Victor Khosla

Khosla mentions investing in a large European steel company and using AI to "buy better scrap metal," which he claims can create "$50-75 million in EBITDA." This is the "Real Economy" application of AI. Instead of AI *replacing* the product (as it might with software code), AI is used here to optimize logistics and procurement for a physical product. Old-economy industries that successfully integrate AI for efficiency will see significant margin expansion without the existential threat facing the tech sector. LONG. Look for industrial firms with heavy operational costs where AI can drive efficiency. Global economic slowdown or Chinese oversupply could crush steel prices regardless of operational efficiency gains.

Khosla states, "If private equity firms have bought software businesses at 20 plus times cash flow relying on growth, you are going to see large pockets of problems." He notes that software is a "big stick" that could make the credit market buckle. The traditional LBO model for software relies on sticky, recurring revenue (seat-based pricing). AI agents and automation threaten to destroy this "knowledge work" pricing model. If revenue growth stalls or reverses due to AI efficiency, the massive debt loads placed on these companies by PE firms become unserviceable, leading to defaults that will hit Private Credit lenders and equity holders. SHORT / AVOID. High-valuation, high-debt software companies are the epicenter of the next credit cycle. AI adoption might be slower than expected, or software companies may successfully pivot to usage-based AI pricing models.

Khosla observes, "High yield spreads today are 300... credit at 300 basis point spreads... is mispriced." He compares this to the 2015 energy crash where spreads blew out to 900 bps. A spread of 300 bps implies a benign economic environment, yet default rates are already at 6%. This is a pricing error. When the market realizes the "fat tail risk" (likely triggered by software/tech defaults), spreads must widen to compensate for the risk. Widening spreads mathematically equal falling bond prices. SHORT / AVOID. The risk/reward in high-yield credit is currently asymmetric to the downside. The Fed could cut rates aggressively, keeping zombie companies alive and spreads artificially tight.

Khosla states, "If private equity firms have bought software businesses at 20 plus times cash flow relying on growth, you are going to see large pockets of problems." He notes that software is a "big stick" that could make the credit market buckle. The traditional LBO model for software relies on sticky, recurring revenue (seat-based pricing). AI agents and automation threaten to destroy this "knowledge work" pricing model. If revenue growth stalls or reverses due to AI efficiency, the massive debt loads placed on these companies by PE firms become unserviceable, leading to defaults that will hit Private Credit lenders and equity holders. SHORT / AVOID. High-valuation, high-debt software companies are the epicenter of the next credit cycle. AI adoption might be slower than expected, or software companies may successfully pivot to usage-based AI pricing models.

SVP is investing heavily (40% of portfolio) in "real assets, power plants, airplanes, real estate, toll roads." Khosla explicitly states, "We are not in the high knowledge growth game." This is a defensive rotation against AI disruption. While AI threatens "knowledge" jobs and SaaS cash flows, it cannot digitize a toll road, a power plant, or an airplane. Furthermore, AI data centers require massive amounts of power, creating a tailwind for power generation assets. LONG. These assets provide inflation protection and insulation from the technological deflation AI brings to software. A deep recession would lower demand for travel (airplanes) and energy (power), hurting cash flows despite their "real" nature.

Khosla observes, "High yield spreads today are 300... credit at 300 basis point spreads... is mispriced." He compares this to the 2015 energy crash where spreads blew out to 900 bps. A spread of 300 bps implies a benign economic environment, yet default rates are already at 6%. This is a pricing error. When the market realizes the "fat tail risk" (likely triggered by software/tech defaults), spreads must widen to compensate for the risk. Widening spreads mathematically equal falling bond prices. SHORT / AVOID. The risk/reward in high-yield credit is currently asymmetric to the downside. The Fed could cut rates aggressively, keeping zombie companies alive and spreads artificially tight.

SVP is investing heavily (40% of portfolio) in "real assets, power plants, airplanes, real estate, toll roads." Khosla explicitly states, "We are not in the high knowledge growth game." This is a defensive rotation against AI disruption. While AI threatens "knowledge" jobs and SaaS cash flows, it cannot digitize a toll road, a power plant, or an airplane. Furthermore, AI data centers require massive amounts of power, creating a tailwind for power generation assets. LONG. These assets provide inflation protection and insulation from the technological deflation AI brings to software. A deep recession would lower demand for travel (airplanes) and energy (power), hurting cash flows despite their "real" nature.

Khosla mentions investing in a large European steel company and using AI to "buy better scrap metal," which he claims can create "$50-75 million in EBITDA." This is the "Real Economy" application of AI. Instead of AI *replacing* the product (as it might with software code), AI is used here to optimize logistics and procurement for a physical product. Old-economy industries that successfully integrate AI for efficiency will see significant margin expansion without the existential threat facing the tech sector. LONG. Look for industrial firms with heavy operational costs where AI can drive efficiency. Global economic slowdown or Chinese oversupply could crush steel prices regardless of operational efficiency gains.