Open Source Wins, AGI Is Here, and Scorsese’s AI Toolkit with CEOs of Cerebras & Black Forest Labs
Episode Details
The All-In Podcast features interviews with two prominent tech leaders shaping the current AI buildout. First, Andrew Feldman, CEO and founder of Cerebras, discusses the unprecedented scale of Data centers construction and the rapidly increasing Energy demand for AI. He highlights how massive compute requirements are driven by companies like OpenAI, Anthropic, SpaceX, Google, Microsoft, and AWS. Cerebras provides blisteringly fast chips optimized for AI Inference, challenging the legacy x86 architecture of Intel and the standard GPU architecture dominated by Nvidia and its CEO Jensen Huang. This competitive landscape has pushed tech giants like Amazon to build custom silicon. The panel contrasts Open source AI with Closed source AI, noting a trend towards AI Sovereignty. For example, the UAE based G42 builds custom models to protect their Intellectual Property (IP). Feldman discusses advanced reasoning, referencing the Hermes agent processing massive amounts of AI Tokens, and Z.AI producing GLM 5.2 which runs on the decentralized Bit Tensor Tao network. Reasoning engines like Fable indicate we are approaching AGI (Artificial General Intelligence) through recursive self-improvement, a trend anticipated by pioneers like Sam Altman, Ilia Sutskever, and Dario Amodei. With rapid advancements comes the need for AI Safety, as demonstrated by Palo Alto Networks discovering vulnerabilities through AI red-teaming. Feldman draws an analogy to Elon Musk revolutionizing launch costs to illustrate exponential gains. These strategic topics frequently surface at global forums like Davos. In the second segment, Robin Rombach, CEO of Black Forest Labs, discusses the future of generative media. His company pioneered Diffusion Models and created the open-source image generator Flux. Rombach explains how these systems are evolving into Multimodal Models capable of becoming Interactive World Models that predict real-world actions, eventually serving as brains for physical Robotics. He shares his experience collaborating with legendary director Martin Scorsese to use AI for film storyboarding, drawing a parallel to how George Lucas pioneered visual effects for Star Wars. Finally, the conversation touches on how massive IP holders like Disney can leverage proprietary models while balancing fan creation, akin to the licensing efforts surrounding closed models like Sora.
The episode discusses the massive buildout of AI infrastructure, driven by insatiable demand from major tech players and the rapid advancement of AI capabilities towards AGI. It also explores the evolving landscape of generative media, with a focus on multimodal models, robotics, and the potential for AI to revolutionize content creation and intellectual property.
Portfolio lens: A basket of companies and themes positioned to benefit from the exponential growth in AI compute infrastructure and the evolving landscape of generative AI, particularly concerning intellectual property and open-source innovation.
Generated with gemini-2.5-flash-lite on 7/25/2026, 5:13:21 AM. For research only. Not financial advice.AI Infrastructure Buildout and Compute Demand
The unprecedented scale of AI data center construction and compute demand, driven by frontier AI models, presents a significant long-term investment opportunity in the underlying infrastructure and specialized hardware providers.
The episode highlights the immense scale of data center construction, exceeding historical benchmarks, with companies like Cerebras experiencing a $25 billion backlog. Demand from major AI players like OpenAI, Anthropic, Google, and Microsoft is described as insatiable, with orders placed years in advance. This indicates a sustained, high-growth period for AI infrastructure.
- We're talking about individual buildings the size of football fields that have more power coming into them than midsize cities.
- The people who are buying the capacity, the open AI, anthropics, SpaceX, SpaceX AI, the Googles, they are insatiable right now.
- We have a $25 billion backlog.
- OpenAI, Anthropic, Google wants more data centers, Microsoft wants more data centers, AWS wants more data centers.
- Continued exponential growth in AI model complexity and training requirements.
- Further investment by hyperscalers in custom silicon and advanced data center designs.
- Geopolitical initiatives to build domestic AI capabilities, driving data center expansion globally.
- Potential for oversupply if demand projections are not met.
- High capital expenditure and long lead times for building new infrastructure.
- Technological obsolescence as newer, more efficient hardware architectures emerge.
- Geopolitical instability or regulatory changes impacting global data center development.
- Analyze capital expenditure plans of major cloud providers and AI labs.
- Track the growth and backlog of specialized AI hardware manufacturers like Cerebras and Nvidia.
- Monitor the development and adoption rates of custom silicon (e.g., Amazon's Inferent chips).
- Assess the energy and power requirements for future AI data centers.
Open Source Generative Models and IP Control
The increasing sophistication of open-source generative AI models, coupled with a growing desire for 'AI sovereignty' among IP holders, creates opportunities for companies developing flexible, customizable AI solutions that balance open innovation with proprietary control.
