Skip to main content

Future-Oriented Companies: Technology & Mineral Innovation for the Data Center Economy | Sept. 24

Investor TV Oct 8, 2026 54:48

Summary

Deep Sea Minerals' CEO argues polymetallic nodules (seafloor rocks holding copper, nickel, cobalt, manganese) can augment, not replace, land-based mining.

Power shortages are AI's real bottleneck; Chip AI's CEO says transformer, switchgear (grid electrical gear), and memory backlogs push capacity toward 2029.

Panelists see historic US critical-minerals policy support: seabed mining declared a national security priority, backed by multibillion-dollar government financing commitments.

James Deckelman cites an estimated 50,000 tons of copper per new AI data center, with nickel, cobalt, and manganese demand close behind.

FUTR's CEO argues personal data is an investable asset class, with consumer-owned vaults and AI agents capturing value big tech currently keeps.

James Deckelman's takeaways

  • Top takeaway He argues the real returns in the cited $7 trillion data center build-out sit in power infrastructure, cooling and physical real estate — about 25% of that capital, roughly $1.75 trillion, by his reading of the scenario.
  • He argues polymetallic nodules will augment, not replace, tightening terrestrial supply of copper, nickel, cobalt and manganese, with first commercial recovery permits possibly issued within a year of the panel and first commercial production in the coming years.
  • He argues US policy support for seabed mining is real and likely durable, citing the April 2025 executive order declaring seabed minerals a national security priority, BOEM lease sales planned for November and December, and more than $16 billion in US government critical minerals commitments over the prior 10 months.
  • He argues Deep Sea Minerals is partly insulated from any AI slowdown because its target minerals also serve defense and energy-transition demand, including the cobalt, nickel and copper used in phones and laptops.
  • He stresses the company works strictly through NOAA under DISHMA, a review process he says can take over a year from application to potential award and should not be shortcut.

Jon Franklin's takeaways

  • Top takeaway He frames the panel around power constraints, citing the McKenzie estimate of $7 trillion in global data center spending by 2030 and noting a large portion of capacity due to open this year is delayed for lack of power.
  • He says he is a big bull on the critical minerals cycle and highlights cooling-driven siting choices, pointing to Google's $15 billion investment in northern Finland two weeks before the panel.

Alex McDougall's takeaways

  • Top takeaway He argues data is an asset class in its own right and positions FUTR as a 'universal picks and shovels' way to invest in the AI boom — participating directly in the data that feeds compute regardless of how power constraints get resolved.
  • He argues that bringing consumers inside the data value chain with private vaults aligns FUTR with what privacy regulation is trying to achieve, making compliance easier than for extract-and-monetize incumbents.
  • He argues big tech is unlikely to copy the model soon because returning a share of data revenue to consumers — he floats 40% — would upend incumbents' business models and distribution.
  • He argues AI adoption is still early, comparing it to the internet's adoption timeline cut in half to roughly 10 years, so compute demand should keep growing even if some investment runs ahead of real-economy usage.
  • He argues edge compute is a net plus for FUTR's mobile-based app — reducing latency, cost and sovereignty friction — though he expects the shift to take years and require step-function efficiency gains.

Kenneth Cotiamco's takeaways

  • Top takeaway He argues power is the biggest AI bottleneck — the grid cannot support the data centers being planned — and that metal supply is being bought out years in advance as prices take off.
  • He argues the entire supply chain is on back order — talent, transformers, switchgear, PDUs and high-bandwidth memory — and estimates backlogs may not start clearing until around 2029.
  • He describes Chip AI's colocation model — building everything up to the server without GPU or hardware exposure — as keeping capex low, with another 25 megawatts planned for the first half of 2027 and a hoped-for 90-plus megawatts in 2027.
  • He argues Chip AI avoids heavy permitting by building smaller 10-30 megawatt data centers inside existing permitted buildings, and expects a data center or AI tax eventually, for which the balance sheet is being prepared.
  • He argues high-end AI compute cannot run on phones, so the shift to edge and decentralized computing does not threaten centralized data center demand.