To The Moon

Google's Surge Has Amazon Seething With Envy. Will the AWS Conference Spur the Ecosystem to Strike Back?

$Amazon.com(AMZN)$   is one of the few companies besides Google that possesses a full-stack AI value chain. Despite net profit growth of 64.2%, 34.7%, and 38.2% in the first three quarters, the stock is up only 6.85% year-to-date. Its P/E multiple sits at the 28th percentile over the past decade.

In addition, looking at cumulative revenue for the first three quarters of 2025, AWS revenue is more than twice that of Google Cloud. Although AWS is growing more slowly (and is gradually accelerating; see chart below), its larger absolute scale means the value it drives across the industry chain is not necessarily less than Google's.

Amazon's launch event solidified its full-stack capabilities.

At re:Invent, AWS unveiled its next-generation AI training chip, Trainium 3, the first AWS AI chip built on a 3 nm process. Each Trainium 3 chip delivers 2.52 PFLOPs of FP8 compute, with memory capacity increasing 1.5x over Trainium 2 to 144 GB of HBM3e and memory bandwidth up 1.7x to 4.9 TB/s. AWS says the Trn3 UltraServer improves energy efficiency by 40% over the previous generation, with performance-per-watt up 4x.

Amazon also previewed the in-development Trainium 4 chip. This next-generation product will deliver a major performance leap and will support Nvidia's NVLink Fusion high-speed chip interconnect. This compatibility means AWS systems equipped with Trainium 4 will be able to interoperate with Nvidia GPUs and scale performance, while still leveraging Amazon's in-house, lower-cost server rack technology.

On Tuesday, Amazon also introduced its new Nova models. The Nova 2 family spans reasoning, multimodal processing, conversational AI, and code generation, emphasizing price-performance advantages.

The Nova 2 models launched this time are designed for different application scenarios. Nova 2 Lite is a fast, cost-effective reasoning model built for everyday workloads; it can process text, images, and video and generate text. In benchmark comparisons against competitors, it outperformed or matched Claude Haiku 4.5 in 13 of 15 tests and outperformed or matched GPT-5 Mini in 11 of 17 tests.

Nova 2 Pro is Amazon's most capable reasoning model; it can handle text, images, video, and speech and generate text, suitable for highly complex tasks such as agentic coding and long-horizon planning.

In addition, Amazon launched the industry's first “open training” service, Nova Forge. Nova Forge is a first-of-its-kind AWS service that enables enterprises to build customized versions of Nova—what Amazon calls “Novellas.” The service pioneers an “open training” model by giving customers exclusive access to Nova model checkpoints at pre-training, mid-training, and post-training stages, allowing them to blend proprietary data with Amazon Nova–curated datasets at every stage of model training.

AWS supply chain in one chart:

Given its scale and customer diversity, Amazon's industry chain is characterized by an open stance toward multiple technical routes. For example, its CPU purchases include x86 chips from $Intel(INTC)$   and $Advanced Micro Devices(AMD)$   as well as its own Arm-based Graviton4. As AWS states on its website: “We are the first major cloud provider to support Intel, AMD, and Arm processors.”

In 2024, AWS joined with AMD, $Astera Labs (ALAB.US)$ , $Cisco (CSCO.US)$ , and six other companies to launch the UALink Consortium. Its core goal is to establish an open industry interconnect specification as an alternative to Nvidia's NVLink. However, because Trainium 4 will support Nvidia's NVLink Fusion, we believe this move will instead strengthen Nvidia's ecosystem and have some impact on certain UALink partners such as Astera Labs.

Amazon launches 'ultra-fast' 30-minute deliveries

In addition to its cloud business, its traditional retail business has also made progress. Amazon is rolling out an “ultra-fast” delivery service in Seattle and Philadelphia that promises arrival in 30 minutes or less. Prime members can opt in for $3.99 per order, while non-Prime customers pay $13.99; orders under $15 carry a $1.99 small-basket fee. This move could reshape the competitive landscape and pressure the market share of Uber Eats, DoorDash, and Instacart, while expanding Amazon's e-commerce footprint.

@TigerStars  @CaptainTiger  @TigerWire  @Daily_Discussion  @Tiger_chat  @Tiger_comments  @MillionaireTiger