This group sells the consumable raw inputs a chip factory (a "fab") burns through every single day: the polished silicon discs chips are printed on (silicon wafers), the light-sensitive chemicals used to draw the circuit patterns (photoresist), the ultra-pure gases used to lay down and etch away thin layers (specialty gases), plus polishing slurries, filters and gas-handling hardware. Unlike the giant tool-makers who sell a fab a large machine once, these companies sell inputs that get consumed and re-ordered repeatedly — every wafer that runs through a fab eats materials. The named US-listed players are Entegris (ENTG — advanced materials, filtration and gas delivery), Linde (LIN), Air Products (APD) and Air Liquide (AIQUY — the three big industrial-gas companies that pipe gases straight into fabs). Most leading-edge wafer and photoresist supply sits with Japanese firms (Shin-Etsu, SUMCO, JSR) that are thinly traded on US exchanges.
Demand for these materials rises with every extra wafer the world starts, and the scan documents AI pushing wafer starts up — it is a derivative, volume-driven pull rather than the per-unit pricing seen in GPUs (graphics processing units, the main AI compute chips) or HBM (high-bandwidth memory, the stacked memory paired with GPUs). Supply is limited in a few specific niches (ultra-pure 300mm wafers, EUV-grade photoresist, certain electronic gases) where only two-to-four qualified suppliers exist and a new customer must spend a year or more "qualifying" a vendor (testing and certifying that the vendor's material works in their exact process); outside those niches the materials are closer to a steady industrial business. The US-listed industrial-gas names (LIN, APD, AIQUY) generate steady, contracted cash; the one US-listed pure-ish materials play (ENTG) is far smaller and more cyclical. The price the market puts on each is laid out in the price section; the reader judges it.
There are really four product families bundled here, each sold in a different unit:
How the cash actually shows up: most of this is a razor-and-blades / consumables model (sell the equipment or win the slot once, then earn recurring revenue on the inputs forever). The supplier wins a fab "spec-in" (gets its material qualified for a customer's process), then earns recurring revenue for years as that fab runs wafers. For the gas companies specifically, the on-site plant + take-or-pay contract turns it into something close to an annuity — predictable cash in, contracted for 10-15 years, with the customer carrying the volume risk. That contracted, recurring shape is the factual reason the market tends to value these like steady industrials/utilities rather than like boom-bust chip makers.
What drives the unit (wafers consumed): materials demand is almost mechanical — it tracks total wafer starts (how many wafers fabs feed in) and, increasingly, process complexity (each new node — generation of chip manufacturing — adds more layers, more lithography steps, more deposition/etch cycles, so each wafer consumes more material than it did a node ago). Both numbers are rising, per the scan.
The AGI lens: given the AI/AGI compute build-out is happening, the demand pull is real but second-order. AI accelerators (GPUs), HBM memory, and the advanced packaging that stitches them together all consume leading-edge wafers and far more deposition/etch gas and CMP slurry per finished chip than older logic did. HBM in particular stacks many DRAM (dynamic random-access memory) dies, multiplying the material and gas consumed per gigabyte. So the AI build-out raises both the number of advanced wafers and the material intensity per wafer. Physical-AI / robotics, on-device inference and the broader electronics rebuild add a second leg of volume on mature nodes. The factual caveat: materials capture this as volume growth, not as the price increases seen upstream in GPUs and HBM — a fab pays roughly the same per litre of gas whether the chip on the wafer sells for $50 or $50,000.
Who the buyers are: a short, concentrated list — TSMC, Samsung, Intel, SK Hynix, Micron, and the big memory and foundry fabs. A handful of customers drive most demand, which gives those customers pricing leverage but also makes supplier revenue sticky once a supplier is qualified.
Current vs. forward (label: forecast): current worldwide materials spend is on the order of est. $60-70B/yr split roughly between wafers, gases, and the chemistry/slurry/filtration bucket. Forward demand is widely forecast to grow mid-single-digits in volume over the coming years est., with the AI-exposed leading-edge and advanced-packaging slices growing faster than the mature-node average. These are forecasts, not contracted backlog; the contracted piece sits mostly in the gas companies' multi-year take-or-pay agreements.
✓ VERIFIED — the following figures were confirmed from primary sources after initial publication:
Supply splits into "constrained niches" and "comfortable commodity," and the difference is the whole story.
Market-share structure / who controls supply: wafers = Japanese duopoly; advanced photoresist = a few mostly-Japanese chemists; bulk + specialty gases = a three-firm oligopoly (a market controlled by a few sellers — Linde / Air Liquide / Air Products) that between them are estimated to control the large majority of global industrial-gas supply est. The recurring theme: few qualified suppliers, high switching cost, lumpy capacity additions.
Source: company market caps from /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/ai-stock-universe.html; supplier structure and qualification dynamics from the 500-stocks scan plus general knowledge (not live-verified).
