On Oct 15, Amazon released ALoDLM-1.7B and 8B, adaptively looped diffusion language models built on Qwen3.
Official source: Amazon · DailyCited record
- Reported by 36Kr
Amazon CEO says AWS cloud capacity is largely sold out through end-2027 on AI demand, per 36Kr
Not confirmed by the company — reported on Oct 23, 2026 (source: 36kr.com).
36Kr, citing Jiemian News, reports Amazon CEO Andy Jassy saying AI infrastructure demand has largely sold out AWS cloud capacity through the end of 2027, with many orders already stretching into 2028, as Amazon races to serve AI labs and enterprises.
Source: 36Kr - Reported by 36Kr
Amazon is buying billions of dollars of Nvidia GPU racks ahead of time, but some sit idle in warehouses for lack of power, per 36Kr
Not confirmed by the company — reported on Oct 21, 2026 (source: 36kr.com).
An internal Amazon email shows a new early rack-acceptance program under which the company will take about $5 billion of Nvidia GB300 server racks for US data centers before the sites are ready, then keep adding about $3 billion of hardware a month, according to a Cailian Press report carried by 36Kr. Several AWS sites do not yet have enough power, so many of the GPUs are sitting unused in staging warehouses.
Source: 36Kr - Official
Amazon releases ALoDLM-1.7B and 8B, adaptively looped diffusion language models built on Qwen3
Amazon published this on Oct 15, 2026 (source: huggingface.co).
Amazon published ALoDLM-1.7B and ALoDLM-8B on Hugging Face under CC BY-NC 4.0, diffusion language models initialized from Qwen3 that reuse shared transformer layers and let unresolved tokens keep refining their hidden states while confident tokens commit early, with a learned policy deciding when to stop looping. On the model card's table of eleven reasoning, knowledge, math, science and code benchmarks, ALoDLM-8B averages 80.3 versus 78.5 for Qwen3-8B, and the card plots single-stream throughput on one NVIDIA B200 against autoregressive and diffusion baselines. The weights are for noncommercial research and ship with an ALoDLM-aware nano-vLLM engine.
Source: Amazon - Reported by TechCrunch
Amazon devices chief Panos Panay says ambient AI devices will anticipate needs and reduce reliance on apps, per TechCrunch
Not confirmed by the company — reported on Oct 15, 2026 (source: techcrunch.com).
Speaking at TechCrunch Disrupt, Amazon devices chief Panos Panay described a future where ambient AI devices work in the background and anticipate users' needs instead of waiting for commands, reducing the need to open apps. The remarks frame Amazon's device strategy around always-available AI rather than app-centric interfaces.
Source: TechCrunch - Official
AWS shows how to activate Amazon Bedrock private pricing offers across many accounts with License Manager entitlements
AWS published this on Oct 13, 2026 (source: aws.amazon.com).
AWS has published a guide for organizations whose accounts must each accept AWS Marketplace Private Pricing Offers before using Amazon Bedrock models. It shows two ways to activate the offers' entitlements in AWS License Manager across dozens or hundreds of accounts: a lightweight script for a one-time fix, and CDK automation for ongoing event-driven processing.
Source: AWS - Reported by Wall Street Journal
Amazon is in talks to buy AI startup Decart for around $7 billion, per Wall Street Journal
Not confirmed by the company — reported on Oct 11, 2026 (source: wsj.com).
The Wall Street Journal reported that Amazon is in talks to acquire Decart, an AI startup known for live-video and simulation models including Lucy and Oasis, in a deal that could be valued around $7 billion. Talks could still fall apart or draw another bidder; neither company has publicly confirmed an agreement. Decart was valued near $4 billion in a May round that included Nvidia.
Source: Wall Street Journal - Official
AWS shows BlockRun and Incarna using Bedrock AgentCore payments for pay-per-inference agents
AWS published this on Oct 8, 2026 (source: aws.amazon.com).
An AWS Machine Learning blog post describes Amazon Bedrock AgentCore payments giving agents a managed way to pay for services on demand with infrastructure-enforced spend limits. It shows Incarna agents paying BlockRun for model inference one request at a time over x402, cutting the work of adding x402 support from months to days.
Source: AWS - Reported by The Decoder
Zenity says one public Bedrock AgentCore agent could hijack every agent in the same AWS account; AWS patched it, per The Decoder
Not confirmed by the company — reported on Oct 8, 2026 (source: the-decoder.com).
Zenity Labs researchers say a single publicly accessible Amazon Bedrock AgentCore agent was enough to take over every AgentCore agent in the same AWS account and region by abusing an internal interface for temporary credentials. AWS has patched the issue and tightened default permissions, per The Decoder.
Source: The Decoder - Official
AWS shows how to evaluate multi-agent systems with AgentCore Evaluations
AWS published this on Oct 5, 2026 (source: aws.amazon.com).
An AWS blog post builds a Strands-based multi-agent supply chain decision system and evaluates tool selection, constraint handling and explainability with Amazon Bedrock AgentCore Evaluations.
Source: AWS - Reported by TechCrunch
Amazon open-sources Strands Decider 2B, its own Jev-style decision model, per TechCrunch
Not confirmed by the company — reported on Oct 1, 2026 (source: techcrunch.com).
AWS's Strands Labs released Strands Decider 2B as fully open source — a high-speed, low-cost decision model built on the torso of Qwen3.5-2B that picks between pre-defined options and outputs a confidence score rather than generating text. Engineer Marc Brooker started it as a homebrew project after seeing TypeSafe's Jev, and it briefly topped the Jevbench ranking for its size class before Amazon formalized the release. It lands the same week OpenAI announced its own decision-model offering, as dozens of Jev-inspired clones appear across the industry. TypeSafe CEO Diogo Almeida downplayed the competition, saying the copycats seem more like ML people implementing a cool architecture than teams making the models genuinely smart.
Source: TechCrunch