The world is too loud. Read what matters.

The Twenty Minute VC

Data Centre Prices Don't Rise—the Data Shows Local Rates Fall

Crusoe founder counters with data: data centre construction catalyses new generating capacity, spreading additional megawatts across the same transmission network. Community electricity prices fall as a result, not rise.

Data centresComputing powerEnergyAI infrastructureChip depreciationMoats

The video won't play here. Listen to the audio instead:

From power constraints to chip depreciation to vanishing moats, Crusoe's founder dissects AI infrastructure's true cost structure with concrete data—information-dense, little filler.

The argument · tap a timestamp to hear it

12:15

Bitcoin was never a pivot, just a stepping stone

Chase's goal from the start was an AI platform, but early capital went mostly into monetising stranded energy through Bitcoin mining—Bitcoin was simply the best energy-monetisation engine available then. The AI platform ran in parallel internally the whole time; only after ChatGPT launched did he rebalance resources toward AI infrastructure. What looked to outsiders like a business pivot was internally just a reordering of priorities.

— Chase Lochmiller
15:23

Bitcoin mining's cost curve forecasts AI data centre economics

Bitcoin mining began with laptop hobbyists, graduated to professional facilities with five-nines reliability, then evolved into almost-no-redundancy cage-style buildings that slashed data centre costs by 98%. Chase expects AI infrastructure to follow the same path—no need to cluster in traditional hubs like Virginia, since inference latency is dominated by in-facility compute, not data transmission. Compute can scatter to regions with cheaper, more abundant power.

— Chase Lochmiller
18:34

Vertical integration compressed a 100-week bottleneck to 28 weeks

Abilene's first two buildings top 200 megawatts; Crusoe promised one year. Of 34 competing data centre developers, the fastest bid was two and a half years. The critical bottleneck was equipment like switchgear. External suppliers quoted 100-week lead times; Crusoe made its own electrical equipment instead, delivering in 28 weeks. He stresses this wasn't about extracting manufacturing margins (they're thin) but breaking the supply-chain chokehold and hitting the schedule.

— Chase Lochmiller
24:55

Data centres barely use water—local electricity prices fall

Chase counters two widespread claims: modern AI facilities use closed-loop cooling, so a 140-megawatt building consumes about as much water annually as ten typical homes. Data centre investment catalyses new generation capacity, spreading more megawatts across the same transmission network. His data shows community electricity prices falling, not rising. The real cost is construction-phase traffic and dust—both temporary.

— Chase Lochmiller
28:00

Half of announced data centre projects never get built

Asked what fraction of announced data centre projects never reach operation, Chase says 50%. The real bottleneck isn't technology—it's planning-phase work: land acquisition, permitting, signing interconnection agreements with the grid. A landowner refusing to sell, a utility refusing to sign—any broken link stalls the whole project.

— Chase Lochmiller
34:12

Three markets: sell facilities, sell GPUs, sell tokens

Chase likens Crusoe's business to oil's upstream-midstream-downstream—drilling, pipelines, refining and retail. He wants to build an AI supermajor, vertically integrated like ExxonMobil. ExxonMobil doesn't hedge oil prices because when crude falls, upstream margins shrink but downstream margins (petrol, plastics) rise—the whole chain self-hedges. Right now the thinnest margins are in managed GPU clusters. Reason: supply is still too scarce.

— Chase Lochmiller
38:17

Chip buyers underestimate application developers' ingenuity

Crusoe depreciates GPUs on a six-year cycle. When it bought Hopper aggressively in 2023, the market reaction was ‘who knows whether these will be worth anything in three years’. Three years on, Hopper's utilisation rates are actually higher than when it launched. His lesson: early-stage financing pressures everyone to chase fast payback and high debt coverage, underestimating developers' ability to turn compute into valuable services.

— Chase Lochmiller
54:52

In the past year he concluded: most moats are illusions

Asked what changed in the past 12 months, Chase points to his view of long-term competitive moats—most are illusions, temporary, especially as model capability iterates rapidly. The key is moving fast and adapting to a shifting board. Everything starts as ‘a shell with no moat’; moats are built layer by layer as you pile on value.

— Chase Lochmiller

In their own words · checked verbatim

there's this saying in mountaineering that getting, getting up to the top is optional, getting down is mandatory.

Chase Lochmiller6:06

So absent having a purpose, I was like, man, I guess I should get rich. Like that's maybe the next best thing to like having a monastic calling that you're devoted to.

Chase Lochmiller10:10

we felt like the infrastructure to support AI wasn't going to be centralized. It was going to be distributed where energy was low cost and abundant.

Chase Lochmiller17:32

The data has shown that energy prices actually come down. It's actually the opposite of the narrative that's being told.

Chase Lochmiller25:59

there's really like three products that we ultimately sell to customers where we're making money. We can sell data centers, we can sell GPUs, and we can sell tokens.

Chase Lochmiller34:12

Well, you know, the GPU is actually the most valuable thing in the entire data center, right?

Chase Lochmiller45:36

I think what I've changed my mind on is like most motes are a illusion. Most motes don't exist.

Chase Lochmiller54:52

Figures

Crusoe latest funding round$3.9 billion (valuation $30.9 billion)1:01
Abilene first phase construction timeline1 year (industry fastest competitive bid: 2.5 years)18:34
Switchgear delivery time: external vs internal100 weeks vs 28 weeks18:34
140-megawatt facility annual water useequivalent to 10 typical homes24:55
Fraction of announced data centres never built50%28:00
GPU depreciation cycle6 years38:17
Crusoe share of Taylor County tax revenueover one-third27:00
Tax contribution to local school systemmore than doubled27:00

Glossary

KV cache
Cached key-value pairs of attention weights computed during inference, reused to skip recalculation—critical for throughput and latency.
Take-or-pay contract
Customer commits to pay a fixed capacity fee regardless of actual consumption.
Idiot index
Elon Musk's concept: finished-product price relative to raw material cost, measuring manufacturing efficiency.
RDMA fabric
Remote Direct Memory Access interconnect enabling GPU clusters to bypass CPU and access memory directly for high-speed communication.

How to listen

Who it's for

Founders, investors and engineers focused on AI infrastructure, data centre investment or energy policy—anyone seeking real cost structures rather than headline narratives.

Skip

The opening mountain-climbing metaphor and closing rapid-fire questions about child-rearing—lower business information density.