The episode discusses the rapid closing of the gap between open-source and closed-source AI models, with a trend towards 'AI sovereignty' where entities like the UAE's G42 build custom models to protect IP. Companies like Black Forest Labs, with their open-source Flux model and expertise in diffusion models, are well-positioned to cater to this demand for both advanced capabilities and controlled deployment.
- We need more domestic open source models. We need to give the world a choice.
- Companies... have concerns with the ambition of the frontier models and maybe sharing their data data leakage and sovereignty of intelligence and they're saying hey our company is going to choose maybe we're in a regulated industry... we need to have this on prem domestically and we'd liken an open-source version where we have a little bit more control.
- We run models for say Galaxos Smith Klein, which they wrote and developed.
- We run models for our partner in the UAE G42 and MBZ UAI... that are are their models that they designed.
- Increased demand for on-premise or domestically controlled AI solutions in regulated industries (finance, healthcare).
- Major IP holders (e.g., entertainment studios) seeking to leverage generative AI for content creation while maintaining strict control over their intellectual property.
- Advancements in multimodal and interactive world models that require tailored training on specific datasets.
- The development of more robust and user-friendly open-source frameworks for custom model training.
- The inherent challenges of balancing open-source accessibility with the need for IP protection.
- Competition from large, closed-source models that may offer superior performance or ease of use for certain applications.
- Difficulty in scaling custom model development and deployment for diverse IP holders.
- Potential for misuse of open-source models if not adequately governed.
- Analyze the market for enterprise-grade, customizable generative AI solutions.
- Evaluate the capabilities and limitations of leading open-source models compared to proprietary offerings.
- Assess the strategies of major IP holders (e.g., Disney, Hollywood studios) regarding AI integration and IP protection.
- Track the development of regulatory frameworks surrounding AI and data sovereignty.
Watchlist
- Cerebras's order backlog and delivery pace.
- Nvidia's GPU supply and demand dynamics.
- Adoption rates of custom silicon by hyperscalers (e.g., Amazon, Google).
- Growth of open-source AI model usage and community contributions.
- Partnerships between IP holders and AI model developers.
- Development of multimodal and action-prediction models.
Open Questions
- What is the long-term energy sustainability plan for the massive AI data center buildout?
- How will the competitive landscape between specialized AI hardware (Cerebras, Nvidia) and custom silicon evolve?
- What are the most effective strategies for IP holders to leverage generative AI while safeguarding their intellectual property?
- To what extent will open-source models achieve parity with frontier closed-source models for complex reasoning tasks?
- What are the key milestones in the development of AI for robotics and real-world interaction?
- How will governments and regulatory bodies balance AI innovation with safety and ethical concerns?
Key Topics & People
.jpg?width=300){kind=avatar}
CEO of OpenAI, referred to as an example of a top-tier founder.
.jpg?width=300){kind=avatar}
Proprietary AI models heavily controlled by their creators, existing alongside open models.
AI models with accessible source code, which Ek advocates for to drive innovation.
Cybersecurity company highlighted as critical for the inevitable rise in enterprise and government AI defense systems.
Critical infrastructure for AI compute, increasingly viewed as national security assets requiring massive energy generation.
CEO of Anthropic who attended the summit despite prior political objections, invited directly by Donald Trump.
.jpg?width=300){kind=avatar}
CEO of Nvidia who played a key role in drafting the AI safety accord's internal and external audit mechanisms.
.svg?width=300){kind=logo}
An AI paradigm where models iteratively self-correct and optimize based on a reward function.
Next-generation AI systems processing text, vision, and audio, driving the need for choice.
The principle that enterprises must maintain independent control over their AI models and data.
Artificial intelligence that is as smart as any human across a wide range of tasks.
The massive power requirements necessary to sustain the computing workloads for training and running AI models.
Valuable technological patents and trade secrets that must be protected from foreign theft.
Legendary film director exploring generative AI for storyboarding and creative visualization
An autonomous AI agent capable of sophisticated reasoning and task execution
Co-founder and CEO of Black Forest Labs, pioneer of latent diffusion models
Startup developing open-source multimodal visual and video AI models
Filmmaker noted for pioneering collaborative visual ideation and inspiring fan fiction
AI systems that predict physical interactions and simulate real-world actions
Algorithm used to compress natural data into efficient representations for AI generation
AI pioneer who correctly anticipated the need for supercomputing and AI safety
A distributed crypto project providing decentralized AI compute capacity
The computational process of running live AI models to generate reasoning and answers
The massive mobilization of capital and resources to build data centers globally
CEO and founder of Cerebras, a company pioneering AI inference chips