In partnership with
Dear Readers,
Sometimes it feels like we are witnessing a new industrial revolution—only in real time and accelerating with every headline. Today, we are seeing developments that amaze not only experts: gigantic data centers that consume as much energy as entire cities, models that effortlessly combine text, images, audio, and video, and robots that suddenly play badminton with humans. It all seems like something out of science fiction – except that it's happening now.
In this issue, you can expect stories that show how closely innovation and infrastructure are now intertwined: from the billion-dollar pact between OpenAI and NVIDIA to Qwen's leap to true multimodality to a robot dog that can keep up on the playing field. It's about power, speed, and creativity—and who will ultimately determine the rules of this new world. Stay tuned, the most exciting twists and turns are yet to come.
In Today’s Issue:
⚡ Nvidia and OpenAI are joining forces
🏸 Scientists just taught a robot dog to play badminton
💰 Figure AI just raised over $1B
🤖 Nvidia and Abu Dhabi are teaming up
✨ And more AI goodness…
All the best,
Qwen3-Next 80B FP8 Launch
Qwen has released two new 80B models—Qwen3-Next-80B-A3B-Instruct-FP8 and Qwen3-Next-80B-A3B-Thinking-FP8—optimized with FP8 precision for much faster inference. They’re fully compatible with Transformers, vLLM, and SGLang, and available now on Hugging Face and ModelScope.
UK and NVIDIA launch AI industry offensive
NVIDIA is expanding a nationwide AI infrastructure with Nscale, CoreWeave, and Microsoft: up to 120,000 Blackwell Ultra GPUs and up to £11 billion in investments in UK data centers by 2026—the largest expansion in the country's history; At the same time, Nscale is scaling 300,000 Grace Blackwell GPUs globally, 60,000 of which are in the UK. With “Stargate U.K.” for OpenAI models (including GPT-5), a Microsoft/Nscale supercomputer with >24,000 Blackwell Ultra GPUs, and a renewable-powered CoreWeave data center in Scotland, the country is strengthening its sovereignty in high-end computing.
Qwen3-Omni: One model for all modalities
Alibaba Cloud introduces Qwen3-Omni, the first native end-to-end omni-modal AI model that combines text, image, audio, and video in one system without trade-offs. It achieves state-of-the-art results on 22 of 36 audio and AV benchmarks, supports 119 languages in text, 19 spoken languages as input, and 10 as output, with only 211 ms latency and up to 30 minutes of audio comprehension. In addition to tool calling and system prompt customization, several variants, including the low-hallucination Captioner model, are available as open source.
too many things to release. don't know how to plan. alright, let it be.
— #Junyang Lin (#@JustinLin610)
10:23 AM • Sep 22, 2025
This week will likely see some major AI releases again. We'll keep you up to date!
Unitree impresses every week with new updates on what their G1 robot series can do. Incredible!
I've been observing this "NF2" experiment in the Sora web app for a while now, but there now seem to be related new subdomains and more mentions like new NF2 Feed Page, Web B, Composer, Recording, Accepted Invite, Fast Pass and more, possibly hinting at a new version of Sora
— #Tibor Blaho (#@btibor91)
11:20 AM • Sep 22, 2025
The evidence pointing to Sora-2 is mounting. Not long ago, OpenAI stole the show from Google when they released Sora, and Google's Veo underperformed in comparison. Times have changed. Now the question is: Can Sora 2 keep up with Veo-3 or, soon, Veo-4?
The Takeaway
👉 OpenAI and NVIDIA commit to building at least 10 GW of compute, powered by millions of GPUs
👉 NVIDIA will invest up to $100B, released as each gigawatt is deployed
👉 First gigawatt comes online in H2 2026, built on the Vera Rubin platform
👉 Hardware and software roadmaps will be co-designed, accelerating both performance and AI capability
An AI infrastructure buildout on the scale of ten nuclear reactors is now underway. OpenAI and NVIDIA have announced a partnership to deploy at least 10 gigawatts of NVIDIA systems, a move that could reshape the backbone of artificial intelligence. NVIDIA is committing up to $100 billion, invested gradually as each gigawatt of compute power comes online, with the first stage planned for the second half of 2026 on the upcoming “Vera Rubin” platform. This isn’t just about more GPUs - it’s about creating physical capacity for models that will demand unprecedented power and coordination.
For the AI community, the message is clear: the frontier is no longer just algorithms or datasets, but the massive infrastructure needed to train and deploy them. Ten gigawatts means millions of GPUs, energy grids stretched to their limits, and a new phase of co-design where hardware and software evolve together. This partnership signals a future where compute scale will dictate the speed of breakthroughs, and where ownership of infrastructure could determine who leads the next era of AI. The optimism lies in what this foundation could unlock - models capable of deeper reasoning, more fluid multimodality, and agentic behavior that feels closer to intelligence itself. How might this scale change the way we design, trust, and apply AI tools in the coming years?
Why it matters: This deal underscores that infrastructure is now the decisive battleground for AI progress. By tying innovation to massive energy and hardware commitments, OpenAI and NVIDIA reveal what it truly takes to scale intelligence to the next level.
Sources:
The holiday targeting playbook marketers are using this season
Black Friday and holiday ad campaigns are make-or-break moments. While competitors burn budget on broad audiences, you can zero in on shoppers ready to buy exactly what you sell.
Speedeon’s Holiday Digital Audience Guide unlocks 100+ hyper-focused segments—from Amazon holiday shoppers and Cyber Monday fanatics to gift card purchasers and families starting new holiday traditions.
These aren’t generic lists. They’re proven, conversion-driven digital audiences designed for immediate activation across Meta, Google and more.
Let us help you wrap up your best-performing holiday campaigns yet.
Mag-7, which is ultimately artificial intelligence, is keeping the S&P 500 alive.
Scientists trained a robot dog to play badminton against humans
Engineers taught ANYmal, a four‐legged robot (~50 kg, with an added arm, stereo camera), to play badminton vs. humans. Trained via simulation for ~50M trials, then transferred to real hardware. It tracks shuttlecock, swings at ~12 m/s (roughly half speed of an amateur human), returns to center etc.
Robotics startup Figure valued at $39 billion in latest funding round
Figure AI raised over $1B in Series C, pushing its valuation to ~$39B. The company aims to scale humanoid robots for commercial & household use, improve its AI platform Helix, build out manufacturing (BotQ), and collect more robot data.
Nvidia and Abu Dhabi institute launch joint AI and robotics lab in the UAE
Nvidia has partnered with Abu Dhabi’s Technology Innovation Institute (TII) to create the Middle East’s first Nvidia AI Technology Center. The lab will work on humanoids, robotic arms, four‐legged robots, etc., using Nvidia’s new Thor chip to power advanced robotic systems.
Launch, grow, & monetize your newsletter with Beehiiv
It's the simplest platform to start, yet powerful enough to scale with your ambitions. Effortlessly build your audience and transform your content into a thriving business.
