ExpertsFM20VC: 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 Lochmil
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch · October 3, 2026
5 Selected Lessons:
reasoning_pattern · 2 framework · 1 case · 1 heuristic · 1
Think like a mountaineer: contingency-stacked planning culture
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 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 Lochmil (October 3, 2026) · at 4:28 ▸
When this applies
when running any high-stakes operation where conditions can change rapidly and failure is costly
The takeaway
Build explicit fallback plans in advance and rank safety above speed or ambition, because the environment will deviate from plan A.
Why it works
Having plans B through D pre-built means when (not if) something breaks, the team pivots instead of freezing, and a safety-first hierarchy prevents irreversible mistakes.
One level up
Resilience is designed in before the crisis, not improvised during it.
Where it stops holding
This stops working when the cost of extensive contingency planning outweighs the velocity needed in a genuinely low-stakes, reversible environment.
Detail
procedure:
- Set a primary plan (plan A) with a clear goal
- Pre-build plan B, C, D for likely failure modes (weather, gear, partner illness, avalanche equivalents)
- Embed a safety-first mindset: safety first, second, third
- Accept that enduring hardship/arduous stretches is part of execution, and dig in rather than quit prematurely
rationale: Mountaineering teaches that getting to the top is optional, getting down is mandatory — survival and reliability outrank glory.
In their words
There's this saying in mountaineering that getting up to the top is optional, getting down is mandatory. We try to practice that safety culture here at Crusoe and really try to avoid any sort of physical injuries or challenges with digital attacks on the platform as well.
Counter misinformation with local, data-backed community impact framing
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 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 Lochmil (October 3, 2026) · at 23:17 ▸ · at 25:52 ▸ · at 27:59 ▸
When this applies
when a large infrastructure or industrial project faces local political resistance driven by emotional misinformation
The takeaway
Address local political resistance with concrete, locally-specific data on the exact concerns raised (jobs, costs, environment), not abstract macro arguments.
Why it works
Since politics is hyper-local and driven by concrete worries (jobs, energy costs, water, pollution), countering emotional narratives with specific local data (water-use-per-building, tax-revenue-share, energy price trend data) converts the debate from sentiment to verifiable fact.
One level up
Emotional opposition to large projects is best met with hyper-local factual transparency rather than top-down messaging.
Where it stops holding
This stops working when the data itself is unfavorable (e.g., a project genuinely does raise costs or strain resources locally) or when opposition is driven by non-addressable concerns like job displacement anxiety.
Detail
approach: Identify exactly what locals fear (jobs, energy costs, water usage, pollution, community benefit) and answer each with specific data from your own projects.
signals:
- Closed-loop water architecture reduces usage to ~10 households/year for a 140MW facility
- Markets with data center investment saw energy prices fall due to amortized transmission/distribution infrastructure
- Tax revenue: Crusoe will be >1/3 of Taylor County tax revenue, doubling school funding
rationale: Politics are local. And what people care about is: are there jobs being created? Is this going to drive up my energy costs? Is this going to take all of my water?
In their words
the number of people that have told me that data centers use all the water in the world is crazy. It's also just false.
Most moats are illusory in fast-moving technology — speed and adaptability substitute
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 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 Lochmil (October 3, 2026) · at 53:15 ▸
When this applies
when evaluating whether a company's competitive advantage is durable in a rapidly advancing technology sector
The takeaway
In fast-moving technology markets, assume most claimed moats are temporary, and treat organizational speed/adaptability as the actual durable advantage, not any specific structural barrier.
Why it works
During periods of accelerating technological capability, any specific structural advantage (technical, product, positional) is likely to be eroded or leapfrogged quickly, so the real durable advantage becomes the organizational capacity to move fast and adapt to new conditions ('the ever-changing chessboard').
One level up
The faster a field's technological frontier moves, the shorter the half-life of any specific competitive moat, shifting durable advantage toward adaptive capacity itself.
