In partnership with
In Today’s Issue:
🚀 NVIDIA’s Jensen Huang doubled demand forecasts to $1 trillion and unveiled the Vera Rubin supercomputer platform alongside the new Groq 3 inference chip
🐶 A Sydney tech founder used ChatGPT to help create a personalized mRNA cancer vaccine that successfully treated his dog's tumor
🧠 Kimi's new attention residual architecture boosts a 48B model's performance on complex benchmarks
💼 OpenAI is reportedly in advanced talks to launch a $10 billion joint venture with major private equity firms
✨ And more AI goodness…
Dear Readers,
Jensen Huang just walked onto a stage in San Jose and casually doubled NVIDIA's demand outlook to one trillion dollars, and that wasn't even the biggest surprise of the night.
GTC 2026 kicked off with a keynote that rewrites the rules: Vera Rubin, a seven-chip supercomputer platform promising 10x the performance per watt; the Groq 3 LPU, born from a $20 billion acquisition and already shipping this year; and NemoClaw, NVIDIA's bold play to become the operating system for agentic AI.
We're on the ground in San Jose covering every angle, and today's issue breaks down exactly why this keynote matters more than any before it, but the AI world didn't stop spinning while Jensen was on stage.
OpenAI is reportedly building a $10 billion joint venture with private equity giants to crack enterprise adoption wide open, Kimi dropped a deceptively simple architectural trick that boosted a 48B model by 7.5 points on GPQA-Diamond, and a Sydney tech founder used ChatGPT to help create what may be the first personalized mRNA cancer vaccine for a dog - yes, a dog - and it's working.
Grab your coffee, this one's packed.
All the best,
Kim Isenberg
🐶 AI Helped Save Rosie
A Sydney tech entrepreneur (not an expert in biology!) used ChatGPT, genomic sequencing, and top Australian researchers to build what is described as the first personalized mRNA cancer vaccine for a dog, after standard surgery and chemotherapy failed. The breakthrough appears to have cut one major tumor by about half and improved Rosie’s energy and comfort, showing how AI, citizen science, and fast-turn personalized medicine could eventually reshape cancer care for both pets and humans.
🚀 Attention Residuals Boost Deep Models
Kimi introduced a new way for AI models to reuse information from earlier layers. Instead of treating every past layer the same, each new layer can pick the most useful earlier signals, which helps deep models avoid important information getting washed out.
To keep this fast enough for huge models, they group layers into blocks and apply the smart selection across those blocks. In their tests on a 48B Kimi Linear model trained on 1.4T tokens, this improved results on benchmarks like GPQA-Diamond (+7.5), Math (+3.6), HumanEval (+3.1), and MMLU (+1.1), with less than 2% extra inference latency.
💼 OpenAI courts PE mega-venture
OpenAI is reportedly in advanced talks to launch a joint venture with private equity firms including TPG and Bain Capital, aimed at accelerating AI adoption across their portfolio companies. Reuters says the venture could carry a pre-money valuation of about $10 billion, with PE investors committing roughly $4 billion, a big signal that enterprise AI monetization is becoming the next major battleground. OpenAI is under pressure to turn huge demand into durable business revenue, especially in high-value sectors like finance and healthcare.
We're there live for you, but Jensen's legendary keynote at the GTC will also be live streamed! Watch the NVIDIA GTC Keynote 2026 here:
NVIDIA GTC Day 1:
A groundbreaking keynote.
The Takeaway
👉 Nvidia doubled its demand outlook to $1 trillion through 2027, powered by Vera Rubin - a seven-chip platform that delivers an order of magnitude improvement in performance per watt over Grace Blackwell. The first system is already running in Microsoft Azure.
👉 The $20 billion Groq acquisition already has a product: the Groq 3 LPU ships in Q3, and its LPX rack paired with Rubin boosts inference throughput by 35x per megawatt - solving the latency-vs-throughput tradeoff that has plagued large-scale AI deployment.
