Amazon May Start Selling Its Custom AI Chips to Outside Companies. Should Nvidia Investors Be Worried?

Amazon is reportedly in early talks to sell Trainium AI chips for use in other companies' data centers, potentially creating a new competitor to Nvidia in AI accelerators. Amazon's custom-silicon business has already reached a more than $20 billion annual revenue run rate, while Nvidia's data center revenue still rose 92% year over year to a record $75.2 billion in its latest quarter. The news is notable for long-term competitive dynamics, but the article frames the immediate threat to Nvidia as limited.

Analysis

The market is treating this as a gradual broadening of the AI supply chain rather than a direct Nvidia displacement event. The more important second-order effect is that Amazon validating external Trainium sales effectively turns custom silicon from a captive cost-center optimization tool into a strategic pricing anchor for hyperscaler procurement teams, which should pressure GPU ASP growth at the margin over the next 12-24 months. That matters most where buyers have the scale to dual-source and the willingness to absorb software-porting costs; it matters least for smaller enterprises that remain locked into CUDA and Nvidia’s ecosystem.

Near term, the real beneficiary is Amazon, not because chip sales become a huge standalone profit pool immediately, but because the option value of external monetization lowers the hurdle rate on every incremental Trainium wafer and increases AWS’s bargaining power versus Nvidia, Broadcom, and Marvell. MRVL is an underappreciated swing factor: if AWS expands Trainium beyond internal deployment, design-win breadth and attach rates for networking and custom semiconductor content likely rise, but the timing is the key risk since early customer conversations rarely translate into volume for quarters.

For Nvidia, the bear case is not unit loss; it is mix and pricing. If large buyers conclude that second-source capacity exists, Nvidia may need to defend share with packaging, software, or financing concessions, which would show up first in gross-margin expectations before appearing in headline revenue. The counterpoint is that Nvidia’s demand curve remains supply-constrained, so any share pressure should be deferred until Amazon actually ships racks externally and proves deployment economics outside AWS.

Contrarian take: the consensus is over-focusing on competitive threat and underestimating how much this announcement confirms a still-insatiable market. Amazon can be both a Trainium seller and a massive Nvidia customer because AI capex is expanding faster than substitution is occurring; that argues for relative value rather than outright bearishness. The cleanest expression is long the diversified monetization of AI infrastructure versus short the most valuation-sensitive parts of the GPU trade, not a blanket short on Nvidia.

Key Decisions for Investors

• Maintain a tactical long AMZN / short NVDA pair for 1-3 months: Amazon gains optionality and bargaining power from external Trainium commercialization, while Nvidia faces the first credible narrative of pricing pressure; cover the short if NVDA gross-margin guidance re-accelerates.
• Add MRVL on weakness over the next 2-6 weeks: any broader Trainium deployment outside AWS should increase custom-silicon content and networking attach rates; downside is a delay in external shipments, so size as a catalyst trade, not a core position.
• Avoid chasing NVDA strength into the next print; use call spreads instead of outright longs if you need exposure: the stock still has fundamental support, but multiple compression risk rises if hyperscalers keep advertising alternative silicon paths.
• Long GOOGL as a relative-value hedge against custom-chip diversification: TPU externalization is an underappreciated proof point that hyperscalers can monetize in-house silicon, which supports valuation durability across the platform cohort.
• If AMZN rallies on the headline, wait for a pullback to build exposure rather than buying strength: the near-term upside is mostly narrative, while the real financial impact from Trainium external sales is likely a 2026-2027 story.