Technology and Innovation Community

Latest insight from the Team @ $GS available through the link below:

AI and Data Centers

Includes brief overviews of:

#ai #innovationcommunity

Thanks Todor

Another Bloomberg NEF article complementing the previous GS report:

Key Insights:

• Electricity demand from AI training and services is expected to quadruple by 2035, reaching about 1,600 TWh - roughly 4.4% of global power use.

• If counted as a country, AI-focused data centers would rank fourth in electricity use, behind China, the US, and India.

• The United States remains the largest market, with data-center demand in key hubs growing faster than that of electric vehicles, hydrogen, and other new technologies.

• AI workloads are far more energy-intensive than traditional computing.

• Data centers are scaling up: in 2010 most used under 5 MW; new sites now plan for hundreds of megawatts, with some approaching gigawatt scale.

• Efficiency gains can't keep up: PUE (Power Usage Effectiveness) is expected to improve from 1.4 to 1.2 by 2030, but the growth in computing power outweighs these savings.

• Short-term: growth is largely met by natural-gas-fired power, which is abundant and flexible in major US markets.

• Long-term: operators face pressure from policymakers, net-zero targets, and customers to expand using cleaner energy sources.

Latest insight from the Team @ $GS available through the link below:

AI and Data Centers

Includes brief overviews of:

#ai #innovationcommunity

Great info. Thanks Carlos. The focus going forward will be on ramping up energy capacity and compute. This on top of everything else.

-Todor

Thanks Todor

Another Bloomberg NEF article complementing the previous GS report:

Key Insights:

• Electricity demand from AI training and services is expected to quadruple by 2035, reaching about 1,600 TWh - roughly 4.4% of global power use.

• If counted as a country, AI-focused data centers would rank fourth in electricity use, behind China, the US, and India.

• The United States remains the largest market, with data-center demand in key hubs growing faster than that of electric vehicles, hydrogen, and other new technologies.

• AI workloads are far more energy-intensive than traditional computing.

• Data centers are scaling up: in 2010 most used under 5 MW; new sites now plan for hundreds of megawatts, with some approaching gigawatt scale.

• Efficiency gains can't keep up: PUE (Power Usage Effectiveness) is expected to improve from 1.4 to 1.2 by 2030, but the growth in computing power outweighs these savings.

• Short-term: growth is largely met by natural-gas-fired power, which is abundant and flexible in major US markets.

• Long-term: operators face pressure from policymakers, net-zero targets, and customers to expand using cleaner energy sources.

Latest insight from the Team @ $GS available through the link below:

AI and Data Centers

Includes brief overviews of:

#ai #innovationcommunity

New article on this topic.

Exec Summary:

AI's energy intensity and data-center expansion are transforming electricity demand, with nuclear power emerging as a critical, premium-priced, low-carbon baseload option. Limited new nuclear capacity and persistent hyperscaler demand could sustain elevated contract premiums and drive multi-billion-dollar EBITDA growth for leading operators.

Key Insights:

• AI-driven power demand:
  Growth in artificial intelligence, particularly large language models (LLMs), could significantly increase U.S. electricity consumption by 11–26% by 2030, requiring 131–310 GW of new generation capacity to support an additional 345–815 TWh of demand.

• Energy intensity:

  • Generative-AI queries consume up to 10× more energy than standard internet searches.

  • Nvidia Blackwell GPU: 1,200 watts vs. ~150 watts for a CPU.

  • Major investors include OpenAI, Amazon, Google, and Meta, each planning multi-billion-dollar IT infrastructure outlays.

• AI Capex forecasts (BI Technology team):

  • Training scenario: Spending peaks in 2026 after a rapid 2023–26 phase.

  • Inference scenario: Capex grows through 2032, with accelerated growth from 2023–26 and slower gains thereafter.

• Energy source trade-offs:

  • Nuclear: Highest capacity factor (90%) but most expensive ($12,500/kW).

  • Coal: 60%, $5,000/kW.

  • Gas (CCGT): 60%, $2,500/kW.

  • Wind: 35%, $1,500/kW.

  • Solar: 25%, $1,500/kW.

  • Renewables' intermittency requires storage or backup; fossil fuels face emissions constraints.

• Nuclear premium pricing:

  • Data centers are paying $15–$25/MWh above market.

  • Example: Constellation Energy's $840 million 10-year VPPA with the U.S. General Services Administration covers 1 million MWh/year (~$80/MWh).

  • EBITDA uplift potential: Constellation +$1.8B/year, Vistra +$500M/year if half of nuclear capacity is contracted at midpoint premium.

• Nuclear capacity ownership (MW):

  • Constellation 22,000, Vistra 6,500, PSEG 3,800, NextEra 2,300, Talen 2,200, Dominion 2,000.

• Contract structures:

  • Shift from behind-the-meter (BTM) to front-of-the-meter (FTM) VPPAs, avoiding regulatory hurdles but adding ~$7/MWh in transmission costs.

  • BTM offers better load control, avoids interconnection delays, and may lower capex.

  • Example projects:

    • Microsoft–Constellation (835 MW), Crane restart in 2027.

    • Meta–Constellation (Clinton plant) post-subsidy in 2027.

    • Talen–AWS (Susquehanna) expanded from 960 MW → 1.9 GW, including 300 MW BTM.

Great info. Thanks Carlos. The focus going forward will be on ramping up energy capacity and compute. This on top of everything else.

-Todor

Thanks Todor

Another Bloomberg NEF article complementing the previous GS report:

Key Insights:

• Electricity demand from AI training and services is expected to quadruple by 2035, reaching about 1,600 TWh - roughly 4.4% of global power use.

• If counted as a country, AI-focused data centers would rank fourth in electricity use, behind China, the US, and India.

• The United States remains the largest market, with data-center demand in key hubs growing faster than that of electric vehicles, hydrogen, and other new technologies.

• AI workloads are far more energy-intensive than traditional computing.

• Data centers are scaling up: in 2010 most used under 5 MW; new sites now plan for hundreds of megawatts, with some approaching gigawatt scale.

• Efficiency gains can't keep up: PUE (Power Usage Effectiveness) is expected to improve from 1.4 to 1.2 by 2030, but the growth in computing power outweighs these savings.

• Short-term: growth is largely met by natural-gas-fired power, which is abundant and flexible in major US markets.

• Long-term: operators face pressure from policymakers, net-zero targets, and customers to expand using cleaner energy sources.

Latest insight from the Team @ $GS available through the link below:

AI and Data Centers

Includes brief overviews of:

#ai #innovationcommunity