The valuation sweepstakes
Don’t translate into revenue
Projections are mostly delusional
More fiction is written in excel these days than in word. Ai has now become so good that it can generate a spreadsheet that arrives at any target valuation.
One of our highly respected IIT think tank leads Prakash Hebalkar send me the Bloomberg article on forthcoming Chinese ipos for foundational ai players.
What stands out is that the projected revenue numbers are very small. Sub USD 100m per year. On the other hand the Silicon Valley majors talk of USD 10 billion. Whilst it is true that like Hollywood, Silicon Valley is far more global and dominant – the Chinese market is not one to be sniffed it. It is way way more than 1% of the global market in anything. It would thus seem that the revenue potential may be highly inflated and may be out of sanity range. This is broadly on expected lines.
But let’s look at the implications for India. We don’t have a trillion dollars to loose. So there is no risk of losing it. There are over zealous politicians and technically illiterate industrialists who will announce GW scale neoclouds. But for now it is only land allotment. The businesses will pocket the land and the politician gets some shouting points. Nothing is actually going to be build.
Land and capital are in short supply. People and dreams are not. Politicians have to seek re election.
End result is that some states will announce subsidies which add up to 100 times the total income of the state and smart businessmen will grab land.
But the answer to ai leadership is not in bullish PR or land mongering. It will have to be in technology. That is where we are missing the plot. Sitting on a laptop with a good internet connection doesn’t cost too much money. We have millions seeking employment. So what can they do.
One option is that we deploy them in teaching ai models. There are a lot of fancy words used like supervised learning, reinforcement learning etc. But net net it is akin to teaching a circus animal new tricks. It is a lot of effort and labour. Now what tools do we need to provide for the millions to earn. We need to give them compute power and storage. Facilities typically provided by a data centre.
The answer is not to fund expensive laptops. We need to allow top end functionality with entry level terminals. The way to do so is thin clients working on a client server architecture. The personal device is just a screen with input devices. The compute power and storage is in the cloud – physically in a data centre.
As a country it doesn’t make sense to invest in neoclouds. We don’t have the capital or the weather or the indigenous equipment. We don’t even have surplus green electricity. We should not join the race to lose money.
The Americans have more capital than talent. Capital bleed should be left to them. We need to focus on asset light models. And employment generating models.