Atlas Prime

Applied sells the machines and process steps that fabs use to build chips — deposition, etch, inspection, and more. When the industry adds leading-edge capacity or new packaging flows, WFE (wafer fab equipment) spend is usually involved.

The chokepoint read is not one secret widget but **share and timing in process steps** that every new node or advanced packaging line needs. Demand inflects with capex cycles; estimates move when bookings and backlog language inflect.

Use the full ticker page for numbers; here the point is **where they sit**: upstream of the wafer, selling to foundries and memory makers who must keep spending to stay on the roadmap.

AXT supplies substrate wafers used in compound semiconductors — the material layer before laser or RF devices are built. When indium phosphide or gallium arsenide demand runs hot, substrate vendors are an early **upstream** tell.

Smaller cap and more volatile than integrated laser or module vendors. The thesis hinge is often **tight supply** in substrates when downstream photonics or wireless ramps.

Good for mapping **who feeds whom** in the chain; pair with names like LITE or COHR to see the full stack from wafer to module.

Ciena builds optical and routing systems for carriers and increasingly for cloud-scale networks. It is **systems-level** exposure — different margin and cycle shape than a laser component vendor.

When hyperscale and telecom both push coherent speeds (800G, 1.6T class buildouts), system vendors participate, but the stock often trades on **order cadence and mix**, not a single component shortage.

Useful as a **second opinion** in the same theme bucket as datacom optics: if CIEN and component peers are all guiding up, the theme is broad; if only one leg moves, dig for idiosyncratic drivers.

Coherent spans lasers, components, and systems for industrial and communications markets. In AI-infrastructure narratives, investors often focus on **datacom photonics** — the parts that sit inside high-speed optical links.

Chokepoint logic applies where specs are tight and **qualification cycles** are long: customers cannot swap suppliers quickly. That supports pricing power until new capacity or dual-sourcing appears.

Bear case to keep in view: mix shift, pricing in commoditized lines, and customer concentration. Conviction should tie to **which segment** is growing and whether leadership is sustained across quarters.

Entegris sells specialty materials and contamination control for semiconductor manufacturing — chemicals, filters, handling. Fabs treat yield as sacred; **purity and process materials** are hard to rip out once qualified.

This is classic **razor-and-blade / consumables** exposure to wafer starts rather than a single gadget cycle. Demand grows with fab activity and complexity (new nodes, new packaging).

Less flashy than optics headlines, but often **sticky** revenue. Watch for inventory corrections in the broader semi supply chain that hit materials before they hit equipment.

Corning is best known for glass science; in connectivity, it supplies optical fiber and related products that carry light over distance in networks. It is **infrastructure** exposure more than a laser chip play.

The read-through for AI clusters is **more fiber miles and higher-spec cable** as data centers scale. Margins and mix depend on consumer electronics segments too, so separate **optical inflection** from TV or mobile cycles.

Use GLW to anchor “**photons in the cable**” while LITE/COHR cover “**photons in the module**.”

Lumentum makes optical components and modules used in high-speed data center links. When bandwidth per rack rises, more optical ports and higher-speed optics tend to follow — that is the **datacom** demand engine.

Chokepoint framing: only a handful of suppliers can hit certain power, speed, and reliability specs at volume. **Capacity and yield** often cap how fast revenue can grow even when orders are hot.

Risk is always **mix and competition**: transceiver modules can see price pressure even when laser or component supply is tight. Match the story to **segment disclosure**, not a generic “AI optics” label.

Micron is a primary memory manufacturer — DRAM and NAND. The classic supercycle pattern is **supply discipline + demand surprise** (PC, server, AI-related memory content). Pricing and bit growth drive the PnL.

For thematic work, memory is the **commodity end** of the stack: huge leverage to ASP (average selling price) turns. Chokepoints show up as **allocation, lead times, and fab timing** rather than a single sole-source part.

Pair memory with equipment (AMAT, TER) and materials (ENTG) to see **capex → wafer → bit shipments** in one picture.

onsemi focuses on power management, silicon carbide, and image sensing for auto, industrial, and energy. **Power efficiency** matters everywhere data centers add GPUs — not just optics.

The chokepoint story here is often **SiC substrate and fab capacity** for high-voltage applications, plus design wins in electrification that compete for management attention with datacenter narratives.

Good cross-check when the theme is “**electrons**” (power delivery, conversion) versus “**photons**” (optical links). Both show up in large-scale AI infrastructure.

Teradyne builds automatic test equipment (ATE) — machines that verify chips before they ship. More complex silicon and advanced packaging usually mean **more test time** and higher equipment load, within the limits of the semi capex cycle.

Chokepoint is **capacity at test** and **share in key sockets** when fabs and OSATs expand. Revenue is cyclical: it moves with industry capex, not a smooth SaaS curve.

See the public **thesis** page on this site for a deeper peel-the-onion Q&A on how testers fit the AI build-out story.