Juicing AI with Photons — Behind the Markets

PodcastBehind the MarketsCarly AndersonOct 16

The discussion examines the infrastructure demands of AI, from data-center power and cooling to optical links, grid technology, and the financing of capital-intensive startups. Anderson also points to opportunities and constraints in fusion, batteries, recycling, and materials, while Siegel discusses inflation, GDP, deposits, and the role of AI investment in long-term rates. A separate market discussion flags power as a possible data-center bottleneck and outlines the potential downside for stocks if AI investment disappoints.

AI Infrastructure Energy Markets Semiconductors Venture Capital Inflation Interest Rates

Preview · 24 more points in the app

Read the Full Summary

Key Takeaways

Siegel says AI spending has not yet produced visible productivity gains or better margins, making the sustainability of the build-out important to long-term rates. He argues that continued growth and funding demand could push long rates higher, while disappointment in token use or build-out decisions could pull rates lower and trigger a 20–30% Nasdaq correction and a 5–10% S&P 500 decline.

Data-center power is emerging as a construction bottleneck, with a speaker suggesting it could become the key shortfall next year; customers are responding with 20-year PPAs and faster interconnection. Anderson says neo-cloud operators also have to secure interconnection while managing water and energy use, alongside cooling approaches such as GPU liquid cooling and systems using CO2 refrigerants.

Anderson describes several tools for managing a more flexible grid: a buffer that smooths data centers’ sub-second power swings, solid-state transformers that could enable bidirectional flows, and virtual power plants that aggregate commitments to turn equipment off. She expects some tools to appear in certain areas within three years, with broader grid changes in five years and mainstream adoption beginning around 2030.

Carly Anderson says deep-tech startups face a smaller funding gap between Series A and pre-IPO, even though some capital-intensive companies may need more than $100 million to build. She also says AI and simulation tools can compress engineering iterations for turbine and electrical designs from weeks or months on supercomputers to minutes or hours on a MacBook.

The discussion highlights recycling as a potential supplement to mined materials, given the host’s estimate that opening or reopening a mine can take 20 to 30 years. Anderson points to aluminum scrap—about 2 million tons of which the U.S. exports annually—as a refining opportunity, while copper demand was described as potentially enormous and Anderson said securing access to copper over the next five years was worthwhile.

TECHNOLOGY

AI Productivity — Hyperscalers and companies using AI tokens are spending, but available data has not yet shown productivity gains or improved margins. Siegel says he hopes those gains begin to appear.

“I wanna see productivity of those firms come through, which in the data we have not seen yet. They're spending, they're maintaining, but they're not improving margins at this point…”—

Deep-Tech Financing — Carly Anderson says financing for capital-intensive technology startups has improved, narrowing the gap between Series A and pre-IPO funding. Some companies may need more than $100 million to build and could transform large markets.

“The capital stack has also finally shown up…”—

AI Design Tools — Simulation, AI, and machine-learning tools let engineers iterate on turbine components and electrical designs in minutes to hours. Tasks that once took weeks to months on supercomputers can now run on a MacBook, speeding development of physical-world technologies, Anderson says.

“And so that's everything from, you know, the simulation, AI, and machine learning tools where you can now iterate on designs for a turbine component or an electrical design, you…”—

ECONOMY

Inflation Estimate — Goldman Sachs raised its estimate of the core consumption deflator due later that month by roughly 40–60 basis points. It expects the measure to move lower in November and December.

“And Goldman Sachs really raised maybe by 40, 50, 60 basis points its estimate of the core consumption deflator that is gonna be reported at the end of this month…”—

GDP Expectations — Siegel says third-quarter GDP looks likely to post a three-handle. He notes that the St. Louis Fed and Goldman Sachs also expect a figure beginning with 3.

“GDP in the third quarter looks like a three handle. Both the Fed in St. Louis and Goldman Sachs think it's gonna be in in a three handle.”—

MARKETS

Anthropic IPO — The speaker says reactions to Anthropic's news were mixed and expects its IPO the following month could be the biggest IPO in existence and the stock event of November.