Where it stops holding
This stops working in slow-moving, regulated, or network-effect-dominated markets where structural moats (regulatory licenses, network effects, switching costs) genuinely persist for long periods.
Detail
approach: Question whether a claimed 'moat' is actually structural or just currently uncontested; assume erosion is likely; invest in organizational speed and adaptability as the real defensible asset.
signals:
- Accelerating technological progress in the sector
- Capabilities/models changing the competitive landscape repeatedly
- Everything 'starts off as a moatless wrapper' until value is built on top
rationale: I think what I've changed my mind is like most moats are an illusion. Most moats don't exist. Most of them are ephemeral.
In their words
I think what I've changed my mind is like most moats are an illusion. Most moats don't exist.
Energy and skilled labor are the true AI data center bottlenecks
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 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 Lochmil (October 3, 2026) · at 20:53 ▸
When this applies
when diagnosing the real bottleneck in any capital-intensive physical buildout that looks compute- or capital-constrained on the surface
The takeaway
Identify the physical/human resource constraint underlying a capital-intensive buildout, since capital alone cannot solve for scarce energy or scarce skilled labor overlapping in the same location.
Why it works
The limiting resource isn't the chip or even capital, but whether there is a place to plug the chip in (power availability) and enough skilled tradespeople to build that place in a specific location simultaneously.
One level up
In physical infrastructure races, the bottleneck is usually not the headline asset (chips) but the enabling substrate (power, labor, permits) around it.
Where it stops holding
This stops working in regions/scenarios where power and skilled labor are abundant and the actual constraint shifts elsewhere (e.g., permitting, capital, or chip supply itself).
Detail
rule: Treat energy availability and skilled trade labor as the binding constraints for AI infrastructure buildout, not GPU supply alone.
threshold: Supply constraint manifests as 'not places to plug in GPUs' — i.e., energized, buildable sites with available skilled labor overlap.
rationale: Power that's available for compute is a key bottleneck... there are a finite amount of skilled trade workers, electricians, welders, plumbers, construction workers.
In their words
Energy is definitely a key constraint. Power that's available for compute is a key bottleneck. Alongside that, labor is a very key bottleneck. There are a finite amount of skilled trade workers, electricians, welders, plumbers, construction workers in the United States.
Bitcoin mining's cost-stripping evolution as a template for AI data centers
Source
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch — 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 Lochmil (October 3, 2026) · at 14:24 ▸ · at 16:34 ▸
When this applies
when a new technology's infrastructure requirements are assumed to mirror an older paradigm's, but could actually be radically simplified
The takeaway
Question inherited infrastructure assumptions by examining what the new workload actually requires, not what the old workload required.
Why it works
Bitcoin mining discovered that extreme reliability (five nines) wasn't needed, and stripping redundancy cut data center costs by ~98%, shortening payback; this created a prior that AI infrastructure could similarly shed assumed requirements (centralization, high redundancy) once its actual workload characteristics were understood.
One level up
Technological paradigm shifts create opportunities to strip unnecessary cost/complexity inherited from prior-generation assumptions.
Where it stops holding
This stops working when the new workload genuinely does require the inherited reliability/centralization standard, unlike the prior analog.
Detail
situation: Bitcoin mining data centers evolved from hobbyist laptops to leased five-nines colocation facilities to realizing redundancy wasn't needed.
decision: Crusoe bet that AI compute, like Bitcoin mining, didn't require centralized, high-redundancy facilities and could instead be distributed to cheap/abundant energy sources.
outcome: Crusoe built distributed, lower-redundancy AI-focused data centers ahead of ChatGPT's launch, positioning them well for the subsequent demand surge.
lesson: Examine whether inherited infrastructure norms (reliability tiers, centralization) actually apply to a new workload before building to the old standard.
In their words
They were able to eliminate something like 98% of the total data center cost with like these like chicken coop type data centers, very, very low cost, no redundancy, but ultimately brought down the payback period for this investment in Bitcoin mining. And we felt always that something similar was going to play out in AI.