20VC: The Future of Datacentres: What You Need to Know | Why...
The Twenty Minute VC (20VC)Full Title
20VC: The Future of Datacentres: What You Need to Know | Why Everyone Gets GPU Depreciation and AI's Energy Costs Wrong | Who Really Makes Money From AI & Why Most Moats Don't Exist with Chase Lochmiller
Summary
This episode features Chase Lochmiller of Crusoe Energy discussing the infrastructure needs for AI, challenging common narratives about energy costs and GPU depreciation.
The discussion highlights Crusoe's vertically integrated approach to building data centers and providing compute, emphasizing cost-effectiveness and adaptability in the rapidly evolving AI landscape.
Key Points
- The company culture at Crusoe incorporates principles of mountaineering, emphasizing preparedness for change, resilience, endurance, and a paramount focus on safety, which translates to robust business operations.
- Having achieved financial success prior to founding Crusoe empowered Lochmiller to take bigger risks and pursue ambitious goals, viewing entrepreneurship as a path to wealth after realizing the slow pace of academic discovery.
- Crusoe's initial strategy for monetizing waste energy with Bitcoin was a stepping stone to their core mission of building an AI platform, demonstrating strategic pivots driven by market evolution and technological advancements like the launch of ChatGPT.
- The increasing power density of GPUs necessitates a rethinking of data center design, shifting from centralized hubs to distributed models where low-cost, abundant energy is available, allowing for greater geographic flexibility.
- Crusoe's vertical integration, including in-house electrical manufacturing, allows them to overcome supply chain bottlenecks, such as long lead times for critical components, and provides a unique perspective on end-to-end cost structures.
- Data center development can significantly benefit local communities through job creation, economic stimulus, and substantial tax revenue, countering negative perceptions about their environmental impact.
- The energy price narrative is often misconstrued, as large data center investments can catalyze new energy generation and, in turn, lower energy costs for communities.
- The AI compute market is likened to the oil and gas industry's value chain, with Crusoe aiming to become an "AI super major" by being vertically integrated across data centers, GPUs, and services.
- The most profitable layer in the current AI infrastructure market is managed compute clusters due to extreme supply shortages, driven by take-or-pay contracts that ensure revenue regardless of immediate usage.
- GPU depreciation is often underestimated, as the ingenuity of application developers continues to find value in older hardware, extending its useful life beyond traditional depreciation cycles.
- Forecasting AI demand is challenging due to extended infrastructure build-out timelines, prompting Crusoe's focus on modular, manufactured data centers for more just-in-time deployment.
- The future of AI infrastructure is seen as a blend of open-source and closed-source models, catering to diverse needs like data sovereignty and cutting-edge capabilities, with a move towards domain-specific solutions.
- The "lowest cost producer of intelligence wins" principle is driven by optimizing metrics like dollars per token, throughput, and latency, with GPU utilization and efficient memory management being critical.
- Most perceived "moats" in business are temporary due to rapid technological advancement; adaptability and speed are more crucial for sustained success.
Conclusion
The AI infrastructure landscape is rapidly evolving, with energy emerging as a critical bottleneck, driving innovation in distributed data center solutions.
Vertical integration and adaptability are key competitive advantages, allowing companies like Crusoe to navigate supply chain challenges and capitalize on emerging market opportunities.
The narrative around data centers and AI's impact is often misinformed; these technologies are driving significant economic growth and job creation, with a focus on sustainable and community-beneficial development.
Discussion Topics
- How can the AI industry effectively communicate the positive economic and community impacts of data center development to counter misinformation and build public trust?
- Given the rapid evolution of AI technology, what are the most crucial factors for companies to consider when forecasting long-term demand for compute and infrastructure?
- In an era of accelerating technological progress, how can businesses build sustainable competitive advantages beyond simply being first-movers or early adopters?
Key Terms
- GPU Depreciation
- The decrease in value of Graphics Processing Units over time due to obsolescence or wear and tear.
- AI Super Major
- A term used to describe a highly integrated company in the AI infrastructure space, analogous to "super majors" in the oil and gas industry, covering the entire value chain from resource acquisition to service delivery.
- Take or Pay
- A contractual clause where a buyer commits to paying for a service or commodity, whether they use it or not, ensuring revenue for the supplier.
- NVMe
- Non-Volatile Memory Express, a high-speed interface protocol for accessing solid-state drives (SSDs) connected via a PCI Express (PCIe) bus.
- HBM
- High Bandwidth Memory, a type of RAM that has higher bandwidth and lower power consumption than traditional DRAM.
- KB Cache
- Key-Value Cache, a memory structure used in AI models to store and quickly retrieve frequently accessed data, improving inference speed.
- LLM
- Large Language Model, a type of AI model trained on massive amounts of text data, capable of generating human-like text, translating languages, writing different kinds of creative content, and answering your questions in an informative way.
