ALTi Global — Data Center & Vera Rubin Deployment Tracker

01 — Market Overview

AI Infrastructure Supercycle

The convergence of AI model scaling, sovereign compute mandates, and hyperscaler expansion is driving the largest infrastructure buildout in history.

$690B

2026E AI Capex

5,000+

Data Centers Global

45 GW

DC Power by 2030

$1T+

Infra Investment Need

Demand Drivers

AI Model Scaling — Training compute doubling every 5-6 months. GPT-5 class models require 50,000+ GPU clusters.
Sovereign Compute — 40+ nations launching national AI strategies requiring domestic GPU infrastructure.
Enterprise AI — Inference workloads growing 10x faster than training, driving distributed DC demand.
Data Gravity — Regulatory requirements (EU AI Act, GDPR) forcing data localization in-region.

Supply Constraints

Power — Grid capacity is the #1 bottleneck globally. Average time to secure 100MW+ allocation: 3-5 years.
Land — Permitted, utility-connected sites are scarce. Pre-zoned DC land commands 3-5x premiums.
Cooling — Liquid cooling required for next-gen GPU racks (120kW+/rack). Most existing sites can't retrofit.
Talent — Global shortage of DC engineers and operators, especially in emerging markets.

Sources: Futurum Group (Feb 2026); IoT Analytics (Nov 2025); Goldman Sachs (Dec 2025)

02 — Hyperscaler Capex

The $600B+ Capex Supercycle

Combined hyperscaler capital expenditure is projected to exceed $600B in 2026, with AI infrastructure representing the majority of new spend.

Microsoft

$120B+

Google

$80B

Amazon

$80B

Meta

$60B

Apple

$20B

Oracle / xAI

$30B

Sources: CreditSights (Nov 2025); IEEE ComSoc (Dec 2025); Introl (Jan 2026)

03 — NVIDIA Vera Rubin

Next-Gen GPU Architecture

The Vera Rubin NVL72 platform represents a generational leap in AI compute density, requiring purpose-built facilities with liquid cooling and high-density power delivery.

72 GPUs

Per NVL72 Rack

260 TB/s

Scale-Up Bandwidth

120+ kW

Per Rack Power

$3M+

Per Rack Cost

Key Specs

72 Vera Rubin Ultra GPUs
36 Vera CPUs (ARM-based)
NVLink 6 interconnect
HBM4 memory per GPU
Liquid cooling required

DC Requirements

120+ kW per rack density
Direct liquid cooling (DLC)
High-density power bus
Reinforced floor loading
Advanced fire suppression

Deployment Timeline

CES 2026: Architecture launch
H2 2026: Partner qualification
H1 2027: Volume shipments
H2 2027: Full-scale deployment
2028: Rubin Ultra refresh

Sources: NVIDIA (Jan 2026); NVIDIA Product Page; NVIDIA Dev Blog (Feb 2026)

04 — Global Mega-Sites

Operating & Planned DC Campuses

The largest data center campuses under development or operation globally, ranked by power capacity.

Campus	Operator	Location	Power	Status
Stargate (Abilene)	OpenAI / SoftBank / MGX	Texas, USA	5 GW	Announced
Project Dataville	Microsoft	Wisconsin, USA	2 GW	Planned
Colossus	xAI	Memphis, TN, USA	1 GW	Operating
Switch CITADEL	Switch	Nevada, USA	900 MW	Operating
Hamina	Google	Finland	600 MW	Expanding
Dublin Campus	Microsoft	Ireland	600 MW	Operating
Stargard Campus	ALTi Global	Poland	560 MW	Phase 1 Live
Ashburn Campus	Equinix / DLR	Virginia, USA	500+ MW	Operating
Johor Bahru	AirTrunk / YTL	Malaysia	500 MW	Planned
Jurong / Tuas	ST Telemedia / Equinix	Singapore	400 MW	Expanding

Sources: Data Center Knowledge (Mar 2026); DCD M&A Database; company filings

05 — Operator Landscape

Valuations & Key Metrics

Public and private data center operators ranked by enterprise value, with key financial metrics.

Company	Type	EV	LTM Rev	EV/Rev	EV/EBITDA
Equinix	Colo / Interconnection	$70B	$9.1B	7.7×	22.5×
CoreWeave	GPU Cloud / AI	$67.3B	$5.1B	13.2×	21.4×
Digital Realty	Wholesale / Colo	$55B	$6.0B	9.2×	15.7×
QTS (Blackstone)	Hyperscale	$10B	—	—	—
AirTrunk (Blackstone)	APAC Hyperscale	A$24B	—	—	—
Aligned DC (MGX/BR/GIP)	Hyperscale US	$40B	—	—	—

Sources: Public Comps (Jan 2026); Equinix Q4 PR (Feb 2026); DCD (Nov 2025)

06 — ALTi Global Position

CEE Infrastructure Moat

ALTi Global's purpose-built, sovereign-scale GPU data center infrastructure is positioned at the intersection of the three most powerful trends in AI infrastructure.

Power Secured

560 MW

Phase 1 + Phase 2 confirmed power allocation across Poland portfolio. 480 MW ENEA phased at Stargard.

GPU Capacity

147,456

Vera Rubin GPUs across 2,048 NVL72 racks. Purpose-built for next-gen liquid-cooled AI compute.

Revenue at Scale

$4.3B

Projected annual revenue at full deployment. 60%+ EBITDA margins on GPU-as-a-Service model.

Competitive Advantage

Pre-zoned DC land — Stargard is the only site in Poland with data center in the municipal spatial plan
Power moat — 480 MW phased ENEA allocation with 1 GW+ pathway, years ahead of competition
NVIDIA partnership — Delta 41-year partnership, NVIDIA-accredited installer confirmed
EU/NATO location — Full regulatory compliance, data sovereignty, NATO security umbrella
20-site pipeline — 1.7 – 4.6 GW total capacity across 10 voivodeships

Why CEE, Why Now

60% lower energy costs vs. Western Europe (Poland avg. €0.12/kWh vs. Ireland €0.30/kWh)
Available grid capacity — Post-industrial sites with HV connections Western Europe lacks
SEZ incentives — Up to 40% CapEx offset via Polish Investment Zone tax benefits
Data sovereignty demand — EU AI Act and GDPR requiring in-region compute
Northern climate — 7-8.5°C average = extended free cooling, lower PUE

← Back to Executive Summary Poland Portfolio →

Global Data Center &Vera Rubin Deployment Tracker

Global Data Center &
Vera Rubin Deployment Tracker