Putting the two sides together: on the facts available, this group is balanced-to-slightly-short in the niches and balanced elsewhere — not a runaway shortage like leading-edge GPU/HBM, and not a glut. Demand grows with wafer starts and rising material-intensity per wafer; supply is concentrated and slow to add, which keeps the constrained niches tight, but the majors generally build to contracted demand, which limits price spikes of the kind seen upstream.
| Sub-product | Short or long? | Evidence | When could it flip to oversupply? |
|---|---|---|---|
| 300mm silicon wafers | Balanced (was short) | Tightness normalized; duopoly adds in big lumps est. | If duopoly over-builds into a digestion year |
| EUV / advanced photoresist | Tight (know-how limited) | Few qualified makers; long qualification est. | Slow to flip — protected by chemistry barrier |
| Bulk + specialty gases | Balanced (contracted) | Built to take-or-pay demand; oligopoly discipline contracted demand | Rarely — supply follows signed contracts |
| Gas delivery / filtration (ENTG) | Balanced, sticky | Spec-in lock-in; recurring consumables est. | Cyclical with fab utilization, not over-supply |
The factual read: the contracted gas business is the most insulated (the customer carries the volume risk via take-or-pay), the photoresist/know-how niche is the hardest to flood (the barrier is chemistry expertise and qualification time, not capacity), and wafers are the piece most able to swing to oversupply because the duopoly adds capacity in chunks. Across the whole group, the pattern the sources describe is steady absorption rather than a sustained sold-out scarcity premium.
| Company (ticker) | What it makes | Exposure to THIS product | Rough size (mkt cap) | Position |
|---|---|---|---|---|
| Entegris (ENTG) | Advanced materials, filtration, gas-delivery hardware, CMP slurries | Near pure-play on fab materials/consumables | ~$23.7B | Broadest US-listed materials/consumables franchise; spec-in lock-in, recurring revenue |
| Linde (LIN) | Industrial + electronic gases; on-site fab supply | Diversified; semis a slice est. ~10-15% of sales | ~$232.0B | Largest industrial-gas firm globally; on-site plants, take-or-pay contracts |
| Air Liquide (AIQUY) | Industrial gases; semi fab supply | Diversified; semis a slice est. | ~$122.4B | French gas major; electronics-gas franchise, esp. in Asia/Europe fabs |
| Air Products (APD) | Industrial gases for fabs + hydrogen | Diversified; semis a slice, plus large hydrogen projects est. | ~$66.9B | #3 gas major; electronics gas plus capital committed to large hydrogen projects |
| Nippon Sanso (NIPNF) | Industrial / electronic gases | Gas major; competes for fab supply | ~$36.3B | Japanese gas leader; presence in Asian fabs (US-listing thin) |
| Shin-Etsu (SHECY) | #1 silicon wafers + photoresist | Core wafer/resist supplier (foreign-listed) | ~$87.2B | Largest wafer maker by share est.; Japanese listing, thin US trading |
| SUMCO (SUOPY) | #2 silicon wafer maker | Pure-play wafers (foreign-listed) | ~$6.3B | Other half of the 300mm wafer duopoly (Japanese listing) |
Source: company list and qualitative notes from /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/02-semiconductors.html (Sector 11, "Semiconductor Materials"); market caps from /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/ai-stock-universe.html. Revenue-mix percentages and rank/share labels are est. general knowledge, not live-verified.
Translating market values into plain arithmetic — "how much market value am I paying per $1 of this year's sales" — using the grounded market caps and approximate revenue scales (the revenue figures here are est. general knowledge, not live-verified, so the ratios are approximate):
| Name | Mkt cap | Approx. annual revenue est. | ≈ $ of market value per $1 of revenue | Cash shape (factual) |
|---|---|---|---|---|
| Linde (LIN) | ~$232B | ~$33B | ~7x | High-capex; cash from long-term contracted on-site supply |
| Air Liquide (AIQUY) | ~$122B | ~$30B | ~4x | High-capex; contracted gas supply, profile similar to LIN |
| Air Products (APD) | ~$67B | ~$12-13B | ~5x | High-capex; part of near-term cash committed to hydrogen build-out |
| Entegris (ENTG) | ~$23.7B | ~$3-3.5B | ~7x | Lighter assets than gas majors; consumables cash, more cyclical with the chip cycle |
Money-in / money-out shape of the group: the gas majors (LIN, AIQUY, APD) are high-capex businesses — they put large amounts of cash into building plants and pipelines — and those assets then generate contracted owner cash (cash left for shareholders after the business funds itself) over multi-year contracts. APD differs in that a portion of its capital is currently committed to hydrogen projects, so less of its cash reaches the owner in the near term. ENTG is more capital-light and consumables-driven, generating cash on smaller revenue, and it is more exposed to the chip cycle (when fabs slow, consumable orders dip). One neutral arithmetic fact: at ~4-7x sales these names carry lower price-to-sales ratios than the sold-out parts of the AI chain (GPUs, HBM, leading-edge foundry, which the scan rates "High" and which trade at higher multiples). What you are buying is steady, contracted, derivative exposure to wafer-start growth; the price-per-$1-of-sales is in the table above and the reader judges whether it is worth paying.
Source: market caps from the provided universe file; revenue figures and the resulting multiples are approximate general-knowledge est. and not live-verified — verify against the latest 10-K/annual report and a live quote before relying on them.
Where a company-level deep-dive would add the most factual detail, by question:
This is a pointer to where the facts are most concentrated for each question (pure-play, contracted, or conglomerate-with-a-small-slice), not a recommendation.