Amazon Unveils New AI Chips and Tightens Nvidia Ties, Yet Cloud Capacity Remains Key

Amazon Web Services (AWS) sharpened its two‑track AI strategy at the annual Re:Invent 2025 conference in Las Vegas, unveiling a next‑generation custom accelerator and deepening its partnership with Nvidia. The announcements come as the cloud‑computing giant races to expand capacity, fend off intensifying competition from Microsoft Azure and Google Cloud, and capitalize on the soaring demand for artificial‑intelligence workloads.

Trainium 3: A Leap in In‑House Silicon

During the keynote, AWS CEO Matt Garman introduced Trainium 3, the latest iteration of Amazon’s custom AI training chip. Compared with its predecessor, Trainium 3 delivers roughly four times the compute performance, energy efficiency, and memory bandwidth. Early customer tests suggest the new silicon can slash training and inference costs by up to 50 %.

The move follows a broader industry trend where large tech firms develop proprietary processors to optimize AI workloads. Google’s Tensor Processing Units (TPUs), co‑engineered with Broadcom, have gained market traction after the successful launch of the Gemini 3 model, which runs exclusively on TPUs. Similarly, Meta has reportedly evaluated TPUs alongside Nvidia GPUs for its internal AI pipelines.

Hybrid On‑Premise Offering: AWS Factories

In parallel with the hardware rollout, AWS announced AWS Factories, a new on‑premise AI infrastructure service. Factories combine Trainium accelerators with Nvidia GPUs, giving enterprises access to a full stack of Nvidia‑accelerated computing, AI‑optimized software, and GPU‑centric applications within their own data centers. By offering both custom silicon and leading‑edge GPUs, AWS aims to capture a broader slice of the AI market and sustain revenue growth amid a hyper‑competitive landscape.

Capacity Expansion as a Growth Engine

Capacity has become the focal point for investors. After a year of supply constraints that limited cloud growth, AWS reported a resurgence in revenue momentum in Q3 2025. CEO Andy Jassy highlighted a 20.2 % year‑over‑year growth rate and more than 3.8 gigawatts (GW) of new compute power added in the past twelve months.

Analysts at Wells Fargo argue that Trainium 3 is “critical to supplementing Nvidia GPUs and CPUs in this capacity build” and will help close the gap with Azure and Google Cloud. Their model projects AWS will add over 12 GW of compute by the end of 2027, potentially unlocking up to $150 billion in incremental annual revenue if demand holds. Oppenheimer’s research echoes this sentiment, estimating that each additional gigawatt of compute in recent quarters translated to roughly $3 billion of annual cloud revenue, and forecasting a 14 % upside to 2026 AWS revenue and a 22 % upside to 2027.

Strategic Implications

• Technology differentiation: Trainium 3 positions AWS to offer cost‑effective, high‑performance AI training at scale, reducing customers’ reliance on expensive Nvidia GPUs.
• Supply‑chain resilience: By diversifying its silicon portfolio, AWS mitigates the risk of GPU shortages that have plagued the industry since early 2024.
• Revenue diversification: The hybrid Factories service opens a new on‑premise revenue stream, appealing to regulated industries that require data sovereignty while still leveraging AWS’s software ecosystem.
• Competitive positioning: The combined hardware offering strengthens AWS’s value proposition against Azure’s AI supercomputer initiatives and Google Cloud’s TPU‑first strategy.

Market Reaction

Following the earnings release that included the capacity and chip updates, Amazon’s shares rallied nearly 14 % to $254, though they have since retreated. As of the latest close, the stock is up 6.5 % year‑to‑date, lagging behind the broader “Magnificent Seven” and underperforming the S&P 500’s 16 % gain in 2025. The market appears to be pricing in the execution risk of scaling both custom silicon and compute capacity at a pace that matches accelerating AI demand.

Outlook

While new chips are an important differentiator, analysts agree that the decisive factor for AWS’s future growth will be how quickly and efficiently the company can translate expanded gigawatt capacity into billable AI services. Given Amazon’s logistical expertise and deep pockets, the consensus is that AWS is well‑positioned to turn today’s supply‑side constraints into a long‑term tailwind.

In sum, AWS’s dual strategy—accelerating its own silicon roadmap while deepening ties with Nvidia—signals a robust response to the AI wave. If the company can sustain its capacity buildout, it could capture a sizable share of the projected $1 trillion AI‑infrastructure market by the end of the decade.