- Prioritizing strict margin targets too early forces resource constraints that curb experimentation and speed, which slows innovation during a hypergrowth phase.
Prioritizing strict margin targets too early forces resource constraints that curb experimentation and speed, which slows innovation during a hypergrowth phase.
- Fast model releases and frequent new hardware SKUs make older chips lose value quickly, which shortens the effective economic life of infrastructure and makes build-versus-buy timing much more risky.
Fast model releases and frequent new hardware SKUs make older chips lose value quickly, which shortens the effective economic life of infrastructure and makes build-versus-buy timing much more risky.
- Once agents act autonomously for users, incentive alignment will matter more because agents that handle purchases, messages, and finances must prioritize the user's goals over the provider's to avoid conflicts and harmful choices.
Once agents act autonomously for users, incentive alignment will matter more because agents that handle purchases, messages, and finances must prioritize the user's goals over the provider's to avoid conflicts and harmful choices.
- Technologies follow S-curves, so AI progress may naturally slow if architecture, data, or hardware limits appear, which would cap runaway growth and redirect competition to other layers.
Technologies follow S-curves, so AI progress may naturally slow if architecture, data, or hardware limits appear, which would cap runaway growth and redirect competition to other layers.
- If attractive acquisition targets are scarce, buybacks return cash to shareholders because repurchasing shares reduces float, offsets dilution from stock compensation, and redeploys idle capital more effectively than sitting on cash.
If attractive acquisition targets are scarce, buybacks return cash to shareholders because repurchasing shares reduces float, offsets dilution from stock compensation, and redeploys idle capital more effectively than sitting on cash.
- Savings from running a proprietary model versus frontier or open-source ones vary because cheaper alternatives help most on simple tasks, but complex, high-quality workloads keep per-inference costs high, so total savings often range from a few percent up to around 30%.
Savings from running a proprietary model versus frontier or open-source ones vary because cheaper alternatives help most on simple tasks, but complex, high-quality workloads keep per-inference costs high, so total savings often range from a few percent up to around 30%.
- Regulatory barriers, not engineering limits, are the key bottleneck to building more nuclear because safe reactor designs exist but slow, politicized licensing and extreme guarantee demands make new construction costly and time consuming.
Regulatory barriers, not engineering limits, are the key bottleneck to building more nuclear because safe reactor designs exist but slow, politicized licensing and extreme guarantee demands make new construction costly and time consuming.
- Capping insurer profits can create perverse incentives because when margins are limited firms must grow the premium base or allowable spending to increase absolute profits, which encourages higher total spending rather than efficiency.
Capping insurer profits can create perverse incentives because when margins are limited firms must grow the premium base or allowable spending to increase absolute profits, which encourages higher total spending rather than efficiency.
- Even when models can run on‑prem, geopolitical trust blocks adoption because companies fear hidden backdoors or political risk from models originating in adversarial jurisdictions.
Even when models can run on‑prem, geopolitical trust blocks adoption because companies fear hidden backdoors or political risk from models originating in adversarial jurisdictions.
- Enterprise AI usage follows a power law because a few teams experiment intensively and consume most tokens while the majority stick to basic summarization, concentrating costs and advanced use cases.
Enterprise AI usage follows a power law because a few teams experiment intensively and consume most tokens while the majority stick to basic summarization, concentrating costs and advanced use cases.
Keep the best of 20VC with Harry Stebbings
A free account saves these cards to your Collection, and Korva resurfaces them so you actually remember.