AI Infrastructure

AI infrastructure is the picks-and-shovels layer of the artificial-intelligence build-out: the chips, networking, optical interconnects, and power that hyperscalers buy to train and run models. This pillar maps the theme, starting with optical networking, the part that moves data between and inside AI clusters at the speed of light.

Key takeaways
  • AI infrastructure is the picks-and-shovels of the AI build-out: the compute, networking, optical interconnects, and power that hyperscalers buy to train and run models.
  • Optical networking moves the data between and inside AI clusters; demand is scaling from 800G to 1.6T transceivers and toward co-packaged optics.
  • The signal that interconnect (not just compute) gates AI scaling: in March 2026 NVIDIA committed about $4 billion to the optical layer, investing $2 billion each in Coherent and Lumentum.
  • The build-out is capital-driven and supply-constrained, which can mean both fast growth and sharp cyclicality.
  • Representative U.S.-listed names span optical networking (Ciena, Coherent, Lumentum), with the Akros AI Optical index tracked by the KODEX US AI Optical Communication Network ETF.

What is AI infrastructure?

Behind every chatbot sits a vast physical plant: racks of accelerators, the switches and optics that wire them together, and the power and cooling that keep them alive. AI infrastructure investing owns that plant rather than the model makers. The bet is simpler than picking the winning model: whoever wins, they all need more compute, more bandwidth, and more power.

What are the layers of AI infrastructure?

The theme breaks into a few distinct layers, each with its own suppliers:

  • Compute: the accelerators (GPUs and custom silicon) and the high-bandwidth memory that feed them. The custom-silicon side runs through hyperscaler supply chains, covered via the Google (Alphabet) AI value chain concept.
  • Networking & optical interconnect: the transceivers, lasers, switches, and data-center interconnect that move data between thousands of chips. This is the layer ETFpedia covers in depth, via the optical networking concept.
  • Power & cooling: the electrical and thermal plant, increasingly the binding constraint as clusters grow. The power side, including the nuclear revival, is covered via the AI electricity & nuclear SMRs concept.

The bottleneck keeps moving between these layers, which is why the most investable opportunity shifts over time. Right now, a major constraint is moving data fast enough between accelerators, which is why optical interconnect demand has surged.

Why does it matter now?

Hyperscaler capital spending has stepped up sharply to build AI clusters. The AI optical-transceiver market alone is set to jump from about $16.5 billion in 2025 to $26 billion in 2026, roughly 57% growth (TrendForce), and the silicon-photonics interconnect market is forecast to grow from about $3.69 billion in 2025 to $33.8 billion by 2034 (Global Market Insights).

AI creates a new wave in demand for optical transceivers and accelerates LPO/CPO adoption.

— LightCounting

Why is NVIDIA funding the optical layer?

The clearest signal that interconnect, not just compute, gates AI scaling came in March 2026, when NVIDIA committed about $4 billion to the optical layer, investing $2 billion each in Coherent and Lumentum (NVIDIA IR). The chip leader is helping fund the very suppliers that build the optics around its GPUs. The opportunity behind that bet is large: in May 2026, Bank of America raised its 2030 AI-data-center systems TAM to roughly $1.7 trillion (24/7 Wall St.).

Explore this theme

Concepts

Companies

Key terms

How can you invest in it?

The optical networking concept covers the U.S.-listed leaders in this layer. The Akros U.S. AI Optical Communication and Network Infrastructure Index is tracked by the KODEX US AI Optical Communication Network ETF, a Korea-listed fund that rose about 34% in its first month on this demand (Samsung KODEX).

What are the risks?

This is a capital-cycle theme. Spending that ramps fast can also pause, and many suppliers depend on a handful of hyperscalers and on Nvidia’s roadmap, so revenue can be lumpy. New approaches like co-packaged optics can also reshuffle which suppliers win the next generation.

Frequently asked questions

What is AI infrastructure investing?

AI infrastructure investing focuses on the companies that supply the physical build-out behind artificial intelligence: data-center compute, networking, optical interconnects, and power. It is a picks-and-shovels approach, owning the suppliers rather than the model makers.

Which parts of AI infrastructure are investable?

The main layers are compute (accelerators and memory), networking and optical interconnects that move data between chips and servers, and the power and cooling that keep clusters running. Optical networking is the layer covered here in depth.

How big is the AI data-center opportunity?

Estimates vary widely, but in May 2026 Bank of America raised its 2030 AI-data-center systems TAM to roughly $1.7 trillion. Within that, the AI optical-transceiver market alone is projected to grow from about $16.5 billion in 2025 to $26 billion in 2026.

Is there an ETF for AI optical networking?

The Akros U.S. AI Optical Communication and Network Infrastructure Index is tracked by the KODEX US AI Optical Communication Network ETF, a Korea-listed fund. Always check fees, holdings, and risk before investing.

Sources & references

  1. AI creates a new wave in demand for optical transceivers and accelerates LPO/CPO adoption · LightCounting, 2025-01-15
  2. Global AI Optical Transceiver Market to Reach US$26 Billion in 2026 · TrendForce, 2026-04-20
  3. NVIDIA strategic partnerships and $2 billion investments in Coherent and Lumentum · NVIDIA Corporation, 2026-03-02
  4. BofA hikes Coherent target on a $1.7 trillion AI data-center forecast · 24/7 Wall St., 2026-05-13
  5. Silicon Photonics Market size and forecast · Global Market Insights, 2026-01-01