Theses›Datacentre & AI compute›Rack-scale design automation

Rack-scale design automation

Rack-scale multiphysics design automation captures a layer above chip-EDA by 2030, but only as an AI-native tool for the non-hyperscaler builders (neoclouds, sovereign-AI, enterprise AI factories) the incumbents underserve

  • Bear
  • Closing
  • $1.9B by 2035

The callThe rack is now the unit of compute and designing it is a coupled physics problem, but Cadence and Synopsys paid about $48B to own that layer before any startup began.

D1D2R1R2R3C1foragainst D1D2R1R2R3C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the restraints.
  1. D1The rack is the unit of compute, shipping at 120kW and heading to 600kW-1MW, and past 120kW power, cooling and interconnect have to be solved as one problem.
  2. D2The incumbents are priced for chip teams and hyperscalers, and NVIDIA's blueprint covers only NVIDIA's rack, leaving neoclouds and sovereign builders underserved.
  3. R1The incumbents bought the layer three times in 20 months, about $48B including Synopsys-Ansys, before any startup started.
  4. R2NVIDIA gives away a design-and-operations blueprint, and recent exits nearby are acqui-hires: Emmi AI, Run:ai, CentML.
  5. R3The 10x revenue multiples here come from franchise purchases nobody pays a young company, and buyer willingness to pay is unproven.
  6. C1NVIDIA's Kyber Rubin Ultra NVL576 racks ship at about 600kW on 800V DC, turning every rack into a coupled power, cooling and interconnect design problem.H2 2027
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
20272028today1234predictionswatch 20272028today1234predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1By end-2027 no independent rack-scale design-automation startup raises a Series A >=$20M with a defensible, non-absorbable multiphysics rack-design moat selling to neoclouds / sovereign-AI / enterprise AI-factory builders.Dec 2027 · open
  2. 2An independent rack-design company raises a $20M+ Series A with a moat the incumbents cannot absorb; I expect none does.Dec 2027 · open
  3. 3Kyber ships at about 600kW: the demand checkpoint.Dec 2027 · open
  4. 4Cadence's system design and analysis growth; single digits would mean the layer was capex beta rather than a structural shift.Dec 2027 · open
Expressionslisted, private, technologies
  • SNPSSynopsys
    • full
     EDA duopoly with Cadence.
  • SIE.DESiemens Digital Industries Software
    • expensive
     Sells electronics-cooling and CFD simulation covering packages through enclosures to whole data centres (Siemens AG unit; ticker is the parent's).
  • CDNSCadence Design Systems
    • not screened
     Sells system thermal, signal/power-integrity and data-centre CFD digital-twin tools (Future Facilities, acquired 2022) to system and data-centre designers.
  • DSY.PADassault Systèmes
    • not screened
     Sells SIMULIA CFD and electromagnetic simulation used in data-centre and electronics design, and contributes to NVIDIA's DSX blueprint.
  • NVDANVIDIA
    • cheap
     Supplies the reference rack and AI-factory designs (Vera Rubin DSX) plus free simulation libraries that rack builders design against.
Private · 2 private companies, 2 of them in the US.

AlphaTango · cove (Vitras.ai)

Technologies
Who has already bet on itprivate rounds, from public sources

raised by 1 company, $25M of it in the last three years. Early: Little money in so far; a lead investor would set the pace.

  • ·cove (Vitras.ai) Rounds not itemised