Incumbents improve exponentially (e.g., 2x annually). A startup taking 5 years to reach scale needs a 100x better product at launch just to be competitive, accounting for the incumbent's 32x improvement during that development time.
When creating a fundamentally new product like Cerebras's wafer-scale chip, supporting vendors don't exist. This forces you to own the full stack, from chip to software. The expertise gained by building this ecosystem yourself becomes a powerful, hard-earned competitive advantage.
During the long, expensive period before a hardware product works, progress can feel stalled. The key is methodical problem-solving: ensuring each failure is a new one, not a repeat. This demonstrates progress to the team and investors, even without a final working product.
The core AI supply constraint is a fundamental mismatch in velocity. AI model demand grows exponentially at the speed of software, but building the necessary infrastructure (fabs, data centers) moves at the slow, multi-year pace of massive construction projects.
In a rapidly evolving field like AI, hardware architectures must be flexible. Cerebras succeeded by accelerating the underlying algebra of AI, not a specific model type like CNNs. This forward-thinking decision allowed their chip to excel with transformers, which were invented after their architecture was set.
Instead of a rigid long-term plan, operate like an internet router. Focus on reaching the next milestone ("hop"). This provides a new vantage point from which to re-assess the landscape and determine the optimal next move, allowing for agility in an unknowable future.
The source of NVIDIA's dominance isn't just CUDA or its chip architecture. It's the intense, gritty culture developed during a decade (2004-2014) as an underperforming public company. This relentless "fight" is in their DNA and is a more powerful moat than any single technology.
Founders of challenger companies should embrace being a "professional David" battling Goliath. This underdog mentality turns external skepticism and shifting goalposts into fuel. Each "you can't do that" moment becomes a new mountain to climb, strengthening the team's resolve and focus on doing the impossible.
Focusing on speed is a creative act, not just an optimization. There was no market for "slow search." When internet speeds increased, Netflix didn't just mail DVDs faster; it became a movie studio. Radical speed improvements enable entirely new business models and applications.
In highly technical hardware companies, a board member's role isn't to solve engineering problems. It's to operate within their own circle of competence: keeping the company financed, asking strategic long-game questions, and providing the patience required for multi-year R&D cycles.