AI inference lifts a broad watchlist
Surging AI inference demand is pushing investor interest across the stack — names called out include NVIDIA, AMD, TSMC and Micron alongside optics and power suppliers (x.com). The same commentary also flags smaller plays tied to optics and power management, like LITE and TXN, as beneficiaries of inference growth (x.com).
Artificial intelligence inference — the step where a trained model answers a prompt — is widening the chip trade beyond the biggest graphics processors. NVIDIA, Advanced Micro Devices, Taiwan Semiconductor Manufacturing, Micron, Lumentum and Texas Instruments all have current filings or product updates tied to that buildout. (nvidia.com, amd.com, investor.tsmc.com, investors.micron.com, investor.lumentum.com, ti.com) Inference is the serving side of artificial intelligence: a model is already trained, and the system must answer millions of user requests cheaply and fast. NVIDIA said on February 12 that Baseten, DeepInfra, Fireworks AI and Together AI cut cost per token by as much as 10 times on Blackwell-based systems running open models. (nvidia.com) That shifts attention from one chip to the whole machine. Faster answers require compute chips, high-bandwidth memory, advanced packaging, optical links that move data between racks, and power circuits that convert electricity efficiently inside servers. (amd.com, investors.micron.com, nvidia.com, ti.com) NVIDIA remains the center of that spending. In its fiscal 2026 annual report, the company said Blackwell products represented the majority of Data Center revenue, and its March 2026 networking update said co-packaged silicon photonics can deliver 5 times the power efficiency and 10 times the resiliency for large artificial intelligence networks. (marketscreener.com, nvidia.com) Advanced Micro Devices is pitching the same inference problem from the challenger position. Its Instinct MI350 page says the chips support low-precision formats aimed at faster, more efficient inference, and the company’s 2025 annual report said Data Center revenue rose 32% to $16.6 billion on demand for fifth-generation EPYC processors and Instinct accelerators. (amd.com, ir.amd.com) Taiwan Semiconductor Manufacturing sits underneath both camps because it makes the most advanced chips and packages many of them together. The company’s April 2026 first-quarter materials showed revenue of $34.6 billion, above its prior guidance range of $32.2 billion to $33.4 billion, after March sales jumped 46.5% from a year earlier in New Taiwan dollar terms. (investor.tsmc.com, cnbc.com) Micron is the memory leg of the same trade. On March 18, the company reported fiscal second-quarter 2026 revenue of $23.86 billion, up from $8.05 billion a year earlier, and said those record results reflected “the strategic value of memory in the AI era”; separate coverage of the release said Micron’s 2026 high-bandwidth-memory output is already sold out. (investors.micron.com, finance.yahoo.com) Lumentum shows why optics keeps appearing on investor watchlists. The company reported fiscal second-quarter 2026 revenue of $665.5 million on February 3, up from $402.2 million a year earlier, and said it received an additional multi-hundred-million-dollar order in co-packaged optics for delivery in the first half of calendar 2027. (investor.lumentum.com) Texas Instruments represents the power side. On March 16, it unveiled an 800-volt direct-current architecture for next-generation artificial intelligence data centers with NVIDIA, saying the design cuts conversion steps from incoming power to processor power to two; in its 2025 annual report, Texas Instruments said its Analog segment generated $14.01 billion in revenue. (ti.com, investor.ti.com) The counterargument is that not every company tied to artificial intelligence spending will benefit equally. Texas Instruments still sells broadly into industrial and automotive markets, Lumentum’s optical orders can be lumpy, and Taiwan Semiconductor Manufacturing does not report first-quarter 2026 earnings until April 16, leaving investors to wait for updated margin and capital spending details. (sec.gov, investor.lumentum.com, investor.tsmc.com) What has changed in 2026 is the market’s focus from training a model once to serving answers constantly. As inference workloads spread through cloud services, search, coding tools and agents, the watchlist has spread with them — from graphics processors to memory, foundries, optics and power. (nvidia.com, amd.com, investors.micron.com, nvidia.com, [ti.com](https://www.ti.com/about-ti/newsroom/news-releases/2026/2026-03-16-ti-unveils-complete-800-vdc-power
