The Bridge Ep. 23: Can AI Deliver Returns, Not Just Hype?

Summary

Torsten Slok of Apollo argues that AI’s real question is not whether demand exists, but whether the cash flows can justify the massive capital being deployed in a higher-rate world. He says higher interest rates hurt long-duration AI investments by lowering the present value of future cash flows, while AI spending is already contributing meaningfully to GDP through data centers, energy, tokens, tools, and wealth effects. Slok also breaks the ecosystem into distinct buckets—frontier labs, hyperscalers, energy/grid, and chips/equipment—and notes that competitive pressure from open-source models is compressing the window for pricing power. He sees AI as supportive of productivity, employment, and business formation, but warns that markets, private credit, and rate-sensitive sectors like housing could all be stressed if AI growth slows while the broader economy remains bifurcated.

Notes

• Higher rates raise the discount rate on far-future AI cash flows, making venture-style AI bets less attractive.
• Slok treats AI as a macro driver, estimating about 1 percentage point of roughly 2% GDP growth is AI-related.
• AI’s GDP footprint extends beyond data centers into energy, tokens, consumer tools, business tools, and wealth effects.
• He splits AI investing into four buckets: frontier labs/models, hyperscalers/data centers, energy/grid, and silicon/equipment.
• Frontier labs are the most exposed to consumer adoption and model monetization.
• He says compute demand is rising sharply as users interact with chatbots and more agentic systems.
• Open-source models are gaining usage and pressuring closed-source pricing power.
• He says open-source and closed-source capability gaps have narrowed to roughly three to four months.
• That short capability gap implies a much shorter window for moats and monetization.
• He reads market stress in widening CDS spreads on hyperscaler names.
• Consensus still expects hyperscaler operating cash flow to rise from about $600 billion to roughly $2 trillion.
• He argues AI capital formation is mostly happening outside public markets because the largest beneficiaries are private.
• He says 90% of U.S. businesses with more than $100 million revenue are private, so the AI economy is underrepresented in the S&P 500.
• Data-center buildout is now central to GDP forecasting, not a minor capex line item.
• He thinks AI should still lift productivity and employment, not just replace labor.
• He cites census data showing record business formation, including solo-founder firms.
• He calls AI a productivity and employment “miracle drug,” with no major displacement visible yet.
• Recent labor reports remain strong, suggesting AI displacement is not yet large in the data.
• He says rate-sensitive sectors such as housing are weak while AI-related spending props up the broader economy.
• Mortgage rates rose from about 2.7% in the pandemic to about 7%, worsening the bifurcation.
• He says the Fed is stuck balancing inflation still around 3.5% against a 2% target.
• The last-mile inflation problem limits how quickly policy can ease for rate-sensitive assets.
• He says government debt dynamics are constrained by entitlement-heavy spending and politically difficult fiscal adjustment.
• About 75 million people receive Social Security, making entitlement reform politically difficult.
• Households are moving into money funds and T-bills while selling longer-duration government debt.
• That flow pattern supports the front end and leaves long bonds vulnerable.
• A recession would likely widen the fiscal deficit from about 5% to about 10%.
• Treasury buybacks matter more as a signal than as a size effect because they are tiny versus $40 trillion of debt.
• The Treasury’s buyback ramp from $2 billion to $6 billion is symbolic relative to the debt stock.
• Private credit is especially exposed in software, where direct lending is concentrated.
• Software is about 25% of direct lending outstanding, or roughly $500 billion of a $2 trillion market.
• Higher-for-longer rates are particularly painful for 2020-2022 private credit vintages.
• Seven-year bullet structures from 2020-2022 are resetting into 2028-2029, exposing weak underwriting.
• In private equity, he favors businesses with current earnings that can absorb higher debt service.
• The practical private-markets takeaway is to prefer quality and cash generation over distant growth.