NNuggets
BookmarksCollections
  • About Us
  • Terms of use
  • Privacy policy
  • Disclaimer
  • Copyright & Takedown Policy
  • Community Guidelines
  • Cookie Policy
  • Contact

© 2026 Nuggets

NuggetsMarket PulseCollections

On this page

Speakers & Credentials

  • Speakers & Credentials
  • 1. Executive Summary
  • 2. Chronological Table of Contents
  • 3. Detailed Thematic Summary
  • The Magnitude of Hyperscaler Budgets and Power Demand [00:00:55]
  • Agentic Systems as Token Furnaces [00:03:20]
  • The 6 P's Framework and The Severe Human Capital Constraint [00:05:19]
  • Overcoming Inertia: Behind the Meter Solutions & Political Affordability [00:07:59]
  • The 2030 Vision: A "Yes, And" Energy Mix [00:12:26]
  • The Reference Vault
  • 4. Data & Figures
  • 5. Core Frameworks & Mental Models
  • 6. Anecdotes
  • 7. References & Recommendations

On this page

  • Speakers & Credentials
  • 1. Executive Summary
  • 2. Chronological Table of Contents
  • 3. Detailed Thematic Summary
  • The Magnitude of Hyperscaler Budgets and Power Demand [00:00:55]
  • Agentic Systems as Token Furnaces [00:03:20]
  • The 6 P's Framework and The Severe Human Capital Constraint [00:05:19]
  • Overcoming Inertia: Behind the Meter Solutions & Political Affordability [00:07:59]
  • The 2030 Vision: A "Yes, And" Energy Mix [00:12:26]
  • The Reference Vault
  • 4. Data & Figures
  • 5. Core Frameworks & Mental Models
  • 6. Anecdotes
  • 7. References & Recommendations
Technology/April 7, 2026/9 min read/youtu.be

AI Exchanges: Power Problems?

Source
Source
Watch on YouTube ↗

"We have seen the hyperscalers’ capital budgets and R&D budgets in 2026 and 2027 combined increase by more than $300 billion." - Brian Singer [00:01:23]

"If you just observe the way that these agentic systems work, they are furnaces for tokens." - George Lee [00:03:30]

"We're now assuming about 220 percent global growth in AI... 2030 versus 2023... that would be like adding another top ten consuming country to the mix." - []

References

  1. Original source (youtu.be)

Disclaimer: Orignal content owned by or sourced from third parties. It does not represent the views of 'Nuggets' platform or it's team. AI is used extensively across this platform including for summaries. Accuracy is not guaranteed, there can be mistakes. Any info or content on this platform is not a financial, legal, or investment advice. Do your own research. Refer for complete disclosures:- Terms of Use · Full Disclaimer

Related nuggets

Jun 2, 2026

AI Is Escaping the Screen | 01 Jun 2026 | Coatue

Coatue : AI is entering a new phase: moving beyond digital tools and into fully autonomous systems operating in the physical world. From advanced manufacturing and surgical robotics to robots in the home, the next wave of innovation will b…

Jun 2, 2026

Kalshi Monthly Volume - Politics ($M) | Chart of the Day | Coatue

Coatue: Kalshi's political volume has scaled dramatically, and the American Power Index KPOW is what that scale enables: a single number gauge of the current balance of political power and where markets expect it to move, which Kalshi bill…

Jun 2, 2026

The BlackBerry Problem |18 May 2026 | The Mistakes Series | Malcolm Gladwell's Revisionist History

"My mistake and naivity was to think that people are were with me so you're flying around the world you're trying to get people on side and you think they're on side but they're not mhm mhm and you get blindsight" Jim Balsillie 00:01:34 ht…

Jun 2, 2026

Partnership Perspectives: Network International | 2 Jun 2026 | Brookfield Perspectives

Actions

Reading

Published
April 7, 2026
Read time
9 min read
Progress0%
Brian Singer
00:04:03

"Of all of these... we're most worried from a constraint perspective on the people side." - Brian Singer [00:05:48]

"To use an improv comedy phrase, we are in a 'yes and' environment here. Not a 'no or' environment." - Brian Singer [00:14:30]


Speakers & Credentials

  • Allison Nathan: Co-host of Goldman Sachs Exchanges.
  • George Lee: Co-head of the Goldman Sachs Global Institute and co-host of the series exploring the rise of AI.
  • Brian Singer: Head of GS SUSTAIN in Goldman Sachs Research, specializing in the investment opportunities and constraints around the AI energy infrastructure build-out.

1. Executive Summary

  • The massive capital deployment by hyperscalers is triggering unprecedented growth in power demand to support the physical infrastructure of artificial intelligence.
  • With hyperscaler R&D and capital budgets expanding by over $300 billion for the 2026-2027 window, global data center power demand is projected to grow by 220% by 2030, effectively adding the equivalent of the world's 6th largest country to the grid.
  • Despite concerns over supply chain parts and rising energy costs, the most severe constraint limiting growth is human capital—specifically, a massive shortage of 500,000 infrastructure workers and highly trained electricians in the US.
  • To bridge immediate power shortfalls, hyperscalers are adopting "behind-the-meter" natural gas solutions, as grid upgrades will take 3 to 5 years to materialize.
  • Moving toward 2030 and beyond, the energy solution will not be singular; it will rely on a "yes, and" blend of 60% thermal (natural gas) and 40% renewables, eventually scaling up nuclear capabilities as parts and policies catch up.

