The bubble burst is now under 18 months away. Under could mean three or a maximum of 18 months. Bubbles do not taper. They burst. And till they burst they grow. If they grow fast, more and more people enter, egged on by the high returns notionally realised by the early movers.
Till the music stops.
All valuations assume a 10x improvement in price performance every year. Or 1 will become 100 in two years. Somewhere between the 1 and the 100, the bubble will burst. Nothing becomes 100x in two years. Because there is a diffusion time, an acceptance time and then a learning curve.
What we are seeing is not the result of any productivity gain realised or any viability achieved – it is usual Wall Street hype playing the American drums. Chips don’t improve at 100x in two years. They improve as per Moore’s law. The most powerful artificial general intelligence that exists is not even comparable to a domestic pet – leave aside a human kid. However, as a computer it is very good – the result of improvements in chip technology, automation of the software industry and the sudden availability of stolen data available with BigTech. They call it Big Data.
Yet one needs to understand that AI is a disruption like no other. Not for technological reasons but for other reasons. First, it signals the end of Big Tech. Apple and Microsoft are not even runners in the race. Meta, Google and X have access to self updating data and that makes them relevant. Till they are not.
But finally it is the new players like Anthropic or others who don’t have legacy revenues to protect or migration challenges – who will win. And whilst the next Googles and Facebook successors arrive, it is the bubble and not AI that is showing results.
People are using AI apps because the smartphone is making them accessible. And this generates some revenue. Not enough to make the data centres viable. But enough to make Nvidia profitable and reinforce his rhetoric that progress of a nation is proportional to its contribution to Nvidia profits. India has bought into their rhetoric – but fortunately won’t lose much because it doesn’t have much money. The day someone can make a Netflix of AI models – all models are dead. An eat all you can will anonymise the user and take away the model’s ability to directly extract revenues. If data centres have to pass costs to customers – there won’t be any customers left.
The radical price of the internet is free. And barter enables tax efficient operations. Google maps barters location data for right to use the maps. Meta sells your attention in return for letting you use the app free. Finally government regulators will catch on and transfer pricing will get enforced. Already, no country in the world allows tax free barter. You have to pay taxes on what you buy and sell. Even if the net is zero.
You can fool some people all the time.
And all people some of the time.
But you can’t fool all, all the time.
