This issue explores major AI breakthroughs transforming industries, including xAI's acquisition of Hotshot for AI video, Google's chip pivot with MediaTek, and Mistral Small 3.1's success in multimodal AI. Highlights include logistics advancements with Digit at ProMat 2025, Intel's AI overhaul under CEO Lip-Bu Tan, and a legal battle between ChemImage and J&J. Discover more, plus creative hacks with Gemini 2.0 Flash!

Google is reportedly teaming up with Taiwan’s MediaTek for its 2026 Tensor Processing Units, aiming to slash costs while keeping Broadcom in the mix, according to The Reuters. MediaTek’s appeal? Cheaper chips and a strong tie to TSMC, the world’s top chip maker. This move builds on Google’s sixth-gen TPU rollout last year, which already challenged Nvidia’s dominance. Sources say Google shelled out $6–9 billion on TPUs in 2024, so this cost-cutting collab could save billions.

Unlike Nvidia’s one-size-fits-all approach, Google’s crafting its own AI chips for research and cloud clients, with MediaTek now joining Broadcom to boost production flexibility. The shift isn’t a Broadcom breakup—think of it as Google diversifying its bench.

Elon Musk’s xAI has nabbed Hotshot, a startup crafting AI video generation models, to supercharge its tech lineup. The deal keeps financial details hush-hush, but it’s clear Hotshot’s video skills will scale up fast on xAI’s Colossus supercomputer—loaded with 200,000 Nvidia chips—setting the stage to challenge OpenAI’s Sora and Google’s Veo 2, especially with Musk’s “Grok Video” hint from January.

Hotshot, launched as Natural Synthetics Inc. in 2023, started with AI photo tools before pivoting to video in 2024, spinning 10-second, 720p clips from 600 million videos. Fueled by cash from Reddit’s Alexis Ohanian and SV Angel, it trained its model on Nvidia A100s over four months, using bfloat16 for speed and a second AI to caption clips—now, paired with xAI’s Colossus, it’s primed to redefine video creation.

Mistral AI dropped Small 3.1 and it’s a beast—outrunning Gemma 3 and GPT-4o Mini with better text, multimodal smarts, and a 128k-token context window, all under an Apache 2.0 license. Clocking 150 tokens per second, this lightweight model shines across benchmarks, from instruction tasks to multilingual chats, making it the top open-source pick for speed, power, and versatility.

Need a fast, flexible AI? Mistral Small 3.1 runs on a single RTX 4090 or a 32GB Mac, perfect for on-device tricks like image processing, diagnostics, or chatbots. Startups can fine-tune it for niche gigs—think legal bots or tech support—while its low-latency function calling and multimodal skills (text, images, you name it) make it a go-to for everything from quality checks to customer service, live now on Hugging Face and La Plateforme!

AI’s not just talk—it’s transforming epilepsy care! A study from King’s College London and UCL shows MELD Graph, an AI tool, catches 64% of brain abnormalities (focal cortical dysplasia) that radiologists miss. Trained on 1,185 MRI scans, it’s speeding up diagnoses, slashing NHS costs by up to £55,000 per patient, and paving the way for life-changing surgeries.

Hungry for More AI Healthcare Wins?
Dive into AIHealthTech Insider – Issue #40 for the full scoop on AI’s medical revolution.

Stay Ahead with Medicine’s Future—Subscribe Now!

At ProMat 2025 in Chicago, Agility Robotics wowed crowds with Digit, its next-gen humanoid robot, flexing autonomous muscle in a warehouse demo. The bot tackled tote loading, recycling, and palletizing with slick precision, running solo for over six hours—showing off upgrades like a curvier torso and beefier batteries. “Digit’s here to solve real logistics headaches,” said a company rep, hinting at its warehouse takeover potential. Fresh off moving 1,000+ totes in a day, Digit’s proving it’s no prototype—it's a supply chain game-changer.

Intel’s fresh CEO, Lip-Bu Tan, is diving in to shake up the chip giant, eyeing big cuts to middle management and a bold pivot in AI and manufacturing strategies, per Reuters. After quitting Intel’s board in August 2024 over clashes with ex-CEO Pat Gelsinger, Tan’s back with a vengeance, warning staff of “tough decisions” to revive the struggling company—think leaner teams and smarter chips.

Adobe’s latest Analytics report reveals a 1,200% spike in U.S. retail site traffic from generative AI sources since July 2024—skyrocketing 1,950% on Cyber Monday. Drawing from 1 trillion+ site visits and a 5,000-person survey, Adobe shows 39% of shoppers now use AI for research, deals, and gift ideas. AI-driven visitors linger 8% longer, browse 12% more pages, and bounce 23% less, proving chatbots are turbocharging online engagement—though conversions lag 9% behind traditional channels.

In a David-and-Goliath showdown, Pittsburgh’s ChemImage is battling Johnson & Johnson in Manhattan federal court over a $1.5 billion breach-of-contract lawsuit. The small biotech, armed with AI-powered imaging tech to revolutionize surgical robotics, claims J&J ditched their 2019 deal—worth $7 million upfront and up to $1.5 billion in royalties—leaving it high and dry by 2023. Testimony kicked off Monday with founder Patrick Treado defending their work, while J&J insists ChemImage flopped on milestones. With $180 million and key patents at stake, this trial pits cutting-edge AI against corporate muscle

Google’s Gemini 2.0 Flash is gaining attention for more than just its AI-powered image generation—users have discovered its ability to remove watermarks from copyrighted images with alarming accuracy. Social media reports show the AI seamlessly erasing watermarks from stock photos and protected media, automatically filling in missing details.

Controversy aside, Gemini 2.0 Flash is a startup superpower—crank out slick logos, ads, or mockups in minutes via Google AI Studio. Just log in, pick the model, drop a prompt like “neon fitness banner,” tweak it live, and snag pro-grade visuals, no design skills needed. It’s fast, free, and a branding game-changer—dive into our full guide at Medium to level up your hustle!