2. Chronological Table of Contents

  • [00:00:00] Introduction & Welcome [00:00:00]
  • 00:00:55 Hyperscaler Spending & Power Demand Growth [00:00:55]
  • 00:02:08 The Shale Revolution Analogy & Supply/Demand Imbalance [00:02:08]
  • 00:03:20 Driver of Demand: Agentic Systems & Token Furnaces [00:03:20]
  • 00:05:19 The 6 P's Framework for AI Power Growth [00:05:19]
  • 00:06:12 The People Constraint: Labor Shortages & Grid Upgrades [00:06:12]
  • 00:07:59 Bridging the Gap: Behind the Meter Solutions [00:07:59]
  • 00:09:16 Policy, Politics, and Affordability Concerns [00:09:16]
  • 00:12:26 The 2030 Vision: Future Energy Mix [00:12:26]

3. Detailed Thematic Summary

The Magnitude of Hyperscaler Budgets and Power Demand [00:00:55]

  • The fundamental driver of physical resource consumption is the aggressive spending by tech giants; capital and R&D budgets for the 2026-2027 period have increased by a staggering $300 billion [00:01:23].
  • This massive capital outlay is destined to "trickle down" heavily into power consumption and broad infrastructure enablement [00:01:40].
  • Drawing a parallel to the Shale Revolution, experts warn of a cycle consisting of an appraisal phase, massive capital spending, and a potential future cyclical downturn [00:02:08].
  • However, unlike the late stages of shale, the current AI landscape is deeply supply/demand imbalanced; product is not oversupplied, innovators are maintaining returns, and financial flexibility is intact [00:02:31].

Agentic Systems as Token Furnaces [00:03:20]

  • Demand growth continues to surprise analysts due to fundamental shifts in how AI is deployed—specifically, the rise of agentic systems [00:03:30].
  • These agentic machine-to-machine systems are heavily resource-intensive, described as "furnaces for tokens," which will ultimately dwarf simple consumer-to-chatbot traffic [00:03:49].
  • Driven by this exponential compute usage, Goldman Sachs research significantly increased its global power demand forecast for AI and data centers from its previous estimate of 175% up to 220% global growth by 2030 compared to 2023 [00:04:03].
  • To contextualize the 220% growth, the resulting energy draw is equivalent to adding an entirely new top-ten consuming nation—ranking roughly number six on the global power consumption list [00:04:19].
  • Previously assumed to be relatively low-energy, the inference phase is becoming significantly more energy-intensive per server as newer models deploy [00:04:39].

The 6 P's Framework and The Severe Human Capital Constraint [00:05:19]

  • To evaluate AI growth metrics, analysts utilize the 6 P's Framework: Pervasiveness, Productivity, Price, Policy, Parts, and People [00:05:32].
  • George Lee notes that three additional P's are becoming highly relevant: Pricing, Politics, and Populism [00:09:23].
  • Of all constraints, labor (People) is the most critical bottleneck limiting infrastructure scaling [00:05:48].
  • In the US alone, the build-out requires 500,000 new jobs [00:06:33]. This breaks down into 300,000 jobs for direct power generation and 200,000 highly specialized jobs for grid transmission and distribution [00:06:40].
  • Grid transmission jobs present a profound friction point because electricians require 4 years of skilling [00:06:58].
  • The current pipeline is drastically short: there are only 45,000 energy apprentices active in the US today, requiring an immediate influx of an additional 20,000 to 25,000 apprentices to meet demand targets [00:07:09].

Overcoming Inertia: Behind the Meter Solutions & Political Affordability [00:07:59]

  • Because upgrading the physical grid could take 3 to 5 years, deeply capacity-constrained hyperscalers are circumventing delays by using direct, "behind the meter" power generation [00:07:38].
  • These localized solutions often utilize simple cycle natural gas generators, which are quicker to deploy but far less efficient than combined cycle generators—which are not expected to be widely available until around 2029 or the early 2030s [00:08:27].
  • As data center expansions grow visible, affordability and local politics play a massive role, triggering community "NIMBY" (Not In My Back Yard) resistance over fears of rising consumer electricity bills [00:09:34].
  • To secure social license and green energy, hyperscalers are willing to pay a premium. A simulated premium of $40 per megawatt hour for reliable green energy would only reduce their 2030 EBITDA by a mere 2.5% and lower their corporate return on capital by less than 1% [00:11:10].

The 2030 Vision: A "Yes, And" Energy Mix [00:12:26]

  • Looking ahead to 2030, the global energy profile powering data centers is projected to stabilize around 60% thermal sources (primarily natural gas) and 40% renewables (solar, wind, battery) [00:12:49].
  • Nuclear power is poised to return, but initially only via the de-mothballing of existing facilities; new small and large-scale reactors are delayed by parts availability constraints and will likely not contribute meaningfully until the mid-2030s [00:13:37].
  • Solar and battery tech are viewed highly optimistically because they benefit from a Moore's Law-like dynamic, creating massive natural leverage on volume, price, and innovation curves [00:12:39].
  • Instead of competing technologies, the infrastructure paradigm relies on a "yes, and" model rather than a "no, or" environment, signaling that all available power generation formats will be integrated simultaneously to meet demand [00:14:30].

