Korva is a scroll of the internet’s best ideas.
Everything works, but most companies won't
The popular framing asks how to cut up the pie: how much goes to the labs, how much to the semiconductors, and whether one lab takes 98 percent of the value and does all the drug discovery. That view doesn't hold. The major cloud providers do well, some of the newer GPU clouds do well, and companies like Fireworks do well. NVIDIA does well, and some chip startups do. Inference runs on phones, near the edge, and in big data centers.
That does not mean every company in those areas works. Most do not. Real differentiation matters more than ever, which means taking each idea to its logical extreme and going all the way on it.
Why do most companies fail in booming markets?
Growth in a market attracts far more entrants than it can support. The early automobile industry had hundreds of makers in the United States by the 1920s, and only a few survived. During the dot-com boom, internet use kept climbing, yet Amazon's stock fell more than 90 percent from its 1999 peak to 2001, and many other internet start-ups shut down.
How does AI inference run on phones?
Phones run smaller models, often compressed through quantization, which stores weights in fewer bits. Dedicated hardware helps. Running on the device keeps data local and removes the delay of a network round trip. Apple added its Neural Engine to the A11 chip in 2017 to run machine learning efficiently.
Why does differentiation matter in crowded markets?
Michael Porter's 1980 book Competitive Strategy described three generic strategies: cost leadership, differentiation, and focus. He argued that a firm without a clear position on any of them is stuck in the middle and earns weak returns. When many firms offer the same thing, price becomes the only difference. Porter was a professor at Harvard Business School.