This is a sample of CapRelay's research. CapRelay provides similar research and more for every US-listed company.
Start free trialA look at Jensen Huang’s life story and his position at the center of the AI race. The episode examines why he believes AI is safe despite being one of the people with the greatest ability to slow its development, and whether that stance reflects conviction, insulation, or salesmanship.
In a conversation with Brad Gerstner and Clark Tang, Jensen Huang addresses whether AI revenues can justify the industry’s spending, the compute and power bottlenecks expected in 2027, and the prospects for AI demand beyond two leading labs. They also discuss the path to the singularity, open versus closed models, responsible AI development, data centers in space, and NVIDIA’s outlook for the decade ahead.
In a conversation with Brad Gerstner and Clark Tang, Jensen Huang addresses whether AI revenues can justify the industry’s spending, the compute and power bottlenecks expected in 2027, and the prospects for AI demand beyond two leading labs. They also discuss the path to the singularity, open versus closed models, responsible AI development, data centers in space, and NVIDIA’s outlook for the decade ahead.
This episode is likely a duplicate. We show potential duplicates if they're published by different podcast series.
A technical guide to NVIDIA’s speech recognition models, including Parakeet, Canary, and Nemotron Speech. The episode compares model performance and capabilities, explains trade-offs such as streaming latency and clip-length limits, and covers licensing and a Parakeet v2 fine-tune for mixed Hebrew-English speech.
Greg Barbone, who leads developer relations on NVIDIA’s XR team, joins Logan Lawler to discuss AI applications in extended reality. They cover use cases for reducing training time and human error, NVIDIA XR AI and CloudXR, local and cloud GPU requirements, agentic AI, and open-source XR tools.
Liz Hoffman joins Ed Elson to discuss NVIDIA’s AI holding-company strategy and why she believes the company has become too essential to fail. James Taylor also weighs in on why he expects the upcoming earnings season to be especially important.
a16z General Partner Martin Casado interviews MakerMaker.AI co-founders Dhaval Adjodah and Owen He about the concept of recursive self-improvement (RSI) in AI. The conversation distinguishes true RSI from AI-assisted development, explores the technical architecture of closed-loop improvement systems across the AI stack, and discusses the challenges of measuring genuine progress, such as accounting for efficiency gains versus resource consumption. They also address technical hurdles like variance and reward hacking in self-improving systems.
MakerMaker is running the efficiency problem on live NVIDIA hardware
a16z general partner Martin Casado interviewed MakerMaker co-founders Daval Ajada and Owen He about recursive self-improvement. MakerMaker's commercial target is inference: making it cheaper to serve the newest open-weight models, with Kimi K3 as the example. They define real self-improvement narrowly. Humans using AI tools to build better AI is ordinary automation. Their bar is a closed loop in which the system being improved also does the improving, across prompts, the harness, model weights, and eventually chips. Their test has two parts. The numerator is capability. The denominator is the…
Log in to read the complete analysis.
Cisco President Jeetu Patel discusses Cisco’s partnership with NVIDIA on next-generation pluggable optics, which he says will increase bandwidth and efficiency.
ReflectionAI co-founder and CEO Misha Laskin joins hosts Sarah Guo and Elad Gil to discuss the launch of Beam, a 500-billion parameter open-weight reasoning model. The conversation covers the technical requirements for training, the broader industry shift in enterprise compute from renting to owning infrastructure, and the potential for open models to dominate global token demand. Laskin also addresses the competitive landscape, including the open-source ecosystem in China, safety considerations, and the critical role of data centers and compute capacity in accelerating scientific discovery.
Reflection trained a frontier open model on thousands of NVIDIA GB300s, and spent more compute on reinforcement learning than on pre-training
Misha Laskin, co-founder and CEO of Reflection AI and a former DeepMind researcher, says the company's first open model, Beam, is a 500B-parameter mixture-of-experts model with 23B active parameters, built for coding and agentic work. Beam's pre-training ran on 6,000 GB300s. The original run took a few weeks; with infrastructure improvements Reflection can now do it in about 12 days. Reinforcement learning used a little over 10,000 GB300s for four weeks, and Laskin says that job consumed more FLOPs than pre-training. He has not seen another documented open-source run at 10,000 GB300s for four…
Log in to read the complete analysis.
The Daily AI Show hosts discuss recent advancements in local AI, including speaker identification systems, AI-assisted coding tools, and new business video models. The episode concludes with a discussion on the integration of NVIDIA and Microsoft’s RTX Spark technology into Windows PCs, which the hosts argue signals a fundamental redesign of the computer to prioritize local AI compute and sandboxing for AI agents.
Apple Silicon is still the default local-AI machine, which is the backdrop for the NVIDIA-Microsoft PC move
Daily AI Show co-host Andy says many local-AI apps are still built for Apple Silicon or macOS only. Google’s new on-device note-taking app uses a 740-million-parameter Embedding Gemma 2 model optimized for Macs, and the ChatGPT desktop app runs on his M4 Pro but not on his Intel iMac. That is the setup for the NVIDIA news he flagged: Windows PCs have been the laggard, and Macs are “really the system of choice for AI at this point in time.” NVIDIA and Microsoft are bringing Spark-class local AI to Windows, with a built-in agent sandbox Andy says NVIDIA and Microsoft just announced they will…
Log in to read the complete analysis.
More podcast results are available.
Log in to see more