The Reference Vault

4. Data & Figures

Data PointValueContextTimestamp
Hyperscaler CapEx & R&D$300+ BillionTotal combined budget increase forecasted for 2026 and 2027.[00:01:23]
Previous Power Demand Growth Forecast175%The prior GS SUSTAIN forecast for 2030 global data center power demand growth before the recent revision.[00:04:07]
Revised AI Power Demand Growth220%Projected global growth in data center power demand by 2030 compared to 2023.[00:04:03]
Global Consumption RankTop 6The energy equivalent rank of the AI sector if it were a sovereign country.[00:04:19]

5. Core Frameworks & Mental Models

  • The 6 P's (Plus 3) of AI Infrastructure: A framework utilized by GS SUSTAIN to calculate constraints and drivers in the AI build-out. It comprises Pervasiveness, Productivity, Price, Policy, Parts, and People. The model isolates "People" (specifically skilled labor) as the most severe systemic bottleneck. George Lee notably appends Pricing, Politics, and Populism to this list. [00:05:32]
  • Agentic Systems as Token Furnaces: A conceptual framework for understanding the next wave of AI compute. Instead of simple query-response loops generated by humans, machines talking directly to other machines on complex, automated tasks will consume compute linearly and exponentially, behaving like continuous "furnaces." [00:03:30]
  • Moore's Law-Like Dynamic: A mental model applied to the trajectory of solar and battery capabilities, projecting that their efficacy and cost improvements will compound exponentially, creating a massive natural leverage on volume and price over time. [00:12:39]
  • The "Yes, And" Energy Adoption Model: Borrowed from improv comedy, this mental model rejects zero-sum thinking regarding energy types (i.e., natural gas vs. nuclear vs. renewables). It posits that AI's hyper-demand requires concurrent deployment of all available energy formats rather than choosing between them. [00:14:30]

6. Anecdotes

  • The Shale Revolution Boom-Bust Analogy: Brian Singer compares the current AI capital spending super-cycle to the early days of the Shale Revolution. In shale, an appraisal phase was followed by massive spending and an eventual bust caused by oversupply and eroded capital returns. However, he notes that AI diverges currently because the immense enterprise demand prevents any short-term product oversupply. [00:02:08]
  • Hyperscalers Going "Behind The Meter": To highlight the intense physical friction in the real estate market, Singer explains how tech giants are being forced to sidestep local grids. Because traditional grid upgrades will take up to five years, companies are buying up less-efficient simple cycle natural gas generators just to get their facilities operational today. [00:07:38]

7. References & Recommendations

  • GS SUSTAIN Research: Proprietary Goldman Sachs Research utilized by Brian Singer to model the 220% growth in global AI energy consumption.
  • Hyperscaler Community Statements: Public commitments issued by major technology firms promising to ring-fence utility costs on a "take or pay" basis, aiming to shield everyday consumers from price hikes caused by their data centers.

"Brookfield's the largest infrastructure owner in the world... We drew a pipeline and we showed all the different components of the payments ecosystem on a pipeline and said it's like a pipe that moves any commodity except what it's moving…

Total Labor Demand (US)500,000 JobsTotal number of specialized labor roles needed to fulfill generation and grid goals.[00:06:33]
Power Gen Labor Need300,000 JobsRoles required purely for building power generation facilities.[00:06:40]
Grid Transmission Need200,000 JobsSpecialized roles needed for transmission and distribution infrastructure.[00:06:44]
Electrician Skilling Time4 YearsThe duration of apprenticeship required to properly skill an electrician.[00:06:58]
Current Energy Apprentices45,000The total current count of active energy apprentices in the US.[00:07:09]
Apprentice Deficit20,000 - 25,000The immediate increase in energy apprentices needed to achieve infrastructure forecasts.[00:07:09]
Grid Upgrade Lag3 to 5 YearsThe delay timeline forcing hyperscalers into "behind the meter" independent power grids.[00:07:38]
Combined Cycle Gas Timeline~2029 - 2030sWhen highly efficient combined cycle natural gas generators will hit the market at scale.[00:08:27]
Green Energy Premium$40 / MWhTheoretical cost increase for sourcing entirely green, reliable energy solutions.[00:11:10]
EBITDA Impact of Premium2.5%The minimal impact the $40 premium would have on hyperscalers' 2030 EBITDA.[00:11:27]
ROIC Impact of Premium< 1%Impact on cash return on cash invested, proving tech companies can easily afford energy price hikes.[00:11:31]
Thermal Power Mix 203060%Estimated share of power sourced from natural gas thermal sources by the end of the decade.[00:12:49]
Renewable Power Mix 203040%Estimated share of power sourced from renewables (solar, wind, battery) by the end of the decade.[00:12:52]