📊 Full opportunity report: The gigawatt gap. Why China is structurally positioned for AI power and the US is engineering around its grid. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

TL;DR

China is structurally positioned for AI power deployment due to its centralized planning, extensive renewable infrastructure, and high-voltage transmission network. The US leads in chip performance but faces grid and permitting constraints that limit gigawatt-scale AI data centers. The emerging gigawatt gap could reshape global AI dominance.

China’s AI infrastructure is rapidly scaling through centralized planning and massive renewable energy deployment, positioning it to close the gigawatt power gap with the United States, which faces grid and permitting constraints.

Current frontier AI data centers now require gigawatt-scale power, with Chinese efforts leveraging 45 ultra-high-voltage transmission projects spanning over 40,000 kilometers to transmit renewable energy from western hubs to eastern demand centers. In 2025, China added over 430 GW of wind and solar capacity—eight times the US’s additions—raising its total renewable capacity above 1.8 TW and overall capacity to nearly 3.9 TW. Despite Chinese AI chips performing at about 60% of NVIDIA’s H100 inference levels and lacking native FP8/FP4 support, the system-level advantage lies in China’s ability to substitute raw power for chip performance due to its extensive renewable infrastructure and centralized grid. Conversely, the US leads in chip performance and AI models but is constrained by a fragmented grid, regulatory hurdles, and slower renewable buildout, which limits the scale of its gigawatt-level data centers.

The Gigawatt Gap — Thorsten Meyer AI
GIGAWATT
● DISPATCH / MAY 2026
THORSTEN MEYER AI · AI ENERGY & INFRASTRUCTURE · § 01
ENERGY & INFRA · 01
US-CHINA · AI POWER STACK
Essay · Structural-Comparison Analysis · 2026-05-17

The gigawatt gap.
Why China is structurally
positioned for AI power
and the US is engineering
around its grid.

