Nvidia reported Wednesday that cloud industry backlog now exceeds $2 trillion, with capital expenditure by the top five hyperscalers expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027. That number is a mandate for every layer of the data-center stack below the GPU. The question the investment committee asked this week: which layer do you actually want to own?
Three names reported in the same cycle and produced wildly different results. The dispersion is the signal.
Equipment: Vertiv (VRT)
Vertiv is the highest-margin business of the three on an adjusted operating basis. Adjusted operating margin reached 22.6% in Q2 2026, up 410 basis points versus the prior year. Adjusted free cash flow hit $925 million, up 234% year over year, with conversion exceeding 150% of adjusted operating profit, and net leverage flipped to negative 0.1 times as Vertiv closed the quarter in a net cash position. That is an industrial company performing like a software business.
The stock fell roughly 17% because revenue of $3.274 billion, while up 24% year over year, fell short of analyst expectations of $3.38 billion. Vertiv attributed the shortfall to timing factors rather than demand weakness, specifically “temporary supply chain congestion and multi-phased project execution as deployments scale in size and complexity.” For the full year, Vertiv raised guidance, now forecasting adjusted earnings of $6.65 to $6.75 a share on net sales of $13.8 billion to $14.2 billion.
The committee’s read: the revenue miss was a phasing event, not a demand event. Capital expenditures are expected to reach about 4% of 2026 sales, which is modest for an industrial company growing at this pace. The margin structure and cash conversion put Vertiv in a different category than traditional equipment suppliers. The risk is valuation: a price-to-earnings ratio above 57x reflects years of expected growth being priced in today, and when a high-multiple name misses on revenue, even slightly, the math magnifies the reaction. A second consecutive supply-chain quarter would force a genuine reassessment.
Interconnection: Equinix (EQIX)
Equinix reported no such drama. Revenue reached $2.625 billion in Q2 2026, up 16% year over year, with a record adjusted EBITDA margin of 53%. Annualized gross bookings reached $424 million, up 23% from the year-ago period, representing the second-highest volume on record. The company added a record 9,700 net interconnections during the quarter, bringing the total to 522,700.
For 2026, Equinix now expects revenue growth of 11% to 12% and AFFO per share growth of 10% to 12%, which management described as the largest single guidance raise in the history of the company.
The structural argument for Equinix rests on one fact: wholesale REITs are renting megawatts; Equinix is renting megawatts plus the network meeting room where tenants connect to clouds, carriers, and each other. That moat is the reason Equinix sustains a 51% to 53% adjusted EBITDA margin while wholesale peers run closer to 40%. Interconnection revenue is the highest-margin product line and growing faster than colocation. The offset is capital intensity. Equinix’s economics only make sense if investors hold two ideas at once: the revenue quality is unusually strong, but the model remains capital intensive. Equinix’s leverage is typical for the REIT structure but worth watching as rate conditions shift.
Integration: Celestica (CLS)
Celestica is the most volatile and the most misunderstood. Q2 revenue reached $4.70 billion, up 62% versus Q2 2025. Adjusted operating margin of 8.2% set another record. GAAP EPS was $3.17. Adjusted return on invested capital reached 55.5%, twenty percentage points better than a year ago. Full-year revenue guidance was revised upward to $20.5 billion, representing 65% year-over-year growth.
Then, on August 5, Celestica announced a $3 billion treasury offering of common shares. The company priced 9,677,419 shares at $310 per share, and later completed the offering with 11,129,031 shares sold after the full exercise of the underwriters’ option. The raise introduces immediate dilution of roughly 8% to 10% depending on the final share count, and extends return timelines, creating an EPS headwind. That is the trade’s friction. The business is accelerating; the share count is growing to pay for it.
The 8.2% adjusted operating margin is a company record, and it matters more than the EPS line: Celestica is a contract manufacturer, a business model that historically lived on low-single-digit margins. Margin expansion is the re-rating thesis. The model is capital-light relative to Equinix, but earnings dilution from the offering makes this a position to size carefully rather than add aggressively at current levels.
Committee Decision
The committee allocates fresh capital to Equinix as the core, permanent holding within the data-center theme. The 53% adjusted EBITDA margin, record interconnection density, and recurring revenue base compound through cycles regardless of which GPU architecture wins next year. Vertiv earns a meaningful satellite position: the margin expansion and cash generation are real, the supply-chain miss is almost certainly timing, and the full-year guidance raise confirms demand has not slowed. Celestica is a trade, not an allocation, until the dilution from the August offering is absorbed and the margin expansion proves durable at the new revenue scale. The committee will revisit in November.