👉 NemoClaw positions Nvidia as the enterprise operating system layer for agentic AI, with OpenClaw integration, a Nemotron 4 coalition (Perplexity, Mistral, Cursor), and a clear message: every company needs an AI agent strategy now.
👉 The 2028 roadmap is already taking shape: Feynman architecture with a new Rosa CPU, next-gen LPU, and Kyber rack design - signaling that Nvidia’s annual hardware refresh cycle is accelerating alongside AI’s insatiable compute demands.
Jensen Huang just turned a hockey arena into the epicenter of the AI world. At GTC 2026 in San Jose, the Nvidia CEO unveiled a staggering $1 trillion demand outlook for Blackwell and Vera Rubin systems through 2027, double the $500 billion estimate from just a year ago. The centerpiece: Vera Rubin, a seven-chip, five-rack supercomputer platform that Nvidia claims delivers 10x more performance per watt than its predecessor, Grace Blackwell. But the real surprise was Groq 3, Nvidia’s first chip from the startup it acquired for $20 billion in December.
The new Language Processing Unit slots into dedicated LPX racks alongside Vera Rubin, combining low-latency token generation with raw GPU horsepower, a pairing Huang says boosts inference throughput by 35x per megawatt (more on that down below). Nvidia also previewed DLSS 5, fusing structured 3D graphics with generative AI for what it calls “probabilistic rendering.”
On the agentic AI front, NemoClaw- an enterprise-grade reference stack for OpenClaw - signals Nvidia’s push to become the operating system layer for autonomous AI agents. With Uber deploying Nvidia’s autonomous driving software across 28 cities by 2028, and a Feynman architecture teased for the 2028 generation, one question looms large: Are we watching Nvidia evolve from chipmaker to the infrastructure backbone of an entirely new computing era?
Why it matters: Nvidia’s GTC 2026 keynote marks the moment the company officially stopped being “just” a chipmaker and became the vertically integrated platform powering the entire AI economy. With $1 trillion in projected demand and seven new chips shipping as one unified system, the infrastructure decisions Nvidia makes today will define the speed, cost, and capabilities of AI for years to come.
Sources:
🔗 https://www.cnbc.com/2026/03/16/nvidia-gtc-2026-ceo-jensen-huang-keynote-blackwell-vera-rubin.html
🔗 https://www.tomshardware.com/news/live/nvidia-gtc-2026-keynote-live-blog-jensen-huang
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
Nvidia doubled its demand outlook from $500 billion to $1 trillion in purchase orders for Blackwell and Vera Rubin through 2027.
Everything revolves around NVIDIA.
NVIDIA's Inference Game-Changer Arrives
The age of waiting for AI to think is over. At GTC 2026, NVIDIA unveiled the Groq 3 LPX - a dedicated inference accelerator built to sit alongside its new Vera Rubin platform. Born from NVIDIA's $20 billion Groq acquisition last December, the LPX rack packs 256 language processing units loaded with 128 GB of ultra-fast SRAM, delivering a staggering 40 petabytes per second of memory bandwidth.
Think of it like adding a turbocharger specifically designed for one job: spitting out tokens at blistering speed. While Rubin GPUs handle the heavy lifting of processing prompts, the Groq 3 LPUs take over the decode phase, generating responses with up to 35x higher throughput per megawatt compared to Blackwell.
This matters because agentic AI systems consume up to 15x more tokens than traditional chatbots. Agents don't just talk to humans, they talk to other agents, and they need answers in milliseconds, not seconds. NVIDIA is essentially building the plumbing for an economy where tokens are the new currency and LPX is the printing press. An amazing start for day 1 and an incredible Keynote as always by Jensen Huang.
Real-World Ads, Simple to Run
With AdQuick, executing Out Of Home campaigns is as easy as running digital ads. Plan, deploy, and measure your real-world advertising effortlessly—so your team can scale campaigns and maximize impact without the headaches.