Timeline
Chase Lochmiller discusses the lessons learned from mountaineering, including preparedness, resilience, and safety, and how these are integrated into Crusoe Energy's company culture.
Lochmiller explains his personal shift from academic pursuits to quantitative finance and then entrepreneurship, driven by a desire for faster-paced work and wealth creation after losing his initial sense of purpose.
Lochmiller clarifies that Crusoe's focus on building an AI platform was always the core goal, with Bitcoin monetization being a means to an end, a strategy informed by a probabilistic view of the future and the impact of events like the ChatGPT launch.
The transcript details the evolution of Bitcoin mining data centers, moving from basic setups to highly efficient, low-cost facilities, and the expectation that a similar pattern would emerge in AI infrastructure.
Lochmiller discusses the critical role of energy in scaling AI and how Crusoe's strategy is to distribute AI infrastructure where energy is abundant and low-cost, rather than concentrating in traditional hubs.
The discussion addresses data center supply constraints, noting that the primary bottleneck is the lack of places to plug in GPUs, and how Crusoe's vertical integration helps navigate these challenges.
Lochmiller provides an example of overcoming a 100-week lead time for power distribution centers by building them in-house within 28 weeks, showcasing the benefit of vertical integration.
The shift from scaling AI infrastructure for training to serving inference is discussed, emphasizing the "time to token" as a critical metric for deploying smaller, more agile compute clusters.
The major supply constraints for AI today are identified as energy and labor, with the difficulty of accessing skilled tradespeople being a significant challenge in the US.
Lochmiller argues that policy and regulation in the US, while needing navigation, are not insurmountable problems for data center development, and emphasizes the importance of good, thoughtful policies.
The discussion challenges misinformation about data centers' impact, particularly regarding water usage, highlighting that modern AI facilities use minimal water, and energy prices typically decrease for local communities due to increased investment.
Lochmiller counters the narrative of AI replacing jobs, arguing that data centers are driving a resurgence in the US economy, creating significant employment in skilled trades and blue-collar sectors.
The economics of providing GPUs and compute are explored, viewing compute as a tradable commodity and discussing payback periods through a portfolio of contracts with varying margins and risks.
Crusoe aims to be an "AI super major," vertically integrated across data centers, GPUs, and services, similar to traditional oil and gas supermajors, to manage margin fluctuations across the value chain.
The highest margin layer currently is managed GPUs due to supply shortages, with take-or-pay contracts providing revenue stability.
Lochmiller addresses GPU depreciation, stating that while the industry standard is a six-year cycle, Crusoe's strategy of offering managed AI services aims to extend this by finding longer-term monetization opportunities.
Lochmiller explains that forecasting AI demand is based on customer conversations and that Crusoe's modular data centers aim to reduce the time to deliver AI infrastructure, especially for startups.
Lochmiller expresses confidence in the future of AI, viewing it not as uncertain but as a certainty of profound change and productivity gains, citing AI's ability to solve complex problems.
The unit of value in managed inference is dollars per token, but efficiency metrics like throughput, latency, and GPU utilization are also critical, especially in managing memory and cache for optimal performance.
The key insight often missed about managed inference providers is that not all are equal; specialized providers and efficient infrastructure management are crucial for optimal performance.
The discussion touches on the balance between open-source and closed-source models, with open-source being valuable for data sovereignty and custom model development, but closed-source frontier models commanding higher spending.
The future of managed inference involves providing companies with easy access to their own intelligence, hosting custom models, and abstracting away infrastructure complexity.
Lochmiller shares advice on balancing work and fatherhood, emphasizing prioritization, spending quality time with children, and being present during dedicated family moments.
Lochmiller believes data centers are beneficial assets for communities, providing economic growth, energy solutions, and vital investments in local services, challenging the negative narrative surrounding them.
Lochmiller has changed his mind on the concept of business moats, now believing most are ephemeral and that adaptability and speed are more critical in an era of rapid technological advancement.
Crusoe's path to becoming a public company is seen as a likely outcome, given the capital-intensive nature of building AI infrastructure and the advantages of accessing scaled capital resources.
Lochmiller expresses admiration for Michael Dell, viewing him as an inspirational figure for his business acumen and family values.
Lochmiller reflects that the 2018 version of himself would find today's scale of Crusoe's AI cloud platform and the realization of the energy-first narrative almost unbelievable.
Episode Details
- Podcast
- The Twenty Minute VC (20VC)
- Episode
- 20VC: The Future of Datacentres: What You Need to Know | Why Everyone Gets GPU Depreciation and AI's Energy Costs Wrong | Who Really Makes Money From AI & Why Most Moats Don't Exist with Chase Lochmiller
- Official Link
- https://www.thetwentyminutevc.com/
- Published
- October 3, 2026