Asymmetrical Bets
· Asymmetrical Bets
· April 08, 2026 at 21:14
· ⏱ 10 min read
| Read on Substack ↗
Summary
The article argues that optical component testing is a critical and underappreciated bottleneck in the AI infrastructure buildout, with testing TAM growing from $2.7B in 2025 to $21.7B by 2030 due to the shift to co-packaged optics. Hyperscalers are tightening testing protocols to avoid catastrophic cluster failures, making testing a structural beneficiary of the $400B/year CapEx cycle.
•Hyperscaler CapEx (Microsoft, Meta, Google) is $400B per year, driving a $73B optical networking market by 2030 at 39% CAGR.
•Testing is currently 19% of the optical market, projected to grow to 30% by 2030, driven by co-packaged optics (CPO).
•Testing requirements are growing 2x-3x the rate of optical port growth because CPO eliminates rework and adds test points per device.
•The testing TAM is estimated at $2.7B in 2025 on a $14B optical TAM, expanding to $21.7B on a $73B optical TAM by 2030.
•A single failed optical component in a 100,000 GPU cluster can cause millions in downtime costs, incentivizing rigorous testing.
•Three key testing layers: wafer-level burn-in (WLBI), module/transceiver test, and field OTDR fiber certification.
Article names Broadcom's AI networking ASICs as central to the CPO wave. Broadcom's custom switch silicon and Tomahawk/Jericho families are key enablers of the 1.6T interconnects that require the test
Article names Broadcom's AI networking ASICs as central to the CPO wave. Broadcom's custom switch silicon and Tomahawk/Jericho families are key enablers of the 1.6T interconnects that require the testing described.
Risk: Cyclical semiconductor demand; competitive pressure from Marvell and other ASIC vendors.
Article explicitly states NVIDIA's next-generation switches use co-packaged optics (CPO) architecture, which drives testing intensity. As the leading GPU and networking supplier for AI clusters, NVIDI
Article explicitly states NVIDIA's next-generation switches use co-packaged optics (CPO) architecture, which drives testing intensity. As the leading GPU and networking supplier for AI clusters, NVIDIA benefits from the infrastructure buildout and the need for reliable optical links.
Risk: CPO adoption may face technical hurdles or delays; NVIDIA's networking revenue is still a small portion of total.
This newsletter, published April 08, 2026,
features Asymmetrical Bets
discussing AVGO, NVDA.
2 trade ideas extracted by AI with direction and confidence scoring.