest falsification condition for the broader bull case. A $103.5 billion market cap built substantially on a small group of hyperscale customers means any slowdown in Meta, Microsoft, or Google’s capex cycles doesn’t just clip Vertiv’s stock, it creates a ripple through every supplier embedded in those data center builds. If hyperscale AI infrastructure budgets compress even 15% in 2027, the companies with diversified enterprise customer bases will prove more durable than the pure hyperscale plays, and that’s the vendor selection lens worth applying now to your own infrastructure stack.
Concept deep-dive: Signed-but-not-started backlog
A signed-but-not-started backlog is contracted revenue that hasn’t been recognized yet because the physical infrastructure, space, power, cooling, hasn’t been delivered to the customer. Think of it as a sold-out concert where the venue is still under construction. For enterprise buyers, it reveals how much of a provider’s near-term capacity is already committed, and how little optionality remains for latecomers trying to secure AI-grade data center space in the next 12 to 18 months.
Based on reporting from Vertiv Stock And Two AI Infrastructure Plays Backed By Data Center Demand, originally published 2026-07-30 08:44:00.
