Amazon Web Services remains the leading cloud provider and relies on Nvidia GPUs for AI workloads, while also developing its own in-house chips to offer cost-conscious alternatives rather than replacing Nvidia.

Amazon’s Trainium and other in-house chips bolster a robust and diverse AI compute market, underscoring that no single company can meet global demand without Nvidia’s foundational role.

CEO Andy Jassy reaffirmed AWS’s continued reliance on Nvidia GPUs for AI, cementing AWS as the preferred platform to run these workloads.

Context: the summary reflects information available up to August 2026 and early 2026 product launches cited in the article.

Trainium2 is delivering roughly 30% better price-performance than comparable GPUs and is nearing full uptake, while Trainium3, launching in early 2026, promises another 30–40% performance gain and is nearly sold out pre-shipment.

Investors have seen shifting sentiment toward Nvidia, with The Motley Fool noting past top picks but current recommendations varying, though Nvidia remains prominent.

Industry analysts expect broad compute demand to require a mix of chipmakers, creating a favorable environment for Nvidia rather than signaling a collapse in its market share.

Across hyperscalers, Google uses TPUs, Microsoft develops Maia, and Meta has MTIA, illustrating a balance between Nvidia dependence and internal chip development for cost and strategic reasons.

Amazon’s GPU deployment pipeline is expanding, with an additional 2 million Nvidia GPUs ordered for 2027–2028 on top of a prior 1 million, bringing the total near 3 million.

Nvidia remains the leading AI chip supplier with strong revenue growth and high gross margins, driven by GPUs and related tools, supported by ongoing chip updates.

Nvidia’s ongoing innovation, including annual chip refreshes and margins above 70%, helps sustain leadership even as peers expand their own AI hardware efforts.

The article frames Nvidia’s competitive risk as coming from customer-turned-competitors, yet it remains central for frontier AI workloads.