“It seems like it's a prelude to the SpaceX Tesla merger one day and that this is part of his, you know, I mean, also, by the way, should mention…”—

Data-Center Power — Power has emerged as a bottleneck for data-center construction. A speaker suggests it may be the key shortfall next year, ahead of chips or memory. EIA figures cited in the discussion show solar and storage as the predominant and fastest-growing additions to grid power, while gas-generation additions are also high. Data-center and hyperscaler customers are willing to sign 20-year PPAs and move quickly to interconnect.

“Right? As power has emerged as a huge bottleneck, for what we can build out.”—

State Power Demand — Anderson says EV connections drive more electricity demand than data centers in some U.S. states. The balance differs across states such as Virginia and Texas.

“Like, while we have an insane number of people connecting EVs, honestly, that is a bigger demand…”—

FORECASTS

AI And Long-Term Rates — Siegel argues that the sustainability of the AI build-out, rather than oil or expected Fed action, is a major driver of long-term rates. Continued growth and funding demand could push long rates higher, while signs the build-out could crack could pull them lower.

“This means that the major source of long term interest rates going up, it's not oil. It's not even what the Fed's gonna do…”—

AI Investment And Yields

Siegel argues that the sustainability of AI infrastructure spending, more than oil or Fed policy, is driving long-term yields and recent growth-value market swings.

Inflation, Deposits, And Growth

Siegel reviews producer-price pressure, slowing deposit growth, resilient GDP and labor data, and the lack of clear productivity gains from corporate AI spending.

AI Correction Scenarios

Siegel says an AI disappointment could pull Nasdaq down sharply while lower yields cushion value stocks and limit the broader S&P 500 decline.

Timescale Ventures Thesis

Anderson says Timescale was founded to back deep-tech hardware as stronger talent, more available capital, and improved simulation tools make ambitious companies more viable.

Data Center Power Bottlenecks

Anderson describes a buffer technology that smooths rapid AI data-center power fluctuations and identifies grid interfaces as a major opportunity for deep-tech startups.

Grid Capacity And Nuclear

Anderson says developers are racing to add power, with gas, solar, storage, and smaller nuclear designs competing to deliver capacity on practical timelines.

Solid-State Transformer Potential

Anderson explains that silicon-based transformers could offer bidirectional power, better monitoring, and flexible grid controls, though cost and deployment timelines remain uncertain.

Modernizing The Grid

Anderson says hyperscalers and data-center operators are helping startups pilot grid technologies, while meaningful transformation is likely to unfold over years rather than months.

Data Center Cooling

Anderson describes rapid innovation in thermal interface materials, liquid cooling, and more efficient cooling equipment as operators compete to control energy and water use.

Critical Minerals And Recycling

Anderson favors recovering usable metals from existing materials, including aluminum scrap, and discusses refining, rare-earth recovery, copper extraction, and the long lead times for new mines.

Battery Recycling Economics

Anderson says metal recycling can be attractive, but cheaper battery chemistries have weakened recycling economics; tracking and repurposing used batteries may help sustain value.

Battery Chemistry And Applications

Anderson notes China’s progress in sodium batteries and emphasizes pack design for high-power applications, while warning that new chemistries face long commercialization paths.

Space Communications And Photons

Anderson sees optical communications as a key space opportunity, with lasers moving large amounts of data and new markets emerging for specialized power systems.

Optical Links And Lasers

Anderson says optical links are already replacing some electronic connections between data-center cabinets, while laser advances also support quantum computing, fusion, and defense.

Fusion Timelines And Materials

Anderson says fusion developers target the 2030s, but prototypes and testing are needed to resolve uncertainties around system performance and first-wall material durability.

Timescale Venture Underwriting

Anderson says Timescale prioritizes capable founders who can demonstrate products early, while relying on industry experience and advisors to assess risk and execution.