The US dominates AI on chips, infrastructure, models, and applications — except on the layer that physically runs them.
Frontier AI data centers now need 100 MW to start and 1–2 GW at full buildout. Meta Hyperion targets 5 GW; OpenAI Stargate 10 GW; AWS 12 GW. The US reaches this scale through behind-the-meter PPAs · off-grid gas · nuclear restarts · ERCOT regulatory arbitrage · because 2,300 GW are stuck in 5-year interconnection queues. China reaches it through the NDRC’s Eastern Data Western Compute initiative · 45 UHV projects · 40,000 km · 340 GW cross-regional capacity · routing demand to western hubs co-located with 430 GW of new wind+solar added in 2025 alone. Even though Huawei’s Ascend 910C runs at ~60% H100 inference perf, the system-level asymmetry inverts the comparison: US perf-per-watt advantage vs. China watts-without-bound advantage. The gap is constitutional, not technical.
3.89 TW
China total installed
power capacity end 2025
2,300 GW
US interconnection queue
5-year average wait
40K km
China UHV transmission
45 projects · 340 GW capacity
~60%
Ascend 910C inference perf
vs. H100 · compensated by watts
STARGATE 10 GW· HYPERION 5 GW· AWS 12 GW· MICROSOFT 2 GW/YR· 2,300 GW QUEUE· 5-YR WAIT· PJM $29→$329/MW-DAY· ON-SITE GAS +1,800%· CHINA 3.89 TW· 1.8 TW WIND+SOLAR· 430 GW ADDED 2025· 4 TRILLION KWH RENEWABLE· 40,000 KM UHV· 45 UHV PROJECTS· 340 GW CAPACITY· ASCEND 910C ~60% H100· CLOUDMATRIX 384 / 300 PFLOPS· HUAWEI 1M DIES 2025· DEEPSEEK ON H800s· NDRC MANDATE· STARGATE 10 GW· HYPERION 5 GW· AWS 12 GW· MICROSOFT 2 GW/YR· 2,300 GW QUEUE· 5-YR WAIT· PJM $29→$329/MW-DAY· ON-SITE GAS +1,800%· CHINA 3.89 TW· 1.8 TW WIND+SOLAR· 430 GW ADDED 2025· 4 TRILLION KWH RENEWABLE· 40,000 KM UHV· 45 UHV PROJECTS· 340 GW CAPACITY· ASCEND 910C ~60% H100· CLOUDMATRIX 384 / 300 PFLOPS· HUAWEI 1M DIES 2025· DEEPSEEK ON H800s· NDRC MANDATE·
FIG. 01 — THE GIGAWATT SCALE
What frontier AI infrastructure now requires
The unit of measure has shifted from megawatts to gigawatts in 24 months · the binding constraint with it
Starter site
100 MW
Single building
~500 MW
Training sweet spot
1–2 GW
Meta Hyperion
5 GW
Stargate target
10 GW
Stargate Abilene’s 1.2 GW peak is half the system peak of El Paso Electric (serving 465,000 customers). AWS Indiana’s 2.2 GW at full buildout = approximately half the residential electricity consumption of all Indiana households combined. The four largest US hyperscalers have committed ~$650B to AI infrastructure across 2025–2026. Capital is not the constraint. The rate at which transformers can be manufactured, transmission permitted, and generation interconnected is.
FIG. 02 — THE AMERICAN BOTTLENECK
2,300 GW stuck · five-year wait · PJM prices 10x
The capacity exists in the queue · it cannot reach commercial operation at the rate AI buildouts require
Capacity in
interconnection queue
2,300 GW
Approx. US total
installed capacity
~1.3 TW
Of 2000-2019 requests
built by end-2024
13%
2026 capacity from
on-site generation
30%
PJM capacity price
DY 2024-25 → 2026-27
$29→$329
Wait times have more than doubled in 15 years. Onsite gas generation capacity has grown ~1,800% since 2025. Stargate Abilene runs 300 MW of on-site simple-cycle gas turbines; Meta Hyperion is anchored on a $3.2B 2 GW combined-cycle gas plant with $550M shouldered by Louisiana residents; xAI Colossus 2 trucks gas turbines into suburban Memphis. The hyperscalers are not solving the grid problem. They are routing around it.
FIG. 03 — THE TWO POWER STACKS
Constitutional fragmentation vs. centralised mandate
The same gigawatt-scale problem · two structurally different state-architectures solving it
UNITED STATES · WORKAROUND STACK
Five layers · routing around the grid
L1
Behind-the-meter PPAs · TMI restart · Talen-Susquehanna · Microsoft-Chevron
L2
Off-grid gas turbines · xAI Colossus · Stargate Abilene 300 MW · Hyperion $3.2B plant
L3
On-site share scaling · 0% → 30% of new capacity in 12 months
L4
ERCOT regulatory arbitrage · Texas HB 1500 · independent of FERC · 2-3x faster
L5
Executive-order acceleration · DOE Section 403 · FERC PJM order · April 30 2026 deadline
CHINA · CENTRALISED STACK
One mandate · five aligned layers
L1
NDRC mandate (2022) · Eastern Data Western Compute · 8 hubs · 10 cluster sites
L2
UHV backbone · 45 projects · 40,000+ km · 340 GW cross-regional capacity
L3
Western renewable hubs · Guizhou · Ningxia · Inner Mongolia · Gansu · co-located
L4
State Grid + China Southern · unified transmission build · single operator
L5
PUE ≤1.25 mandate · 50 intelligent computing centers · 300 EFLOPS target 2025
The US coordination cost runs through Cleanview · RMI · FERC · DOE · 7 ISOs/RTOs · 50 state utility commissions · local zoning. In China the coordination cost is the NDRC’s planning meeting. This produces speed and scale at the cost of democratic legitimacy and local accountability — both costs are real, and both are routed back to consumers downstream.
FIG. 04 — THE RENEWABLE FOUNDATION
The asymmetry under the chip comparison
China’s renewable buildout operates at roughly 8x the US pace · this is the foundation everything else rests on
United States · 2025
36 GW
Wind + utility solar + distributed
solar additions 2025
~1.3 TW
Total installed power
generation capacity
368 GW
Operating wind + solar
installed base
~26%
Renewable share
of capacity
~8×
2025 capacity
add ratio
China · 2025
430+ GW
Wind + solar additions
2025 alone
3.89 TW
Total installed power
capacity end 2025
1.8 TW
Combined wind + solar
installed capacity
>60%
Renewable share
of capacity
Chinese renewable generation reached ~4 trillion kWh in 2025 — exceeding the entire EU-27 electricity consumption (3.8 trillion kWh). China’s single-day peak load (1.506 TW) is now higher than total US installed capacity. 2025 Chinese energy infrastructure investment: ~$500B across generation, grids, and energy security — roughly the same scale as the four-hyperscaler US AI infrastructure commitment, but spent on the foundation AI runs on rather than on AI itself.
FIG. 05 — THE ASYMMETRIC SUBSTITUTION
Perf-per-watt vs. watts-without-bound
Different binding constraints · per-chip comparisons miss the system-level inversion
UNITED STATES STACK
High perf
Low watts
Perf-per-watt advantage at the chip · grid-bounded at the system
Frontier chip
H100/H200/B200
FP precision
FP8 / FP4
Software stack
CUDA / PyTorch
Rack power
130+ kW NVL72
Binding constraint:
grid + transmission capacity
CHINA STACK
Lower perf
More watts
Watts-without-bound advantage at the system · chip-bounded per unit
Domestic chip
Ascend 910C ~60% H100
FP precision
No native FP8/FP4
Memory
HBM2E (older)
System scale
CloudMatrix 384 / 300 PFLOPS
Binding constraint:
chip performance / FP precision
Production scale: ~1M Huawei Ascend dies shipping in 2025 · ~2M in 2026 · Ascend 960 (Q4 2027) projected H200-comparable. DeepSeek V3/R1 trained on degraded H800s at ~1/10 the US comparable-model compute cost — the lesson is not that DeepSeek had better chips; it is that algorithmic efficiency plus power-throughput substitution can produce frontier-competitive models with constrained silicon. If Chinese chips are 60% as performant per-chip but Chinese power can deploy them at 2-3x density without grid constraint, the system-level capability approaches parity.
The US has perf-per-watt advantage. China has watts-without-bound advantage. These are asymmetric substitutes — not the same axis. When the perf-per-watt side is bounded by grid capacity and the watts-without-bound side is bounded by chip performance, the binding constraint differs.
Thorsten Meyer · The Gigawatt Gap · Energy & Infrastructure 01

