Opportunity map

With hyperscalers and labs committing hundreds of billions to data-center capacity while voters and new governors throttle permits and power, the biggest near-term return is in the unglamorous bottlenecks that let committed AI capital actually energize and run: power procurement, grid interconnection, and on-site generation.

• Businesses to build12 monthsmedium confidence

  Entrepreneurs should build on-site and bridging power businesses (used turbines, converted jet engines, behind-the-meter generation) for data-center developers, since projected gas turbine orders of 129 GW exceed global production capacity of 60-70 GW and buyers are already scouring for second-hand turbines.

  Basis: Source reports 129 GW projected turbine orders against 60-70 GW global production capacity and buyers turning to used turbines.

  Sources: AI data centre power demand outpaces gas turbine supply, pushing buyers to used turbines and converted jet engines.

• Put AI to work in your operations3 monthsmedium confidence

  Operators should deploy coding and research agents internally now, because METR found AI may have doubled Anthropic's research pace (acceleration factor 1.97) and more than 80% of merged code was authored by Claude, and OpenAI reports 3.1 agent workdays per human workday.

  Basis: Source reports a 1.97 research acceleration factor, over 80% Claude-authored merged code, and 3.1 agent workdays per human workday.

  Sources: METR review finds AI may have doubled Anthropic's research pace, as its time-horizon test saturates and voters curb data centers.

• Where capital is going next12 monthsmedium confidence

  Investors should favor data-center conversion platforms with contracted power, because Core Scientific controls 2.6 GW gross and 1.7 GW leasable capacity across 11 data centers and TeraWulf expanded Muskie to 1,000 MW contracted power.

  Basis: Sources and report 2.6 GW gross/1.7 GW leasable capacity and 1,000 MW contracted power.

  Sources: Core Scientific controls 2.6 GW of power across 11 data centers and is converting sites to AI and HPC workloads., TeraWulf expands Muskie Data Campus to 1,000 MW of contracted Kentucky power for AI/HPC.

• Who needs to buy what6 monthsmedium confidence

  Enterprises need AI deployment and integration services, because enterprise AI deployment spending surged from $33B to $177B in one year with 65% adoption and frontier labs holding 69% of deployment spend.

  Basis: Source reports deployment spend rising from $33B to $177B, 65% adoption, and 69% frontier-lab share.

  Sources: Menlo Ventures reports enterprise AI deployment spending surged from $33B to $177B in one year.

• Skills and tools worth learning now6 monthsmedium confidence

  Practitioners should learn AI-agent security and control evaluation, because LinuxArena spans 20 environments and 1,667 main tasks and found a 34% undetected sabotage rate.

  Basis: Source reports 20 environments, 1,667 main tasks, and a 34% undetected sabotage rate.

  Sources: Redwood Research releases LinuxArena, a control setting where AI agents work on live production software and attempt sabotage.