Subvolts take Train big AI models cheaply using a global network of computers. May 25, 2026 - Developers
- Learn how to leverage distributed compute and blockchain for cost-effective AI model training.
- Business
- Discover a path to building custom AI models without the prohibitive costs of traditional hyperscale training.
This interview details a practical, cost-saving method for training large AI models that could broaden access beyond major tech players.
Training AI Models Without a Billion-Dollar Data Center | Steffen Cruz of Macrocosmos
Audio streams from Craig S. Smith. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Steffen Cruz of Macrocosmos discusses training large AI models using distributed computing on the BitTensor blockchain. His project, IOTA, leverages thousands of globally distributed devices and surplus energy. This approach aims to significantly reduce training costs, making custom model training accessible to organizations beyond major tech companies.
Imagine using many small computers worldwide to train a giant AI brain, instead of one super-powerful one. This method uses cheap, leftover computer power and blockchain to connect them, making AI training affordable for more people.
This episode explores a novel approach to democratizing AI model training. Cruz details how distributed compute, coordinated by blockchain, can overcome the massive hardware and cost barriers imposed by hyperscalers. It's relevant for businesses and developers seeking to build custom AI without massive upfront investment, potentially challenging the centralized model favored by frontier labs like Google and OpenAI.
- HostCraig S. Smith
- GuestSteffen Cruzco-founder and CTO, Macrocosmos
- Guest: Steffen Cruz, co-founder and CTO of Macrocosmos
- Title: Training AI Models Without a Billion-Dollar Data Center
- Show: Eye on AI
- Length: 47 minutes
- System capability: Reproduces baseline benchmark performance with distributed compute.
- Target model size: 70 billion parameters by mid-2025
Training a frontier AI model today requires hundreds of thousands of GPUs, months of compute time, and a budget that only a handful of companies on earth can afford. Steffen Cruz, co-founder and CTO of Macrocosmos, thinks that model is about to break, and he's spending his time building what comes next. His project IOTA, operating within the BitTensor blockchain ecosystem, uses distributed training to split large language models across thousands of devices located around the world, coordinated by blockchain, and powered by surplus cheap energy wherever it exists. After nine months of research, the system can reproduce baseline benchmark performance using what Cruz calls "wonky vegetables" - unreliable, churning, globally distributed compute - and turn it into something indistinguishable from centralized training if you use the right approach.
The conversation with Craig Smith covers the mechanics of how this actually works, why the blockchain's role is far narrower and more practical than most people assume, and why the Mac mini stockpiling trend creates an unexpected supply of distributed compute that can earn passive income when idle. Cruz's target: a 70 billion parameter model by mid-2025, trained at 10-20% of what it would cost through a hyperscaler, and aimed squarely at the legal firms, hospitals, and cash-strapped startups that have been waiting to train their own sovereign models but couldn't afford the price tag.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.
- GPUs
- Graphics Processing Units, specialized computer chips for parallel processing, essential for AI training.
- Powerful computer chips used for AI.
- hyperscaler
- A provider of cloud computing services that can scale to meet the demands of very large organizations, such as Amazon Web Services, Microsoft Azure, and Google Cloud.
- Major cloud companies like Amazon or Google that offer massive computing power.
- distributed training
- A machine learning technique where the training process is spread across multiple computing devices or nodes.
- Training an AI model using many computers at once.
- sovereign models
- AI models trained and controlled by an organization or nation, often for data privacy or strategic reasons.
- Custom AI models an organization fully controls.
55 min42 min51 min34 minUkraine's Secret Weapon: The Points System Winning the Drone War | Andrii Hrytseniuk
WithAndrii Hrytseniuk
Same guests, other shows
46 min#340 Steffen Cruz: Training AI Without Data Centres
WithSteffen Cruz
56 min
Questions people ask
- How does Macrocosmos train AI models differently?
- Macrocosmos uses distributed training, splitting large language models across thousands of devices worldwide coordinated by blockchain, powered by surplus energy.
- What is the goal of Macrocosmos's IOTA project?
- The project aims to train a 70 billion parameter model at 10-20% of hyperscaler costs, making custom AI accessible to legal firms, hospitals, and startups.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Craig S. Smith.