Reform in Practice: Tanuj Bhojwani (South Park Commons) on Government as a Buyer of Technology

What is one area of regulation in India that should be deregulated?

So, I work in AI, it's very obvious to everyone that data centers and compute are going to be the bedrock of this, you know, coming technology wave. It is still ridiculously hard to set up a data center in India. What makes it even harder is that somehow if you get past all the requirements on getting the power line, getting the land, getting all the utilities in place, all the safety requirements, fire requirements, and you actually set up that data center, the biggest problem is that unless you make a data center of a reasonable size, which is a hundred racks or so, which requires a huge upfront investment, you are not even going to be eligible for government tenders.

You're not, you know, going to be METI impaneled. So especially when it comes to a sector like technology, what we are asking from the government is not even subsidy, what we are asking for is true demand, right, true captive demand for Indian startups, which can only come from, say, the government or sectors like banking, etc. But the government has made the ability to be a supplier of compute, to be a data center of note to this sector, very, very hard, right.

The best thing that government can do for AI is not take 10,000 cores and give GPUs, it is to say, look, as a government, we are going to use AI to transform, we're going to use AI to deliver better services. And our total outlay on AI, therefore, will be, say, 10,000 cores, maybe probably even more. That outlay is going to go to startups, to new companies, we're not going to put the same old procurement rules, we are not going to check for how many previous government contracts have you had, whether you've done 25 cores of sales in the last year, we're going to decide based on whoever's technology has the best outcomes.

If they could just make this change, the government demand could do much more for this industry and its growth than government subsidy or grants ever could.

What would a more effective alternative look like?

The procurement of intelligence in the government, right, is very, very different from the procurement of physical goods, and definitely different from the procurement of software. Because in software, earlier, the way we procured software was procuring, you know, person hours of labour from large services company, and we told them what we'd want.

And, you know, there's this RFP with functional requirements, non-functional requirements, there's a whole list of things they ask for. And anyway, they botch it up, really, honestly, you know, you have stories like the income tax website, etc. Intelligence is going to look even worse, because there is no good way, even for the top leading labs in the world, like OpenAI Anthropic, to know how good their model is.

They're also figuring out through trials. Now, if you are a small startup, right, and you've done something really innovative, and you've beaten these labs at something, you are unlikely to become a vendor to the government, mostly because the government doesn't know how to procure intelligence, and how to define problems and niches where intelligence could really leapfrog some service delivery that they have to do. And then, while in OpenAI may not put in a lot of resources to fill that niche, but some startup in India who cares deeply about, say, farmers or education in Indian language context, may put in the effort and the investment to do that.

But if the government can't pay them or doesn't, you know, find a way to give them money for providing the service, then that market will never get created, those players never get created, and the chances of us one day having a frontier tech lab or technology company all get washed up, because we don't know how to use the government as a buyer, we only think of it as subsidy or grants, etc. We never think of the force of government as being a procurer.