AI Physical Layer

The Physical Layer of AI

What is the physical layer? It's those companies that provide power, cooling, connectivity, and packaging for #AI chips. No matter how powerful Nvidia is, it still relies on these 'shovel sellers' to support the entire ecosystem.

Let's look at the data:
• AI server power consumption has skyrocketed from the traditional 5-10kW to 100kW+
• The liquid cooling market is expected to reach $30 billion by 2030, with a CAGR exceeding 40%
• HBM memory demand is growing 150% annually, with supply unable to meet demand
• Optical modules are upgrading from 800G to 1.6T, with the entire industry chain racing forward

These demands cannot be solved by Nvidia alone. They are the result of dozens of companies across the entire supply chain working together.

👇 Today, let's break down the key 'shovel seller' leading companies in the AI physical layer. Understanding these will make investing in #AI much less confusing! 🧐

High-power, narrow-linewidth lasers Coherent ( #COHR)
Narrow-linewidth lasers are the 'heart' of 1.6T optical modules and silicon photonics technology. Coherent is the global leader in indium phosphide lasers, directly benefiting from the AI optical connectivity upgrade.

Glass core substrates Corning (#GLW)
Glass substrates are the new savior of Moore's Law. With decades of dominance in precision glass, Corning is the preferred R&D partner for chip giants like Intel.

Silicon photonics Marvell (#MRVL)
Although Nvidia also works on this, Marvell's silicon photonics platform and DSP chips are more adaptable in the open market, serving as the core hub for large-scale chip-to-chip optical interconnects.

High-power-density power modules Vicor (#VICR)
Vicor's core is its modular power architecture (FPA), which can embed power directly onto the PCB, solving the challenges of AI chip transient current and power density.

Multilayer ceramic capacitors Murata (#MRAAY)
Although it's a Japanese stock, it has ADRs in the US market. Murata holds about 30% of the global MLCC market share. AI servers require several times more high-capacity, high-reliability MLCCs than traditional servers.

Power distribution systems & UPS Eaton (#ETN)
Eaton is a key player in data center power distribution. As AI chip power consumption surges, demand for its intelligent power distribution units (ePDUs), UPS uninterruptible power supplies, and power management systems is exploding.

High-bandwidth memory Micron (#MU)
One of the purest memory giants in the US stock market. Its HBM3e products are highly competitive in energy efficiency and are a major supplier for NVIDIA and other AI chips besides SK Hynix.

Advanced packaging equipment Applied Materials (#AMAT)
It is the 'toolmaker'. Whether it's 2.5D or 3D packaging, AMAT has the most comprehensive product line for required equipment like thin-film deposition and chemical mechanical polishing.

Data center infrastructure Vertiv (#VRT)
The absolute leader in data center infrastructure. It not only handles cooling; its high-power busways and integrated power racks (Power Rack) are the physical skeleton of AI cabinets. In the trend from air cooling to liquid cooling (cold plate, immersion), Vertiv is deeply tied to NVIDIA's Blackwell architecture, with strong growth certainty.

Hybrid bonding equipment BESI (#BESIY)
Hybrid bonding is the most advanced technology for achieving 3D stacking (e.g., connecting HBM and logic chips). BESI and Tokyo Electron are the main players in this process equipment.

Grid transformers / Gas turbines GE Vernova (#GEV)
An energy giant spun off from General Electric. The ultimate solution for data center power shortages is gas turbine power supplementation and grid transformer upgrades. It is the cornerstone of US domestic power.

ABF substrates & high-end PCBs Ibiden (#IBIDY) (Japanese stock/US ADR)
Although Unimicron is also a giant, Ibiden has consistently been at the top in terms of yield and share for high-end server and AI chip-level ABF substrates. It specializes in producing the ultra-multilayer and high-density interconnect (HDI) PCBs required for AI servers. Financial reports show its data center business segment is benefiting from the volume ramp of next-gen architectures like NVIDIA Blackwell, with a significant increase in order share.

High-end PCB manufacturing TTM Technologies(#TTMI)
TTM is one of the important high-end PCB manufacturers in the US stock market, especially in the complex, high-volume AI infrastructure field, with mature supply chains and technical expertise.

In summary, the money in AI isn't only for Nvidia to make. Across the entire supply chain, from power and cooling to connectivity and packaging, every link holds opportunities.

These 'shovel seller' companies share several common characteristics:
• High technical barriers to entry
• Strong customer stickiness, with huge costs to switch suppliers
• High demand certainty, as AI computing power expansion is a long-term trend
• Relatively reasonable valuations, unlike the crowded trade in Nvidia

Of course, investing in these companies also carries risks:
• Some have small market caps and poor liquidity
• Japanese stock ADRs carry currency risk
• Supply chains are concentrated in Asia, and geopolitical risks cannot be ignored
• Fast technology iteration; today's leader may fall behind tomorrow

Don't put all your eggs in one basket. Nvidia can be part of the portfolio, but these companies in the physical layer are also worth diversifying into. Especially those undervalued by the market but with genuinely strong technology.