Eric Vishria - A Decade of Lessons Investing in Software & Hardware - [Invest Like the Best, EP.486]

Despite using the same open-source models and NVIDIA hardware, specialized infrastructure companies like Fireworks achieve a 5x performance advantage over major cloud providers. This shows running large AI models efficiently is a highly specialized skill, not a commoditized service, creating opportunities for focused startups.

Investors initially dismissed AWS as a low-margin commodity, then later feared it would monopolize software. Both views were wrong. The cloud market was so vast it supported an oligopoly and large niche players (e.g., Snowflake). This same non-zero-sum dynamic is now playing out in the AI market, defying 'winner-take-all' predictions.

In the early 2010s, enterprises were highly skeptical of the cloud. Today, those same companies are actively experimenting with and spending on AI. They perceive it as a more significant opportunity and threat than the cloud was, having learned from their past hesitation, creating a massive demand-side pull for AI solutions.

Traditional product management separates customer problem discovery from technical implementation. In AI, this model fails. Winning teams must deeply understand the nuanced, 'jagged edge' of what models can and can't do, building products that bridge that specific, shifting capability frontier with customer problems.

In the AI era, formal job titles are less relevant than core competencies. The most valuable individuals are those who combine three key traits: a deep understanding of customer problems, strong product taste, and a nuanced grasp of the constantly shifting, 'jagged edge' of AI model capabilities.

Historically, the difficulty of migrating applications made databases incredibly sticky, creating a moat for incumbents. AI agents can automate this monotonous work because database interfaces are well-specified. This shifts the basis of competition from lock-in to cost, zero-to-infinity scaling, and iteration speed.

For established SaaS companies, hitting quarterly plans based on an old playbook is a value-destroying activity during the AI transition. The management team's muscle memory is built around execution, but the moment demands they invert their focus from the core business to figuring out AI, even if it hurts short-term results.

Experienced sales leaders from legacy tech companies often fail at breakout AI startups because their playbooks, like quota capacity models, are designed for pushing demand. When an AI product feels like 'magic,' it pulls demand in. Old assumptions about rep productivity become constraints, not effective models for growth.

Investing in deeply technical hardware, like Cerebras's wafer-scale chip, requires a degree of ignorance about the true difficulty. Unlike software, where a core technical insight gets you 80% of the way, in hardware it's only 2%. The rest is a brutal, multi-year battle against physics and complex supply chains that experts might avoid entirely.

A powerful filter for venture investing is the 'life's work' test: can you, with intellectual honesty, recruit a person you deeply care about to join the company, framing it as a career-defining opportunity? If not, the project may lack the scale of ambition and meaning required for a truly great outcome.

Traditionally, venture investors sought high multiples (e.g., 100x) exclusively at the early stage. However, the sheer scale of modern tech outcomes has changed this. The opportunity for high-multiple returns now extends into later stages, allowing growth funds to pursue the same outlier returns previously confined to seed investing.