Key Takeaways
- • Samir Kaul, founding partner at Khosla Ventures, argues the chip shortage that underpins NVIDIA's roughly 75% gross margins and $10-12 billion in monthly cash flow is not durable, citing Google selling TPUs to Amazon and OpenAI as evidence of surplus supply.
- • Kaul predicts that within a year markets will stop treating semiconductors as the growth bottleneck and instead recognize energy and upstream materials as the real limits on AI growth, a view he expects to be mainstream by 2027.
- • The call coincided with a global chip selloff on November 3: the Philadelphia Semiconductor Index fell 3.37%, Intel dropped 8.80%, and TSMC's US-listed ADR slipped 1.85% to $398.63, per the source's market data.
- • Kaul argues early-stage biotech is a poor risk-adjusted bet, citing a study of roughly 100 biotech IPOs showing about seven years from Series A to IPO and returns under 2x on paid-in capital.
- • Kaul frames sequencing cost declines from over $1 billion to about $100 as turning biology into a data business, and sees AI's clearest near-term value in healthcare and tutoring.
Summary
Samir Kaul, founding partner and managing director at Khosla Ventures, laid out his macro and sector views on the MTS podcast, as reported by BigGo Finance. His central argument is that the semiconductor supply crunch behind NVIDIA's roughly 75% gross margins is not a lasting moat. He points to Google selling its TPUs to rivals Amazon and OpenAI, which have announced purchases. In his view, a firm facing a genuine shortage would hoard capacity. He calls the assumption of a persistent shortage "just bananas" over a five-year horizon.
The argument arrives as investors are already questioning AI returns. On November 3, the Philadelphia Semiconductor Index fell 3.37% after opening higher, Intel dropped 8.80%, and TSMC's US-listed ADR slipped 1.85% to $398.63. In Hong Kong, Lenovo fell 6.9%, SMIC nearly 6%, and Hua Hong more than 7%. According to the source, the selloff was driven by doubts over whether AI capital spending will convert into returns rather than by Kaul's comments. Kaul's point is that the market has not yet decided chips are abundant, and that the question of whether the spending can pay for itself is what matters next.
Kaul's alternative bottleneck is energy and the materials behind it. He cites oil at $90 a barrel as a test case that would keep inflationary pressure elevated despite efficiency gains, making the Federal Reserve's 2% target "going to be tough" to hit. He also highlights minerals used in magnets, batteries, and fusion reactors as underpriced inputs to computation. He frames his one-year forecast as the central test of his thesis, with energy and materials recognized as the binding constraints by 2027.
On biotech, Kaul draws on a dataset from his time at Flagship Ventures covering about 100 biotech IPOs over the prior decade. He reports roughly seven years from Series A to IPO, $100-150 million of private capital raised before listing, a trading value near $250 million at IPO, and returns under 2x on paid-in capital, with IPO support mostly from insiders. He notes that large Series A rounds at firms such as Flagship and Arch now aim to reach $20-30 billion in outcomes to clear return thresholds, and that Khosla Ventures continues to back science-heavy companies facing that bar.
Kaul's view of sequencing is that costs have fallen from over $1 billion and seven to ten years for a human genome in the late 1990s to about $100 today. He cites Ultima Genomics, co-founded with Gilad Almogy, and Guardant Health's blood-based tumor DNA detection as examples of the payoff. On AI, he argues its clearest benefits are in patient-facing informational asymmetry, such as analyzing blood reports, and in low-cost tutoring. He is skeptical that AI will mainly generate new drug targets, arguing its larger role is in patient selection and faster enrollment in trials. He also calls animal testing "a joke" and points to a possible FDA move toward accepting organoids. Kaul, who describes social media as having "zero positivity," offers two falsifiable forecasts, with the 2027 energy call the nearest-term test.
The argument arrives as investors are already questioning AI returns. On November 3, the Philadelphia Semiconductor Index fell 3.37% after opening higher, Intel dropped 8.80%, and TSMC's US-listed ADR slipped 1.85% to $398.63. In Hong Kong, Lenovo fell 6.9%, SMIC nearly 6%, and Hua Hong more than 7%. According to the source, the selloff was driven by doubts over whether AI capital spending will convert into returns rather than by Kaul's comments. Kaul's point is that the market has not yet decided chips are abundant, and that the question of whether the spending can pay for itself is what matters next.
Kaul's alternative bottleneck is energy and the materials behind it. He cites oil at $90 a barrel as a test case that would keep inflationary pressure elevated despite efficiency gains, making the Federal Reserve's 2% target "going to be tough" to hit. He also highlights minerals used in magnets, batteries, and fusion reactors as underpriced inputs to computation. He frames his one-year forecast as the central test of his thesis, with energy and materials recognized as the binding constraints by 2027.
On biotech, Kaul draws on a dataset from his time at Flagship Ventures covering about 100 biotech IPOs over the prior decade. He reports roughly seven years from Series A to IPO, $100-150 million of private capital raised before listing, a trading value near $250 million at IPO, and returns under 2x on paid-in capital, with IPO support mostly from insiders. He notes that large Series A rounds at firms such as Flagship and Arch now aim to reach $20-30 billion in outcomes to clear return thresholds, and that Khosla Ventures continues to back science-heavy companies facing that bar.
Kaul's view of sequencing is that costs have fallen from over $1 billion and seven to ten years for a human genome in the late 1990s to about $100 today. He cites Ultima Genomics, co-founded with Gilad Almogy, and Guardant Health's blood-based tumor DNA detection as examples of the payoff. On AI, he argues its clearest benefits are in patient-facing informational asymmetry, such as analyzing blood reports, and in low-cost tutoring. He is skeptical that AI will mainly generate new drug targets, arguing its larger role is in patient selection and faster enrollment in trials. He also calls animal testing "a joke" and points to a possible FDA move toward accepting organoids. Kaul, who describes social media as having "zero positivity," offers two falsifiable forecasts, with the 2027 energy call the nearest-term test.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.