Specialty Gases: Why the AI Boom Didn't Re-Rate Them
LIN, Air Liquide, APD, and Nippon Sanso are essential inputs to every leading-edge fab on earth. The AI revolution sent NVDA up 600%+ over three years. These four stocks did roughly nothing. Here is the honest reason why.
The price action — three years of "AI everything"
| Ticker | Name | Mkt Cap | 1Y | 2Y | 3Y |
|---|---|---|---|---|---|
| LIN | Linde plc | $232B | +11% | +21% | +44% |
| AIQUY | Air Liquide ADR | $122B | −1% | +5% | +22% |
| APD | Air Products | $67B | +8% | +22% | +15% |
| NIPNF | Nippon Sanso (ADR) | — | +9% | −62% | −42% |
| NVDA | (for reference) | — | +63% | +133% | +608% |
| SPY | (S&P 500) | — | +25% | +42% | +82% |
LIN tracked the index. APD and AIQUY underperformed it. NIPNF was a disaster. None of them got an "AI bid."
Why — five reasons, ranked by how much they explain the flatness
1. Semi gases are a small slice of these businesses
This is the single biggest reason. Linde, Air Liquide, and Air Products are industrial gas companies first, semiconductor specialty gas suppliers second.
- Linde: Electronics (which includes semis) is ~10% of revenue. ~70% is industrial customers — steelmaking, refining, chemicals, hospitals, food & bev.
- Air Liquide: Electronics ~9% of group revenue. Large Industries (refining, chemicals, steel) is the biggest segment.
- Air Products: Electronics ~12% of revenue. They pushed all-in on hydrogen, not semis.
- Nippon Sanso (Taiyo Nippon Sanso): Electronics is a bigger share (~25-30%), but they're listed in Tokyo, illiquid as a US ADR, and the yen has been weak.
If electronics doubled tomorrow, Linde's total revenue grows ~10%. Doubling is not what's happening — AI is adding incrementally to a fab buildout that was already running.
2. The contracts are take-or-pay, not spot — that's a feature and a curse
Semi specialty gas (silane, ammonia, nitrogen trifluoride, tungsten hexafluoride, etc.) is sold under 10–20 year on-site supply contracts. The gas company builds a plant next to the fab and gets paid a fixed monthly fee plus volume. This is great for stability and terrible for AI upside:
- When NVDA's revenue 10x'd, the gas suppliers' revenue from TSMC's existing Fab 18 did not 10x. They got paid roughly the same fixed fee.
- Upside requires new fabs to come online and ramp. TSMC Arizona, Samsung Texas, Rapidus Hokkaido — these are slow, multi-year buildouts.
- Even when a new fab signs, the gas company has to spend capex up-front for the on-site plant. ROIC is steady — 10-15% — not the 60% gross margin re-rating story the market gives to NVDA.
3. AI didn't materially change wafer starts (yet)
Gas consumption scales with wafer starts, not with the price of the chips on those wafers. NVDA selling H100s for $30k each instead of $10k doesn't consume more silane per wafer. SEMI's data shows global 300mm wafer fab capacity grew ~6-8% per year through the AI boom — basically the same trend as before. The dollar value of AI chips exploded; the physical throughput grew at the same boring pace.
Where AI does drive gas demand is HBM (more layers = more deposition steps) and advanced logic (more EUV layers, more etch cycles). That's real, but it's a 15–25% lift in gas-per-wafer at the bleeding edge — not a step-change in revenue across the supplier base.
4. Company-specific damage offset whatever AI tailwind existed
- Air Products (APD): Bet the company on green/blue hydrogen mega-projects (NEOM, Louisiana, Alberta). Costs ran over, customer commitments slipped, the activist Mantle Ridge fight ended with the CEO out in early 2025 and the new team unwinding the hydrogen pivot. Whatever AI bid there could have been was buried by hydrogen write-downs.
- Air Liquide: European listing, weak euro narrative, decarbonization capex weighing on free cash flow. Solid execution but no AI multiple expansion — European industrials don't trade on US AI narratives.
- Nippon Sanso: Strong electronics franchise (Matheson is theirs in the US), but the US ADR is essentially untraded, the parent trades in Tokyo, and the M&A integration of Taiyo Nippon Sanso's European business has been bumpy.
- Linde: The cleanest story. Boring execution, dividend grower, modest electronics tailwind. The market rewarded it with… market-average returns.
5. The oligopoly is too big to re-rate on AI
Industrial gas is a global ~$120B revenue oligopoly with four players. Their combined market cap is ~$450B. For any one of them to double on AI, the AI-attributable earnings would need to grow by an amount comparable to the entire company. Even the most aggressive case — every new advanced fab in the world signing with one supplier — doesn't move the needle that much.
Contrast with NBIS or CRWV: small companies where AI is the entire business. A 5x is mathematically conceivable. For Linde to 5x, you'd need a different planet.
Is the market wrong?
Probably not. The market is correctly pricing these as steady-eddy compounders with semi exposure as a kicker, not as AI plays. The case for owning them is the same as it was in 2019: oligopoly, regulated-utility economics, dividend growth, 8-10% IRR over a cycle. That case doesn't need AI to work, and AI doesn't make it dramatically better.
What would change my mind
- Capacity contention: If TSMC and Samsung start bidding for specialty gas supply because the gas majors can't build plants fast enough, pricing power shifts. No sign of that today — gas capex tracks fab capex with predictable lead times.
- A 2x in electronics segment growth rate: Currently mid-single-digits in line with WFE. If it sustainably runs at 15-20% for 2+ years, the segment becomes large enough to move the parent multiple.
- A pure-play spinout: If one of the majors carves out its electronics business as a standalone listed entity, the market would value it like an AI infra name rather than burying it inside a hydrogen-and-steel company. No one is doing this yet.
Verdict: Pass (for now)
The flat price is rational. These are good businesses being correctly priced as good businesses. They are not hidden AI plays — they're industrial conglomerates with a small, high-quality semi segment. There's no asymmetric setup here today: floor is solid (utility-like cash flows), but ceiling is also low (oligopoly is too big, contracts are too rigid).
The interesting question is not "why didn't they go up" — it's "what would have to happen for an AI-driven re-rating?" The answer is a structural change in either capacity tightness or business structure (spinout). Worth a tickle if APD's new management surprises with an electronics-focused strategy, or if any of them spins out the semi segment. Until then, the watch is passive.