01 / Engineered

Engineered for what's coming.

The next generation of AI compute will require rack densities, power architectures, and cooling profiles most data centers cannot deliver. This is the brief on the reference design ISS Mining engineers for — what it specifies, why, and where it lives in the platform.

02 / The Trajectory

Power density is on a curve the grid was never built to deliver.

What the AI compute roadmap demands. What grid timelines actually deliver.

If you're evaluating where to deploy what's shipping next, you've watched the divergence happen in real time. The AI industry is not constrained by chips, capital, or talent. It is constrained by electricity. New compute generations are shipping every twelve months, each demanding substantially more power per rack than the last.

The utility grid was designed for distributed consumption — millions of homes and businesses drawing modest loads. AI compute is the opposite: concentrated, massive, growing exponentially.

These two curves are diverging.

~380 GW

ERCOT large-load queue

5–7 yr

Current grid interconnection timeline

600 kW

Per-rack demand arriving 2027

1 MW+

Per-rack projected for 2028

A seven-times increase in three years. If an operator files for grid interconnection today, they receive power around 2031–2032 — by which time the industry is shipping the generation after the one currently being designed for. Grid-dependent operators are not late to the next generation. They are structurally unable to deploy even the one shipping now.

03 / The Reference Design

Three specifications. Built for the curve.

What the design specifies, why, and what it enables.

Specification 01 / Native 800 VDC distribution
The power path next-generation racks require.
Gas + steamon-site gen
Medium-voltage AC
IGBT converter
800 VDCdistribution
Compute rack
Longer grid path
Grid
Step-down
AC distribution
48V conversion
Rack
What it specifies
Dual 800 VDC distribution buses with IGBT conversion as primary power delivery to AI compute racks. AC distribution retained for cooling and auxiliary loads.
Why
At megawatt-class densities, traditional low-voltage in-rack conversion consumes the rack — power shelves displace the compute they’re meant to feed. 800 VDC is the emerging industry reference for high-density AI deployments, and the published rack reference for next-generation compute mandates it.
What it enables
Direct compatibility with the rack architectures arriving 2027–2028, without retrofit. On-site generation feeds medium-voltage AC directly into IGBT converters that output 800 VDC to compute racks — the shortest, most efficient power path available.
Specification 02 / Hybrid liquid cooling
Liquid-dominant, designed for the form factors coming next.
Compute chip
Direct-to-chipliquid loop
CDU
Waste-heatrecovery
Absorptionchiller
cooling capacity returned · closed loop
What it specifies
Direct-to-chip liquid cooling for AI compute, with air cooling retained for auxiliary infrastructure. Facility-wide coolant distribution loop. Absorption chillers driven by recovered waste heat.
Why
Next-generation rack designs ship fanless with substantial airflow reduction vs. current generation. Megawatt-class racks eliminate air cooling entirely. Air-cooled colocation design is obsolete on a 24-month horizon.
What it enables
Scales from current rack densities through next-generation deployments without architectural replacement. Waste heat actively converts thermal load into cooling capacity — higher density improves facility efficiency rather than straining it.
Specification 03 / Four-layer power generation
On-site, diversified, sized to the site.
Organic Rankine cycle
Utility-scale solar PV
Combined-cycle steam recovery
Gas turbine baseload
Diversified stack · sized site by site
Battery storage
IT load
What it specifies
A diversified on-site generation stack — gas turbine baseload, combined-cycle steam recovery, utility-scale solar PV, and organic Rankine cycle waste-heat capture — backed by utility-scale battery storage. Sized site by site for the IT load it serves, with deliberate generation headroom for cooling overhead, N+1 redundancy, ERCOT export revenue, and density growth.
Why
Grid-dependent operators face 5–7 year interconnection queues. On-site generation eliminates that bottleneck and delivers power on the timeline AI infrastructure actually needs.
What it enables
Power delivery at platform pace, with bi-directional ERCOT interconnect for ancillary revenue and ultimate backup. Capacity headroom sized to support megawatt-class rack growth without filing for new generation infrastructure.

04 / Where it lives

Available when the deal calls for it.

The reference design is the engineering philosophy. The platform decides where to apply it.

The reference design isn't a campus. It's the engineering standard ISS Mining applies when a site advances through horizontal development with on-site generation. The four deal shapes determine where it lives.

Land & entitlement — included

Horizontal infrastructure — not included

Power generation — not included

Long-term operations — not included

01 / Hand-over after complexity navigation

ISS delivers a controlled, verified site. The buyer brings their own generation and infrastructure. The reference design isn't applied — the buyer engineers from here.

Land & entitlement — included

Horizontal infrastructure — included

Power generation — not included

Long-term operations — not included

02 / Hand-over at horizontal-ready

ISS completes the infrastructure backbone. The buyer brings power. The reference design isn't applied; the buyer's IPP makes the generation choice.

Land & entitlement — included

Horizontal infrastructure — included

Power generation — included

Long-term operations — not included

03 / Develop with power delivery

Reference design delivered

ISS brings on-site behind-the-meter generation built to the reference design — four-layer generation, 800 VDC distribution, hybrid liquid cooling. This is where the reference design is delivered.

Land & entitlement — included

Horizontal infrastructure — included

Power generation — included

Long-term operations — included

04 / ISS retains and operates

Year 2+

ISS holds and operates the site with the reference design as the engineering baseline.

05 / Engaging the design

How the engineering reaches your stack.

Three steps. The reference design is delivered to spec, not to template.

Step 01

Map the requirement

You bring the compute roadmap you’re underwriting against. We map it to the design layers — power, cooling, distribution, generation — and to the site stage that fits.

Step 02

Confirm the deal shape

The four deal shapes determine where the reference design lives. We confirm which one matches your stack ownership preference, your timeline, and your capital model — before any engineering is committed.

Step 03

Deliver the engineering

Once the deal shape is set, the design is delivered to the spec. Hand-over, horizontal-ready, fully powered, or operated — the engineering is the same standard.

Engaged with the platform, you deploy next-generation compute on the timeline AI actually moves on — while operators waiting on the grid are still pre-energization.

06 / Engage

Where to go from here.

Three paths. Match the one that fits your strategy.

01 / Acquire

Acquire a site

Take a controlled site at the funnel stage that fits you — as-is with the complexity navigated, horizontal-ready, fully powered, or operated.

02 / Partner

Partner with us

Capital partnerships, JV structures, anchor tenancies, and strategic co-development across the platform's deal flow.

Discuss a partnership

03 / Portfolio

See the full portfolio

Detailed site information across our controlled positions and pipeline.

See full portfolio