Amazon Plans to Sell Custom AI Chips Directly, Challenging NVIDIA Dominance

What's happening

Amazon Web Services is advancing plans to sell its custom Trainium AI chips to external customers, marking a significant strategic departure from the company's longstanding practice of deploying its silicon exclusively for internal cloud workloads. Amazon AI chief Peter DeSantis told Bloomberg that AWS is in active talks to sell Trainium chips to other companies. AWS spokesperson Doron Aronson confirmed the directional shift, stating: 'While we've historically declined requests to sell chips directly, Andy noted it's quite possible we'll sell racks of them to third parties in the future.' The development was reported by Julie Bort for TechCrunch on June 18, 2026.

The commercial rationale was outlined by Amazon CEO Andy Jassy in his early April shareholder letter, where he wrote: 'If our chips business was a standalone business, and sold chips produced this year to AWS and other third parties (as other leading chips companies do), our annual run rate would be ~$50 billion.' Jassy's letter also noted that capacity for both Trainium and Trainium4 sold out almost instantly, and that the next-generation Trainium4 chip will not be available for more than a year, underscoring the demand pressure the company is navigating.

Why it matters for markets

The potential entry of AWS as an external AI chip vendor introduces a new competitive variable into a market currently dominated by NVIDIA, which operates on a $326 billion revenue run rate. Jassy's $50 billion run-rate projection for a hypothetical standalone chips business, while framed as illustrative, signals the scale at which Amazon believes it can participate in the broader AI accelerator market. NVIDIA's current revenue of $253.49 billion and market capitalization of $5.10 trillion reflect its entrenched position, but the emergence of a hyperscaler-turned-chip-vendor with existing cloud distribution relationships represents a structural shift in how AI compute could be procured and delivered.

Both NVIDIA and Amazon's Trainium line depend on TSMC for manufacturing, and TSMC's foundry capacity sits at the center of this competitive dynamic. NVIDIA has already supplanted Apple as TSMC's largest customer, and any expansion of Amazon's external chip sales program would require additional TSMC wafer allocations. TSMC, which reported revenue of $4.10 trillion and carries a market capitalization of $2.40 trillion, serves as the critical shared infrastructure layer for both companies. Capacity allocation decisions at TSMC could therefore become a constraint on how quickly AWS can scale external chip sales, particularly given that Trainium4 availability is not expected for more than a year.

NVIDIA CEO Jensen Huang has separately identified a brand-new $200 billion market opportunity in selling CPUs for AI, illustrating that the competitive frontier is expanding beyond GPU accelerators. AWS's move to external chip sales adds another dimension to that expanding landscape, as cloud providers with proprietary silicon increasingly seek to monetize hardware development costs that were previously treated as internal infrastructure expenditure.

Sectors and assets to watch

The primary tickers directly implicated in this development are NVIDIA (NVDA), Amazon (AMZN), and Taiwan Semiconductor Manufacturing Company (TSM). For NVIDIA, the materialization of AWS as an external chip vendor would represent a new source of competition in the AI accelerator market, where NVIDIA currently holds dominant positioning with a $5.10 trillion market cap and $253.49 billion in revenue. The degree of competitive pressure would depend on the pace at which Amazon can scale Trainium production and distribution to third-party customers, a timeline constrained by TSMC capacity and the more-than-one-year wait for Trainium4 availability.

For TSMC, the scenario carries a different set of implications. As the sole manufacturer capable of producing leading-edge AI chips at scale for both NVIDIA and Amazon, incremental demand from an AWS external sales program would add to an already constrained order book. TSMC's 52-week price range of $206.20 to $465.12 reflects the market's ongoing reassessment of its centrality to the global AI supply chain. Any acceleration in Amazon's chip commercialization plans would likely require negotiated capacity commitments with TSMC, making the foundry's allocation decisions a key variable for both the NVDA and AMZN competitive trajectories.

What to watch next

Key developments to monitor include whether AWS formalizes commercial agreements to sell Trainium chips or rack-level systems to named third-party customers, and any public disclosures from Amazon regarding the timeline and structure of an external chip sales program. The availability schedule for Trainium4, currently not expected for more than a year, will be a critical gating factor for any scaled external deployment. Observers should also track TSMC's capacity allocation announcements and capital expenditure guidance for signals on how the foundry is prioritizing competing demand from NVIDIA and Amazon. NVIDIA's response — whether through pricing, product roadmap acceleration, or partnership announcements — will indicate how seriously the company views AWS's commercial ambitions in the chip market.