If you would like to get Chasing AI delivered to your inbox every Friday then you can sign up here.

I'm back from a bit of a break. The AI news did not stop for the whole of holiday period so it's a bumper issue and the editing for the shorts section is a little rough - but I have tried grouping by topic to help you more easily sift through it all.

NVIDIA SIGNALS THE GPU MONOCULTURE IS OVER

Nvidia struck a $20B strategic licensing deal with Groq (not the Elon Musk dumpster fire Grok, this is the AI hardware startup), which is being seen as a response to inference overtaking training when it comes to data centre revenue... as well as an acknowledgement that the "LPU"s developed by some ex Google AI hardware pioneers might be better for inference than Nvidia GPUs.
SO WHAT: Nothing says "Google is real a threat in AI hardware" like paying top dollar for the guy (and IP) that helped build Google's original AI hardware playbook. Sadly, the deal seems to be an IP and acquihire, which unfortunately means the Groq business/service will be put into palliative care.
> Read More

META BUYS MANUS FOR $2B+ 

Meta is acquiring Chinese-founded AI start-up Manus, with reports suggesting a price tag above $2bn. Meta says the deal strengthens its push into AI "agents" that can plan and execute complex tasks with minimal user input, and will keep operating and selling Manus' service.
SO WHAT: Meta needs to start making money from AI instead of just giving away AI compute for free so people can make fun selfies or ask questions in WhatsApp - and what better way to do that than to buy a company that already actually makes money selling AI services.
> Read More

MUSK'S RESPONSE TO AI NON CONSENSUAL NUDES AND CSAM? JUST PAYWALL IT

Grok, Elon Musk's AI chatbot, is under scrutiny for generating inappropriate content, including sexualised images of women and children. Despite public backlash and legal threats in some countries, Musk has dismissed the concerns, framing them as attacks on free speech. To try to shush everyone he made the concession of restricting the capability to paid users only.
SO WHAT: Musk is morally bankrupt and needs to make money out of the huge investment that has been made into xAI - and it seems there is a market of cashed up deviants out there for him to sell to. Meanwhile governments around the world seem to be struggling to work out how they can regulate these issues.
> Read More

THE US MILITARY IS GOING TO USE GROK

Fearless all American superstar Pete Hegseth has announced that the Pentagon will integrate Elon Musk's Grok AI into its systems, regardless of concerns surrounding the tool's quality, reliability and accuracy. 
SO WHAT: Given Grok was the first AI to be crowned as "MechaHitler", it is known for parroting Musk's personal values/beliefs and it is also the one that is currently willing to digitally undress minors - it does seem logical that the Republican administration would choose it. 
> Read More

MICROSOFT AND UNIONS INDULGE IN SOME AI PR

Microsoft has announced a partnership with the Australian Council of Trade Unions (ACTU) to ensure "ethical" AI implementation in workplaces. The collaboration will focus on training, consultation, and policy development, aiming to protect workers' rights and address concerns about AI's impact on jobs.
SO WHAT: Last I checked, Microsoft sells AI services to businesses without any constraints on whether they use it to displace staff... and I am pretty sure if they added that clause then people would move to other providers. The agreement is also not legally binding. Call me cynical, but this looks more like the ACTU getting a bunch of free AI training and services in exchange for Microsoft getting some nice PR. 
> Read More

AI SHORTS

AI Assistants & Productivity Tools

• Amazon is taking Alexa+ beyond Echo devices with Alexa.com and a revamped, more "agent-forward" mobile app pitching family workflows (calendars, groceries, smart home) while asking users to hand over emails/docs, which is a big trust leap - more

• Google is testing "CC," a Gemini-powered email assistant that plugs into Gmail/Drive/Calendar to send a daily "Your Day Ahead" brief and take requests via email - another step toward assistants doing the boring admin for us - more

• Salesforce has revamped Slackbot into a gen‑AI "agent" for Business+/Enterprise+ that can allegedly find info, draft emails and schedule meetings across tools - a pitch that'll live or die on real adoption, not demos - more

• Google is rolling out Gemini "Personal Intelligence" (beta) to reason across your Gmail, Photos, Search and YouTube history for proactive, personalised answers - off by default, but clearly another incremental step toward Google's always-on AI assistant - more

• Google is testing a Gemini "Auto Browse" tool that lets AI run Chrome including browse, manage tabs, and execute web tasks - but it may be locked to a premium Gemini Ultra plan - more

• Anthropic launched Cowork (research preview) in the Claude Desktop app, letting non-coders point Claude at a folder to read/edit files and run multi-step actions - more

• Woolworths is upgrading its Olive chatbot with Google Gemini to meal-plan, build lists and auto-fill carts - insisting it won't shove paid brand deals into your basket - more

• OpenAI is rolling out "Your Year with ChatGPT" (a Spotify Wrapped-style recap with awards, poems and images) - I think is likely to scare people by making them realising what info these companies hold about them - more

AI Hardware & Consumer Devices

• Meta's Ray-Ban/Oakley AI glasses add "Conversation Focus" to amplify the person you're talking to in noisy places (US/Canada first), plus a Spotify "play what you see" gimmick - more

• An'An is a wool-and-sheepskin robotic panda for elder loneliness, packed with touch sensors, voice analysis and long-term "emotional AI" memory - cute and a little Black Mirror - more

• A leaked patent suggests OpenAI's $1B Jony Ive hardware play could be an AI "smart pen" with a camera that digitises handwriting and uses ChatGPT to suggest edits in real time - sorry, but this seems naff to me - more

