What the future looks like
If it arrives, superintelligence means AI systems that outperform the best human experts in most fields and can work on hard problems for days or weeks on their own. Anthropic's CEO describes the idea as “a country of geniuses in a datacenter” (Dario Amodei, “Machines of Loving Grace”, Oct 2024).
Whether or not that exact milestone is reached in the next decade, the race toward it is already reshaping the economy: it needs enormous amounts of chips, memory, data centers and electricity, and that build-out is happening now.
How big could it get?
- There is no credible market-size estimate for superintelligence itself. Nobody can size a market for a technology that does not exist yet, so we track the spending and value pools around it instead.
- Data centers worldwide will need about $6.7 trillion of capital investment by 2030, of which about $5.2 trillion is for AI workloads (midrange scenario) (McKinsey, 2025).
- Data-center electricity use is projected to more than double, from about 415 TWh in 2024 to about 945 TWh in 2030 (IEA, 2025).
- Today's generative AI alone could add $2.6–4.4 trillion a year in economic value across 63 use cases (McKinsey, 2023).
Figures are third-party estimates, quoted as published. Different studies use different scopes and years, so they aren't always comparable.
What we're watching
- Frontier models handling long, multi-step tasks reliably without human help.
- Hyperscaler capital-spending guidance, the clearest real-time measure of the race.
- Power: new grid connections, on-site generation and energized data-center capacity.
- Memory supply: high-bandwidth memory (HBM) availability and pricing.
What could go wrong
- Progress could plateau, leaving the industry with excess capacity and a capex hangover.
- Power and grid constraints could slow the build-out regardless of demand.
- Regulation, export controls and safety concerns.
- Customer concentration: a handful of buyers account for much of the spending.
What the future makers say
Predicted in April 2024 that AI smarter than the smartest human would probably arrive by the end of the following year, with chips and electricity as the main constraints.
“Developing superintelligence is now in sight.”
Demand for large language models and the software that makes them useful has “turned into a stampede.”
Quotes are reproduced from the linked public sources; paraphrases are labeled. Inclusion doesn't imply any affiliation with or endorsement by these people or companies.
Companies building it
Only companies we own. Tickers only, never position sizes.
The leading platform for training and running frontier AI: GPUs, networking, systems and CUDA software.
High-bandwidth memory sits beside every AI accelerator, and bigger models need more of it.
Power-rich data-center sites being redirected to AI cloud capacity.
On-site fuel-cell power for data centers that can't wait years for a grid connection.
Has stated its goal of bringing “personal superintelligence” to everyone (Meta, 2025).
AWS hosts frontier models and builds its own Trainium AI chips.
Turns advanced models into real operations for governments and enterprises.
Sources
- McKinsey, 2023
- McKinsey, 2025
- IEA, 2025
- Dario Amodei, “Machines of Loving Grace”, Oct 2024
- Reuters, Apr 2024
- Meta, Jul 2025
- CNBC, May 2025
Last reviewed: 22 Oct 2026.
Not investment advice. This page is opinion and education, researched and presented with the help of AI analyst personas, and may contain errors. The EarlyConviction team holds positions in NVDA, MU, IREN, BE, META, AMZN, PLTR and may buy or sell at any time. Read the disclaimer.