$AAOI

Summary

The analyst warns that Applied Optoelectronics (AAOI) faces major structural risk due to Amazon's RNG and OpenAI's MRC protocols reducing demand for optical transceivers, which will sharply decelerate AAOI's revenue growth.

Reasoning

New network architectures by Amazon and OpenAI reduce the need for active networking equipment and transceivers by 40-50%, undermining the long-term total addressable market expansion for optical infrastructure stocks like AAOI, which relies heavily on these transceivers for revenue growth.

*B. RILEY SAYS AMAZON'S RNG & OPENAI'S MRC PROTOCOLS POSE MAJOR STRUCTURAL RISK FOR OPTICAL INFRASTRUCTURE / "PHOTONICS" STOCKS, NAMELY APPLIED OPTOELECTRONICS $AAOI Full comments: "Following the OFC conference in March 2025, the AI data center boom triggered an unprecedented optical supercycle. As demonstrated by the parabolic re-rating of optical infrastructure stocks such as Lumentum ($LITE – Buy, $1,142 PT ) and Coherent ($COHR – Neutral, $309 PT ), the prevailing investment thesis assumed an aggressive, compounding multiplier: scaling AI clusters meant building deeper, multi-tiered networks (3-tier or 4-tier Clos topologies), exponentially driving up transceiver demand per unit of compute. This thesis, however, is hitting a structural wall. Just a year after the 2025 explosion, the simultaneous market introduction of Amazon’s ($AMZN, NR) Resilient Network Graphs (RNG) and OpenAI’s Multipath Reliable Connection (MRC) protocol marks a major pivot toward network flattening across both training and inference workloads. Microsoft ($MSFT, NR) and Oracle ($ORCL, NR) are actively integrating the open-source MRC protocol into their frontier AI training supercomputing clusters—such as ORCL’s massive GB200 Abilene site—to execute sender-controlled packet spraying and bypass physical network bottlenecks. Concurrently, AMZN is deploying its quasi-random RNG architecture across its general-purpose cloud fabric. While RNG is isolated from back-end training clusters, it serves as the primary backbone for scale-out generative AI inference, data ingestion, and retrieval-augmented generation (RAG). Crucially, as mature AI models transition from training to mass inference, hyperscaler capex is shifting heavily in favor of these front-end, scale-out environments. Because RNG replaces traditional, active multi-tiered aggregation switches with passive optical ShuffleBoxes, it reduces active networking equipment by over 60%. Consequently, as AI workloads migrate to inference, this architectural deflation loops directly into the broader optical market, stripping out switch-to-switch optical links and dismantling the legacy transceiver-to-compute multiplier. For the transceiver market, this dual-front flattening represents a severe structural headwind that will sharply decelerate the sector's long-term TAM expansion, since these technologies reduce structural transceiver counts by 40% to 50%. We believe Applied Optoelectronics ($AAOI – Neutral, $129 PT) will be especially vulnerable to this development. AMZN and ORCL—the primary network provider to OpenAI—are expected to be the anchor customers for the 800G/1.6T transceivers forecast to drive AAOI's quarterly revenue to ~$1B by 2H27. As ORCL deflates training-cluster transceiver ratios via MRC, and AMZN strips active transceiver requirements from inference rollouts via RNG, AAOI's core revenue growth engines face massive structural risk."

Return by horizon