AWS goes full-stack with Frontier agents, Trainium3, and private AI Factories, signalling its biggest push yet into long-running enterprise automation and chip competitiveness.

OpenAI enters internal “code red,” pausing side bets to boost ChatGPT’s latency, reliability and personalisation as Google’s Gemini 3 momentum intensifies.

Global grid operators warn AI could drive a 300% surge in data-centre electricity demand by 2035, accelerating the race for nuclear, gas, and ultra-efficient cooling.

DeepSeek and Mistral escalate the open-weights war with frontier-level small models (V3.2, Mistral 3), pushing GPT-5/Gemini-3-class performance into affordable, offline deployments.

Read more here

OpenAI is testing a new “Memory search” to let users query everything ChatGPT has remembered about them, positioning the assistant as a personal archive rather than just a one-off chat box.
Read More

AWS re:Invent turns fully AI-first: Amazon launches Nova models (Lite, Pro, Sonic, Omni), Nova Forge for custom “Novella” variants, and Nova Act for building task-oriented agents on top of Bedrock.
Read More

AWS previews three “frontier agents” — Kiro for coding, Security Agent, and DevOps Agent — pitched as autonomous teammates that can work for hours or days across GitHub, Jira, CloudWatch and more.
Read More

Anthropic studied its own engineers and claims Claude can double productivity on some workflows, reinforcing the narrative that AI copilots will be embedded into every technical team.
Read More

AI adoption at work is starting to flatten according to several trackers, suggesting we’re shifting from hype to “prove the ROI or cut it” in many enterprises.
Read More

Anthropic’s leaked internal “Soul Overview” for Claude 4.5 exposes the behavioural scaffolding behind its alignment and tone, reigniting debate over how much “personality engineering” is acceptable.
Read More

Multiple outlets report that Google’s Gemini 3 plus Nano Banana Pro rollout is shifting sentiment — markets increasingly view Google as the AI frontrunner on tech and infra, not just OpenAI/Nvidia.
Read More

OpenAI has gone into internal “Code Red,” shelving ad, shopping and health-agent experiments so teams can focus on making ChatGPT faster, more personalised, and better at answering more queries end-to-end.
Read More

Amazon unveils Trainium3 at re:Invent and says its in-house Nvidia competitor chips are already a multibillion-dollar business, doubling down on vertical AI silicon.
Read More

Nvidia showcases its NVL72 Blackwell server stack, claiming up to 10× performance on top open-source frontier models via mixture-of-experts and tightly coupled GPU–network design.
Read More

The US Genesis Mission aims to use AI across 17 national labs to double scientific productivity, effectively guaranteeing another wave of HPC + accelerator demand.
Read More

Marvell moves to acquire optical-interconnect startup Celestial AI for up to $5.5bn, betting that light-based links will be essential to keep AI clusters scaling.

Read More

Raspberry Pi raises board prices as AI-driven DRAM and component costs spike, a micro-example of how AI demand is distorting broader hardware pricing.
Read More

New “designless” chip-design startups (like Ricursive Intelligence) promise to use AI agents to radically compress the time and cost needed to spin up custom silicon, threatening to rewire the chip industry.
Read More