• Meta halts international sales of its Ray-Ban smart glasses, citing overwhelming US demand and long waitlists, or in other words a very different roll out to Google Glass- more

• Ford plans to release an AI assistant for its cars and level 3 autonomous driving, becuase we all want our cars to have their own AI that can talk to us - more

Autonomous Vehicles & Robotics

• Waymo is testing a Gemini-powered in-car "Ride Assistant" for robotaxis (spotted via a leaked system prompt) that can answer questions and control climate/lights/music - more

• Nvidia unveiled some specialised open source self driving car AI models - 2026 will be the year self driving gets real - more

• Google DeepMind is plugging Gemini Robotics into Boston Dynamics' Atlas (and Spot) and will trial them in Hyundai auto factories. BMW already did this with Figure and there will be more to come no doubt - more

AI in Healthcare

• Google Health AI has open-sourced MedASR, a speech-to-text model trained on ~5,000 hours of de-identified clinical audio, beating general models on medical dictation - I think GPs not using AI note taking will be in the minority soon - more

• OpenAI has launched ChatGPT Health in the US, letting users upload medical records plus Apple Health/Peloton/MyFitnessPal data for personalised answers - promising separate storage and no training use - more

• Meanwhile a Guardian investigation reveals a very small minority of Google's AI health summaries can be dangerously misleading, prompting some removals - who would have thought some stuff on the internet could be wrong - more

• Anthropic follows OpenAI into healthcare, offering AI tools for medical uses, albeit with limited real world use cases it would seem for now - more

AI Business, Investment & Acquisitions

• OpenAI is seeing better margins on business sales, but it isn't nearly enough - more

• WSJ reports OpenAI is courting up to $100B in fresh funding, potentially valuing it at ~$830B - an enormous test of investor appetite as the "how do we pay for this?" question still hangs - more

• Anthropic (Claude) is reportedly raising $10B at a $350B valuation - nearly doubling since its last round - more

• ElevenLabs' CEO claims the voice-AI startup hit $330M ARR in 2025 (up from $200M in five months), with enterprises using its voice agents for 50k+ calls/month - more

• Apple has signed a non-exclusive, multi-year deal to use Google's Gemini to power AI features like Siri (reportedly ~$1B/yr) - a big admission its in-house AI isn't landing - more

• Universal Music is partnering with Nvidia to apply its Music Flamingo model to UMG's catalog, promising richer search/discovery (emotion, structure, "cultural resonance") and "responsible AI" tools for artists - but details, as usual, are thin - more

• Mistral, the French AI startup, lands a deal with France's military to provide AI tech - some sovereign AI capability would be nice wouldn't it? more

AI Infrastructure & Compute

• Meta signed nuclear power deals for 6+GW to feed AI data centres - more

• Nvidia is demanding full upfront, no-refund payment from Chinese buyers of H200 AI chips as Beijing drags its feet on import approvals - more

• Meta is launching "Meta Compute" to build tens of gigawatts of AI/data-centre capacity this decade - more

• Microsoft is rolling out a "community-first" data-centre blitz for AI, promising to cover grid costs so locals' power bills don't rise (plus jobs and less water use) - a savvy PR move as backlash and protests mount - more

• Apparently the race is on to build AI data centres in space, with China in the lead - more

AI Technical Research & Models

• DeepSeek published a new training method ("Manifold-Constrained Hyper-Connections") that analysts call a scaling breakthrough, but it looks incremental rather than upending - more

AI Commerce & Search

• Google unveiled Universal Commerce Protocol (UCP), an "open" standard with Shopify/Walmart/etc to let AI agents handle discovery→checkout→support in AI Search Mode/Gemin - more

• AI-based search summaries and chatbots may end the "traffic era" for online content, with big downard trends predicted for 2026 - because people won't need to click to get the info they want - more

• Tailwind CSS' creator said AI is gutting his business: docs traffic down 40%, revenue down ~80%, forcing layoffs of 75% of engineers - then Google's AI team swoops in as a sponsor (arsonist and fire brigade all in one) - more

AI Safety, Ethics & Regulation

• WIRED says Roblox's new AI selfie age-check for chat is already misclassifying kids/adults, is easy to spoof (even with avatars), and "age-verified" accounts are being sold for $4 - more

• Instagram is being flooded by AI "influencers" posting undisclosed fake sex-scandal images with celebs (LeBron, iShowSpeed, The Rock) to drive traffic to paid nude sites - another day, another Meta moderation fail - more

• Reports show "Bush Legend," a viral TikTok "Aboriginal" wildlife storyteller, is entirely AI - raising the ugly new frontier of monetised digital Blakface as synthetic influencers flood socials - more

• Matthew McConaughey is trademarking video clips, photos and his "Alright, alright, alright" audio to deter AI deepfakes and drag offenders into court, this seems to be turning into the preferred protection mechanism for most celebs - more

• Australia shelves plans to let AI train on copyrighted material amidst backlash, opting to wait three years for further review (ie. kick it down the road and let it happen covertly in the meantime) - more

AI & Cybersecurity

• WIRED warns AI models are nearing an "inflection point" in finding software vulnerabilities, forcing a rethink of how we build code; expect CIOs/CTOs to face AI-on-AI cyber warfare with defenders vs attackers both automated - more

AI in Australia

• Suncorp is going all in on AI apparently - some interesting, albeit high level, insights into their approach are in this article - more

• AI tops Aussie business leaders' worry list, overshadowing inflation and housing - as it probably should be given the extistential threat it poses to businesses and humanity - more

That's it for this week. For those asking, yes, all the photos of me in these emails are not really me and are AI generated likenesses.