Amazon’s aggressive spend on custom AI silicon accelerates the market shift away from commodity GPUs, giving Marvell a strategic foothold and a potentially undervalued investment opportunity.
Amazon’s 2026 capital‑expenditure outlook signals a decisive pivot toward proprietary silicon, underscoring the tech giant’s confidence in custom AI accelerators over traditional GPU vendors. By allocating a substantial share of its $200 billion budget to data‑center expansion, AWS is positioning itself to meet surging generative‑AI workloads while retaining control over performance, cost, and supply‑chain dynamics. This strategic shift not only diversifies Amazon’s hardware portfolio but also reshapes the competitive landscape for chipmakers seeking hyperscale customers.
Marvell Technology sits at the heart of this transformation. The five‑year agreement forged in late 2024 makes Marvell the primary designer of the Trainium family—Amazon’s in‑house AI‑training and inference chips. Beyond the accelerator IP, Marvell’s networking silicon powers the interconnect fabric that links thousands of servers, a critical component for latency‑sensitive AI workloads. Recent acquisitions, such as Celestial AI, deepen Marvell’s expertise in AI‑optimized interconnects, positioning the firm to capture incremental licensing and volume revenue as Trainium 3 and Trainium 4 scale through mid‑2026.
From an investment standpoint, Marvell trades at roughly 22.6× forward earnings, a multiple that appears modest given its exposure to the fast‑growing custom‑silicon segment. The company’s diversified product mix—spanning data‑center networking, automotive, and consumer connectivity—provides a buffer against volatility in any single market. However, investors should monitor the pace of Trainium adoption and any potential shifts in Amazon’s supplier strategy. Assuming AWS continues its aggressive AI‑chip rollout, Marvell stands to benefit from both direct licensing fees and indirect demand for its broader silicon ecosystem, making it a compelling play in the evolving AI hardware arena.
By Adam Levy · Feb 19, 2026 at 4:30 PM EST
Amazon is planning a big step‑up in capital expenditures for 2026.
It’s showing very strong momentum in one particular area of Amazon Web Services.
This chipmaker could benefit despite reports that it lost some work with Amazon recently.
Amazon is budgeting $200 billion for capital expenditures this year, nearly a $70 billion increase from 2025. A large portion of the new data‑center capacity will include Nvidia GPUs, but a growing share is going to other chipmakers, including one that investors can buy at a compelling price.
CEO Andy Jassy told investors that Amazon’s custom‑chip business within Amazon Web Services is now at a run‑rate of over $10 billion and still growing at a triple‑digit percentage year over year. The business includes:
Graviton CPUs
Custom AI‑accelerator chips for training and inference – Trainium and Inferentia
Jassy highlighted strong demand for the Trainium chips. The Trainium 2 chip saw its fastest‑ever ramp‑up, and Amazon announced Trainium 3 in December, expecting full supply commitment by mid‑2026. Interest is also high for the upcoming Trainium 4.
Amazon partnered with Marvell Technology (NASDAQ: MRVL) to design its Trainium chips, signing a five‑year agreement in late 2024 for Marvell to supply chips across its portfolio to AWS data centers.
Although Marvell’s stock has faced pressure due to reports that its position with Amazon—and with another major custom‑silicon customer, Microsoft—might be diminishing, the partnership remains significant:
Amazon is reportedly using AIChip for Trainium 3 and Trainium 4 designs.
Marvell’s earlier design accounts for only a small part of the chip’s overall interface, so licensing revenue from new chip sales could be lower than for Trainium 2.
Despite this, Marvell’s five‑year agreement with Amazon still stands. Marvell’s larger business is its networking chips, which Amazon will likely continue to use for interconnect, switching, and storage in data centers. Marvell’s recent acquisition of Celestial AI should bolster its position in AI‑focused interconnect chips.
Fears that Marvell would lose its Microsoft design have not materialized; the company is forecasting a substantial increase in fiscal 2028 custom‑AI‑accelerator revenue, aligned with the expected ramp‑up of Microsoft’s Maia 300 chip.
CEO Matt Murphy emphasized in December that nothing has changed in the company’s outlook since Amazon announced Trainium 3 and the reports about AIChip and Microsoft surfaced.
Before deciding, consider that Marvell is trading at roughly 22.6 × forward earnings estimates (as of this writing), offering a potentially attractive valuation for investors seeking exposure to the rising demand for custom silicon in hyperscale data centers.
Adam Levy is a contributing Motley Fool stock‑market analyst covering technology, consumer, and financial stocks and how policy, economic, and consumer trends shape personal finance, Social Security, and retirement savings. Before The Motley Fool, Adam was a financial advisor at Edward Jones. He studied finance and electrical engineering at Carnegie Mellon University.
Comments
Want to join the conversation?
Loading comments...