Denser AI accelerators raise the value of efficient power delivery near processors, potentially making Navitas one of the most hidden AI winners in Nvidia's architecture. Its 24 kW board targets single-stage 800V-to-12V conversion at about 97% efficiency in Nvidia's 400 mm by 60 mm format. An 800V-to-6V design developed with Nvidia and Claros offer other paths into the power stack. This is a showcase, not a production order; shipment volumes, revenue and competition remain uncertain. x.com 

AI infrastructure bottlenecks could give eight stocks as much as 18x upside. Nvidia supplies frontier-model GPUs; Broadcom and Marvell supply custom silicon and networking; TSMC makes advanced chips; Micron supplies sold-out HBM; Vertiv provides power and cooling; Palantir monetizes enterprise AI; Alphabet spans TPUs, research, cloud and apps. The list tracks where AI infrastructure spending is flowing. x.com  x.com 

Broadcom's roughly 60% margin looks unsustainable because its custom-chip design role is not irreplaceable. Google's TPU v8i has already succeeded, and it is being handled by MediaTek, whose margin is at best in the 30% range. That undercuts the claim that customers need Broadcom's design capability to compete with Nvidia, especially when Broadcom's margin is not far from Nvidia's own. x.com  x.com  x.com  x.com  x.com  x.com 

NVIDIA's liquid-cooling work with Vertiv can deliver more compute from a fixed power budget. Liquid cooling yields roughly 22% more compute and more tokens from the same power on average, a benefit modeled in an April report. NVIDIA's scale, balance sheet and command of its supply chain make it the economic winner of the AI compute cycle, with Broadcom a close-ish second. x.com  x.com 

Google TPU adoption could broaden the AI hardware opportunity beyond Nvidia, with demand reaching custom ASICs, HBM, packaging, optics and testing. Per The Information, OpenAI reportedly plans to buy Google's TPUs and is hiring compiler engineers, but no purchase is confirmed. Broadcom, SK hynix, Samsung, Micron, TSMC, Coherent, Lumentum, FormFactor and Teradyne could benefit. x.com 

Alphabet and Broadcom own the 2027 window because their TPU optical-switching scale-up is the only proven alternative as Nvidia hits packaging and interconnect walls. Per SemiAnalysis, Nvidia's flagship Kyber NVL144 rack slipped over 12 months to 2028 and its NVL72x2 backup was cancelled, while the 2-die Rubin Ultra delivers about half the per-package performance and co-packaged optics won't come until Feynman. That breaks Nvidia's new-rack-every-year promise, the core of its moat premium. x.com 

Navitas's place in NVIDIA's 800 VDC ecosystem carries little pricing power because the silicon tier is crowded. NVIDIA names fourteen silicon suppliers, Navitas among them, and rent comes from scarcity; SiC and GaN also split the socket by stage. Per SemiAnalysis, rollout is phased, and the first stage is a retrofit into existing wiring with the smallest content per site. Unlike optical, no 800 VDC tier has disclosed dollars yet; a first order with a figure and date would change the picture. … Show more

Navitas is an option on Nvidia's 800V DC data-center architecture, where GaN handles the step-down from 800 volts to processor level inside the rack. The design halves power conversion stages so racks can reach the up to 8 megawatts Rubin Ultra needs, with full deployment alongside Kyber in 2027. Navitas, on Nvidia's partner list, is the only US-listed pure-play GaN company. Revenue is minimal until the Kyber ramp, and Nvidia's 29-company partner list means hard pricing pressure. x.com 

Power Integrations addresses the coming AI data-center power bottleneck, where the real alpha lies. Next-generation AI data centers are shifting to 800 VDC, and traditional silicon cannot handle the power density and heat of megawatt-scale racks. Power Integrations is working with Nvidia on that transition, and its 1250V and 1700V PowiGaN switches exceed 98% efficiency while cutting copper and board space. Despite the run, its repricing is in early innings as data-center revenue scales. x.com 

Wolfspeed has yet to show AI data centers can fix its economics. It has gone four quarters without sizing AI data-center revenue, reporting only decelerating sequential growth of roughly 50%, 30% and 20%, and calls the opportunity 'moderate'. Gross margin is -16.2%, with 1.7x debt to equity after a Chapter 11 that closed in September. Yet Lite-On's 800 VDC power racks, due for mass production in November, run on Wolfspeed's 200mm SiC, so a rerate could land on it first. x.com  x.com