The intense pressure to build more powerful models has made frontier AI labs desperate for any performance edge. This has led them to partner with and buy from hardware startups before they have a finished product, providing invaluable early validation and revenue that was previously inaccessible to them.

Startups can make big bets on emerging workloads, like LLMs before they were proven. This is a product risk. In contrast, incumbents like Google or NVIDIA must ensure their next chip serves a wide range of existing customers, forcing them to be more conservative and avoid disruptive product bets.

A robust VC strategy is to identify an inevitable trend, like AI, and invest in the infrastructure that will power it. This means avoiding downstream applications, which are competitive, and instead focusing on upstream suppliers that the entire ecosystem will depend on, ensuring relevance regardless of which application wins.

AI accelerator startups often optimize for the dominant model architecture at design time. However, by the time their chip launches years later, models have evolved (e.g., using smaller matrix multiplies), rendering the specialized hardware inefficient compared to NVIDIA's more adaptable GPUs.

When a new technology stack like AI emerges, the infrastructure layer (chips, networking) inflects first and has the most identifiable winners. Sacerdote argues the application and model layers are riskier and less predictable, similar to the early, chaotic days of internet search engines before Google's dominance.

To stay relevant, tech platform companies must obsessively follow developers and startups. They are the primary source of insight into emerging workloads and platform requirements. This isn't just for partnerships, but for fundamental product strategy and learning.

Unlike pure software, the value in physical AI and hard tech comes from long-term compounding of technology. Startups often fail because they don't survive long enough to see these returns. This makes early commercial discipline and constraints crucial for longevity.

While a dataset may skew towards tech-forward businesses, this is a feature, not a bug. These early adopters signal where the market is headed, allowing for predictive insights into future technology trends before they become mainstream.

Most current VCs come from software backgrounds and lack the deep hardware expertise of 90s-era investors. This knowledge gap creates an arbitrage opportunity for those who can properly vet semiconductor and networking startups, avoiding hype cycles around inexperienced founders.