The layer that was essential and did not rise
Every chip design in the world passes through one narrow chokepoint: EDA. Synopsys, Cadence and Siemens EDA oligopolize the market, and no fabless design and no foundry sign-off becomes silicon without these three companies' tools. Look only at the terrain and this is the narrowest, most essential chokepoint on the chain. Yet in this AI cycle its stock did not rise. Synopsys was negative two years running, -3.2% in 2025 and -3.1% in 2026, and Cadence managed only +4% in 2025 and +17.8% in 2026. Set against the power chain, which re-rated all the way through generation, transmission and cooling, the stall stands out.
This is not to say EDA is an asset that cannot rise. Synopsys was +61.3% in 2023. It simply went still in 2025-26, exactly when the physical-bottleneck layers surged. The location of that stall is the clue.
Essentiality is not sufficient for a re-rating
Set EDA beside the other chokepoints over the same window and the gap is stark. In memory, Micron printed +240.5% in 2025 and SK hynix +361.1%; in equipment, ASML re-rated +55.8% and +72.7% across 2025 and 2026, Lam Research +139.5% and +135.6%, and KLA +94.5% and +113.5%. EDA is, if anything, more essential than these, and its stock still lagged. Essentiality is not sufficient for a re-rating.
This was not weak results
Clear the obvious misreading first. This is not a demand miss. Synopsys FY2025 revenue rose 15% to $7.05B and Cadence 14% to $5.30B. Double-digit growth is not the face of a stalling cycle. Results held; only the stock stood still. The size of the revenue is not the issue. Its nature is, and that nature turns on a single physical condition.
The propagation arrives only where it rides a physical capacity bottleneck
The reason memory, equipment and substrates exploded converges on one thing: their physical capacity has a ceiling. HBM wafers, CoWoS slots, EUV lithography all share it. When demand crowds in, volume cannot expand that quarter, so price jumps, and that price translates instantly into earnings and share price. Ajinomoto's near-monopoly on the dielectric film, traced in the substrate bottleneck, is the archetype. EDA has no such ceiling. Licensed software copies at a cost that approaches zero, so a capacity bottleneck never forms. When customers raise capex, the contracted amount does not spike that quarter; renewals and new-node adoption stack into gentle compounding. The wave of demand does pass through EDA along the order of propagation, but with no capacity bottleneck it cannot become a share-price spike. The propagation arrives only where it rides a physical capacity bottleneck.
Within the same category, only the one holding the bottleneck rises
This mismatch is not a software-versus-hardware split. Arteris, the on-chip interconnect IP, sits in the same software-and-IP category as EDA, yet it rose +181.2% in 2026. Its royalties accrue per unit shipped, so it rides the physical meter of chips actually printed. On the other side are essential assets locked into long-term supply contracts that compounded gently: Teledyne in space IR sensors (+10% in 2025, +22.8% in 2026), and the industrial gases Linde (+3.1%, +23.7%) and Air Liquide (-4.8%, +10.7%). All three are essential chokepoints, yet their prices move only at the pace of contract renewals. What separates what rises from what stalls is whether the billing rides the volume of a physical bottleneck; whether the product is software is beside the point.
The more essential, the later
Siemens EDA is private and cannot be checked by share price, but with Calibre physical verification the de facto sign-off standard at TSMC, Samsung and Intel (85%+), it points to the same conclusion. Top of the essentiality axis, bottom of the cyclical-sensitivity axis. The more essential an asset is, the more stable it is, and the more stable it is, the later it moves. That settles the condition for propagation. The condition is not essentiality but a physical capacity bottleneck.
This is where the anatomy of the money ends. We now know what propagates along the chain and what does not, and why. But the demand we have treated as propagation's origin all along, the hyperscalers' capex, how solid is it? Circular AI financing finds that part of that demand comes from none other than the suppliers' own balance sheets.