The year 2026 marks the watershed transition of AI data center optical interconnects from adoption to mass production. The actual volume driver is not scale-out CPO switches, but GPU scale-up networks. Competition is also elevating from pure PIC foundries to integrated optical engine platforms that combine PICs, EICs, packaging, lasers, and optical coupling. Success in mass production will depend on yield, field-serviceable connectors, InP/CW laser supply, and strategic ecosystem positioning.
HBM 2027 supply talks remain unresolved as suppliers push for steep price hikes; NVIDIA's SOCAMM adjustment reflects supply constraints, not demand reduction.
AI demand and proprietary chip adoption drive upward revisions in global server shipments. Despite supply bottlenecks, memory face severe shortages and soaring prices. AMD and alternative architectures gain market share, putting pressure on the traditional market leader.
Beyond full-rack solutions, NVIDIA is also actively promoting its standalone Vera CPU to capture the AI inference market. However, given the LPDRAM supply bottleneck, NVIDIA has decided to reduce the memory module capacity on its next-generation platforms to ensure shipment volumes align with market share targets. As AI applications continue to expand, the AI server ecosystem has the potential to become the single largest outlet for global LPDRAM, surpassing smartphone applications.
Driven by agentic AI applications, the server DRAM market experienced explosive growth. The need for models to handle complex computations and offload long contexts has led to a severe shortage of key modules. Despite hardware supply bottlenecks, cloud service providers continue aggressive procurement, depleting manufacturers' inventories to critical lows. Contract prices are surging, with this under-supply and upward price trend expected to persist long-term.
TrendForce raised its 2026 global server shipment forecast, driven by surging AI server demand from CSPs and enterprises, with momentum expected through 2027.
HBM remains irreplaceable in AI accelerators. HBM4 ramps, tight supply lifts prices, and ASIC demand diversifies the supplier base.
Server DRAM contract prices continue to rise due to supply deficits and depleted manufacturer inventories. Although premium pricing on high-capacity modules and shifts in processor configurations have led cloud service providers to recalibrate their demand toward low-to-mid capacity options—temporarily halting the growth momentum of large-capacity shipments in the second half—overall server demand remains robust, keeping the market under-supplied.
The explosion in global AI service demand has triggered a severe supply-demand imbalance in the enterprise SSD market, driving quarterly revenue and contract prices to historic highs. Major suppliers are aggressively adjusting their capacity, shifting resources from consumer-grade to enterprise-grade products, and accelerating the development of next-generation, high-speed storage solutions. Consequently, SSDs are gradually transitioning into a computing-assistive role, and the market is expected to sustain a long-term upward trend.
As AI server platforms advance, power architectures evolve into multi-track systems, and cooling accelerates toward full liquid solutions. Delta and LiteOn capture cloud service provider and new platform demands with diverse, high-power architectures, aggressively expanding localized capacity. AVC and Sunon accelerate into system-level liquid cooling integration, driving growth amid computing upgrades.
Research Areas:
Industry News
