Share from 0:00
0:00
/

Hot Rocks and Moving Bottlenecks

Coffee Chats Episode 40, with Ian Nieboer & Graham Bain (Recorded September 14, 2026)

Recorded September 14, 2026.

Prices show where the bottleneck sits. Both halves of this conversation read the same signal: capital pays up for the scarce input it can lock in now.

The first half is the geothermal land market. Federal lease bids set records this year, $701 an acre in New Mexico in June and $1,207 an acre in Utah in August, where one parcel went for $4.2 million. Lease bids are one of the clearest reads on subsurface conviction, and depth is what makes them interesting. If hot dry rock runs continuous, one parcel could carry many stacked EGS benches, so a lease starts to look like an option on future heat. California has since tested that view. The October 27 sale leased 33,362 acres for $3.69 million, with a top bid of $304 an acre. Most winners bid as LLCs, so the serious players stay out of sight.

The second half is GPU rental prices. SemiAnalysis’s H100 one-year rental index rose about 40% between October 2025 and March 2026, and older chips are recontracting at higher rates. If power were the binding constraint on AI, those older chips would be the first unplugged. Their lives are extending instead, which points to silicon binding first. These constraints sit in narrow bands, and helium is on the list too. I expect power to bind again as fab capacity outruns powered shells over the next two years.

Listen for the 20-bench question at 06:21 and the rental chart at 10:54.

Sources: GPU rental chart from SemiAnalysis, “The Great GPU Shortage”. Lease results from the U.S. Department of the Interior and the Bureau of Land Management.

Episode Length:

17:14

Episode Summary:

Coffee Chats returns after a six-week break with two linked questions about scarcity. The first half covers the 2026 run of record federal geothermal lease sales, California’s next sale, and why land rights are the clearest read on where subsurface capital is taking positions, then tests how stacked EGS benches could change what a geothermal acre is worth. In the second half I ask whether power is really the binding constraint on AI, using GPU rental prices on older chipsets as evidence that silicon binds first today. Helium, a core chipmaking input, joins the list, and the conversation closes on a buildout that keeps finding its next bottleneck.

Topics Covered:

Record Geothermal Lease Sales:

New Mexico and Utah set per-acre records in 2026, and California announced its next sale.

Who Is Bidding:

Most bidders use shell companies, which makes the serious players hard to identify beyond a few names such as Ormat and Invenergy.

Stacked EGS Benches:

Continuous hot dry rock could support many development benches on one parcel, which changes the land value math.

Is Power the Binding Constraint for AI?:

GPU rental prices on older chipsets suggest demand for compute is outrunning available silicon.

Helium and Chipmaking:

More complex chips need more helium, and industry reporting now ranks chipmaking ahead of MRI as the largest use.

Key Takeaways:

Land Is the Lead Indicator:

Lease bids show money taking real positions ahead of proven geothermal technology.

Depth Sets the Upside:

The value of a geothermal lease depends on how many benches technology lets a developer reach.

Silicon Binds First Today:

Older GPUs are being recontracted at higher rates, which points to compute supply as the tighter limit (10:59).

Constraints Bind in Narrow Bands:

Relieve one bottleneck and the next appears quickly. I expect chip capacity to outrun powered shells, bringing the power constraint back in the years ahead (12:28).

Every Bottleneck Gets Found:

Power shells, turbine blades and helium all sit in the queue, and a fast-growing economy keeps working around each one (15:18).

Chapters:

00:00 Intro

00:07 Welcome back, Episode 40

02:10 Record geothermal lease bids and California’s next sale

05:16 Who is bidding: shell companies and a few known names

06:21 Stacked EGS benches and what a geothermal acre could be worth

09:24 What would you have to believe?

10:19 Is power really the binding constraint for AI?

10:59 Older GPUs are renting for more

12:28 Constraints that bind in narrow bands

13:17 Helium and chipmaking

15:18 Running into every bottleneck

Comments, questions or things I missed? Send me a note. I would love to hear from you. Thanks for listening!

Discussion about this video

User's avatar