Accelevation's $18 IPO Was Priced 10% Below the Mark, and the Multiple Stack Says Is Worse for the Neoclouds Behind It

The marketing deck reads like a dream. Revenue grew 147% to $447.8M in FY2025 and another 176% to $437.5M in H1 2026 (TTM $726.6M). Backlog stood at $1.1B on Jun 30, 2026. Adjusted EBITDA hit $90.9M in FY2025 and $68.4M in H1 2026. Accelevation was working with three hyperscalers at the full-platform design level.

Strip out the PE-owned narrative and Accelevation is, structurally, a power-distribution and rack-integration shop — the same category as the established incumbents Vertiv and Eaton. Those incumbents earn15–18% operating margins, throw off free cash, and trade at 6–8x EV/Sales. Accelevation cleared at roughly 5x forward sales, which is exactly the multiple an industrial supplier should clear at, not the hyperscaler multiple a Lambda or Nscale might argue for.

Accelevation's IPO clear prices the headline price of capacity at roughly 5x sales. The listed neocloud complex trades at 12–60x. That gap is the read-through — the structural question isn't whether AI capacity gets built, it's whether anyone outside Microsoft, Alphabet, and the rest of the hyperscaler oligopoly deserves a hyperscaler multiple for owning capacity that isn't their own demand.

The neocloud model isn't just a margin story; it's a balance-sheet story. Over $176B of GPU-collateralized debt now sits across hyperscaler and neocloud balance sheets, with private credit funds underwriting neocloud loan facilities against residual values that depend on rental rates staying high and chips staying useful. CoreWeave alone has grown total debt from ~$2B at end-2023 to roughly $35B by mid-2026, anchored by an $8.5B investment-grade-rated GPU-backed facility closed in March 2026. Lambda secured a separate $1B debt facility in August 2026 specifically to buy Nvidia chips it will lease to Microsoft.

The valuation chain runs GPU book value → debt capacity → equity value. If the IPO tape is the new multiple, debt capacity shrinks, which forces equity raises at lower marks, which forces mark-to-market writedowns on the existing private credit book. The first crack in the wall was Groq in August; the second is Accelevation on Sep 30. Nscale and Lambda are the next two IPOs in the queue.

Short-term (days–quarters): the catalyst calendar is the next two IPOs. Nscale has filed for a US listing under NSCL with a $35B aspirational tag; Lambda is working a $12B+ round before a H2 2026 listing. Either deal landing materially below its private mark ratchets the listed names — especially CoreWeave, the only one with public debt spreads to watch — by 10–20%. The 52-week ranges already show the dispersion: CoreWeave is at $87 against a $60–$153 range; Nebius at $236 against a $73–$300 range. Volatility is the upper bound.

Long-term (1–3 years): the structural winner is the established supplier that already trades at industrial multiples — Vertiv at ~8x EV/Sales with 38% gross margins and a $93B market cap, and Eaton at ~6x with 36% gross margins and a $167B market cap. Their thesis (data-center power and rack infrastructure) doesn't depend on the neocloud operators' balance sheets. Their risk is that any AI capex retrenchment cuts order books; their cushion is that they sell into hyperscalers directly and earn their multiples with cash flow, not mark-to-market debt. NVIDIA sits in between: chip demand is more diversified than any single neocloud customer, but neocloud capex cycles move its data-center segment by double-digit points quarter to quarter.