Implications of the Gigawatt Power Divide in AI

This structural difference influences global AI leadership, as China’s ability to deploy power at scale through renewable infrastructure may enable it to operate larger, more energy-intensive AI data centers despite lower chip performance. The US’s fragmentation and permitting delays could become a ceiling, limiting future AI deployment at the largest scales. The outcome will shape which country maintains technological dominance in AI hardware and infrastructure over the next decade.

Amazon

gigawatt-scale data center power supply

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

China’s Centralized Infrastructure and US Grid Constraints

Historically, US AI infrastructure has been built around modular, megawatt-scale facilities constrained by local permitting, grid access, and transmission bottlenecks. In contrast, China’s approach involves large-scale, centralized planning, with the National Development and Reform Commission (NDRC) orchestrating extensive high-voltage transmission projects that connect renewable generation in western regions to demand centers in the east. This systemic difference is rooted in governance: China’s top-down planning versus the US’s fragmented federal and state jurisdictions. China’s rapid renewable expansion—adding 430 GW in 2025—supports its strategy of substituting raw power for chip performance, enabling deployment of less capable chips across vast, renewable-powered grids.

„The gigawatt gap does not stem from chip technology but from structural differences in infrastructure and governance between China and the US.“

— Thorsten Meyer

Extruded Cables for High-Voltage Direct-Current Transmission: Advances in Research and Development (IEEE Press Series on Power and Energy Systems)

Extruded Cables for High-Voltage Direct-Current Transmission: Advances in Research and Development (IEEE Press Series on Power and Energy Systems)

Used Book in Good Condition

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Uncertainties in Future Infrastructure and Policy Developments

It remains unclear whether the US will succeed in closing the gigawatt gap through efficiency improvements, regulatory reforms, or new infrastructure projects. The extent to which China’s renewable buildout and centralized planning can sustain its advantage over the coming years is also uncertain, especially considering potential geopolitical shifts and technological advances.

The BESS Book: A Cell to Grid Guide to Utility-Scale Battery Energy Storage Systems

The BESS Book: A Cell to Grid Guide to Utility-Scale Battery Energy Storage Systems

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Next Steps in AI Infrastructure Competition and Policy

Over the next 24 months, both countries are expected to accelerate infrastructure projects—China through its renewable and transmission expansion, and the US through regulatory reforms and grid upgrades. Monitoring policy changes, renewable deployment rates, and the scaling of gigawatt-level data centers will be key to assessing which country gains a sustained structural advantage in AI deployment.

On-site operation technology of ultra-high voltage transmission lines(Chinese Edition)

On-site operation technology of ultra-high voltage transmission lines(Chinese Edition)

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

Why does China’s renewable energy buildout matter for AI?

China’s extensive renewable capacity allows it to transmit large amounts of power over long distances, enabling the deployment of energy-intensive AI data centers despite lower chip performance. This infrastructure-centric approach shifts the competitive landscape.

What are the main barriers the US faces in scaling AI infrastructure?

The US faces grid bottlenecks, permitting delays, and regulatory fragmentation that limit the size and speed of gigawatt-scale data centers, constraining its ability to deploy energy-intensive AI infrastructure at scale.

Could US efficiency gains close the gigawatt gap?

While technological improvements in chips and data center efficiency could help, the fundamental structural constraints—permitting, grid access, and regulatory hurdles—pose persistent barriers that may not be fully overcome in the near term.

How might geopolitical factors influence this infrastructure race?

Geopolitical tensions could impact cross-border energy projects, supply chains for AI hardware, and international cooperation on renewable infrastructure, potentially altering the pace and scale of both countries’ AI capacity expansion.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

The Skills Marketplace Nobody Is Building Yet

A new open standard for AI skills is established, but a dedicated marketplace remains absent. This gap could shape AI ecosystem dominance in the coming year.

The Ghost Story Became a Forecast.

In May 2026, Clark’s recent essay reveals a bivalent forecast for AI development, shifting the narrative from speculation to a structured probability assessment.

Two Channels: How the Pentagon Just Split Frontier-AI Procurement in Half

The Pentagon announced a split in its AI procurement, placing Anthropic in a separate cybersecurity channel from other vendors, marking a strategic segmentation.

The Regulatory Vacuum.

Google’s May 11, 2026 disclosure of an AI-driven zero-day highlights a lack of regulatory frameworks for emerging AI vulnerabilities, raising global security concerns.