Charlie Songhurst

Then AWS and Azure come along.

And critically, as a startup, you don't know your own future demand curve because you don't know whether you can get product market fit.

So you either had to buy too many servers, in which case you waste capital, or too few, in which case you couldn't beat demand.

Whereas with AWS and Azure, they've effectively taken that risk out of the equation.

And you can almost know that your cost base only scales with your success.

Is the entire industry just sitting on that?

I kind of had an intuitive feel it probably is.

Better luck than good.

Maybe we get an AI boom in the next couple of years.

Maybe that will be written about as that breakthroughs in ML.

Maybe it's actually NVIDIA's Ampere chips and the emergence of sort of five and seven nanometer nodes just drive improvements in throughput that drives ML up.

So it's nothing to do with the companies that are actually working on it.

It's to do with the pickaxes being turned from wood to iron, making their work far more efficient.

But NVIDIA will probably capture some value of that, but probably won't capture the greatest of value.

Same if you were to write a history of early computing in the 90s and 2000s, it'd be interesting to write it from the importance of the evolution of Intel chips and modem speeds as the primary drivers, as opposed to the narrative history of the growth of these great companies.

Oh my God, there are so many.

There are so, so many examples.

I think it's probably a recurrent one of often my enthusiasm has been greater than my competence.

And it's the people that bet on the enthusiasm more than the competence.

I'm internally grateful to them.