Theses›Semiconductor manufacturing›EDA and chip design

EDA and chip design

AI moves chip design onto the engineering bill; physical design and signoff sold per design is the layer the incumbents cannot copy without eating their seats.

  • Bull
  • Closing
  • $62B by 2035

The callThe drivers win, narrowly: AI reprices chip design against the engineering salary bill, a bigger number than the tools budget, and the incumbents cannot sell per design without cannibalising their seats. If this is right the market goes from $11B to about $62B by 2035, with about $2.5B open to newcomers.

D1D2D3R1R2R3C1foragainst D1D2D3R1R2R3C1foragainst
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 drivers.
  1. D1AI agents start doing the work engineers do, so the money moves off a tool-licence budget and onto the much bigger engineering salary bill.
  2. D2Every company with a big workload wants its own silicon, so design starts rise and new entrants need a signoff flow outside the big three.
  3. D3The incumbents sell per seat, and a tool that removes engineers removes seats, so they cannot price per design without eating their own licences.
  4. R1The incumbents own the design corpus, signoff simulators and process kits, so a tool outside their signoff does not ship, and one inside it gets bought.
  5. R2The European lanes are empty or unproven, with the visible companies small.
  6. R3The US entrants are priced as if absorption does not apply to them: Ricursive at $300M raised on a $4B mark.
  7. C1Annual ASIC and SoC design starts rise from about 2,000 to about 5,000, so many new entrants need a signoff flow outside the big three.2032
Rests onthe beliefs underneath
Supported by · confidence
Threatened by · confidence
What would prove itdated tests, open and resolved
20272028today12watch 20272028today12watch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1A European AI-native EDA company raising a €10M+ Series A with a moat outside the absorb zone, say a simulator-agnostic surrogate across process kits.Dec 2027 · open
  2. 2Whether ChipAgents is still independent, which would push the absorption odds down.Dec 2027 · open
  • ·The big vendors charging per design for automated physical design themselves, which would change the view.
Expressionslisted, private, technologies
  • 3529.TWOeMemory Technology
    • expensive
  • RMBSRambus
    • expensive
  • SNPSSynopsys
    • full
     EDA duopoly with Cadence.
  • CDNSCadence Design Systems
    • not screened
     Digital implementation, custom/analog design, simulation (Xcelium), emulation (Palladium) and AI agents (ChipStack), licensed to chip design teams.
  • SIE.DESiemens EDA (Siemens AG)
    • expensive
     Calibre physical verification, Questa verification and custom IC tools (ex-Mentor Graphics), licensed to chip design teams.
  • KEYSKeysight Technologies
    • cheap
     RF, microwave and high-speed design and simulation software (PathWave ADS), licensed to RF and IC designers.
  • 301269.SZEmpyrean Technology
    • not screened
     Analog/custom and flat-panel-display EDA tools, licensed mainly to Chinese chip designers.
  • 688206.SSPrimarius Technologies
    • not screened
     SPICE circuit simulation, device modelling and yield tools, licensed to foundries and chip designers.
Private · 5 private companies, 3 of them in the US; none in continental Europe.

ChipAgents · Cognichip · Ricursive Intelligence · WaveForge · X-Epic

18 more private companies tracked, not shown

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

$593M raised by 5 companies, $535M of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a top-tier venture firm: Ricursive Intelligence, ChipAgents, Cognichip. Backed by a strategic investor: Ricursive Intelligence, ChipAgents.

What changedthe record, newest first
  • 12 Aug 2026Second European chip-design-tools company in a week, and both are working on the same narrow slice - analog. I do not think that is random. Analog circuit design leans on tacit expertise that never got written down, which makes it the slice where the big incumbents'' data advantage is weakest and where their problem of training on customers'' own designs bites hardest. The caution I owe myself: I have been looking for exactly this pattern, so two examples prove very little. The real test is whether it keeps showing up in deals I did not go looking for.
  • 5 Aug 2026I passed on a chip-design company whose core argument I think is right. The technology answered better on the call than it read on paper, and the strategic risk I had written down turned out to be weaker than I thought. It still died, on price: the valuation was asking me to pay for research that has not happened yet. Worth separating the two questions, because they get collapsed constantly - is the bet right, and is this the right price to take it at. A good argument at the wrong entry price is still a bad investment, and the discipline is noticing that before the argument seduces you.
  • 5 Aug 2026I logged a chip-design company two days ago as living inside the incumbent's tool and therefore easy for the incumbent to absorb. On the call that turned out to be the sales channel, not the architecture: they had already connected two large chipmakers' own in-house simulators, in days, at those customers' request, before doing the incumbent integration. The lesson is not about this company. It is that a plugin is visible and an API is not, so reading a deck tells you where a product is SOLD and almost nothing about where it is LOCKED IN - and absorption risk is a question about the second one.
6 earlier
  • 3 Aug 2026The chip-design duopoly shipped agentic verification products within seven months of each other, which should have been the end of the startups attacking that layer. Instead the largest independent in the category just raised again six months after its last round, on six-times revenue growth through exactly the period the incumbents were shipping. Either the market is big enough that the incumbents shipping does not close it, or the growth is coming off a base small enough not to matter yet. Twelve months tells you which, and the tell is whether the company is still independent.
  • 3 Aug 2026The interesting claim in AI chip design this month is not that agents design better circuits - it is that agents break the simulation supply chain. A human designer runs few simulations because intuition prunes the search; an agent substitutes data for intuition and runs ten to fifty times as many. If that is right, the binding constraint on agentic hardware design is simulator throughput, not model quality, and the layer that stands in for the simulator is infrastructure rather than tooling. Two European companies pitched me versions of that argument in a fortnight. Both are betting against a duopoly that owns the simulator they train on.
  • 6 Jul 2026Passed a European 'turn-your-algorithm-into-a-chip' startup this week. Two things kept it a watch not a bet: their own five-year market topped out around €70–150M, and the clever version of the pitch (skip the lossy C-to-hardware step, compile straight from the maths) just moves the moat problem up to the layer NVIDIA and AMD are busy building in-house. The real competitor was never a rival startup, it was an NVIDIA Jetson.
  • 24 Jun 2026The AlphaChip founders raised $300M at $4B to commercialise the approach — the strongest validation of the AI chip design thesis yet. Capital still routes to US teams; the European open-EDA lane remains empty and the right one to back.
  • 23 Jun 2026The capital is real (two US agentic-EDA entrants now past $60-74M each) but it is all flowing to US teams and incumbents, not to a European breakout.
  • 22 Jun 2026The way past Cadence and Synopsys is not a better tool they can buy, it is a free or open EDA flow whose whole business model they cannot copy without blowing up their own licence revenue.
Reading
Show the reading