The winners in AI from India will approach AI as an infrastructure, not as a product: What exactly does Alibaba’s ex CTO mean by “Infrastructure”?

Date: 08-03-2025 at 05:18

I started this essay titled “Personalisation is a tool to make quality healthcare affordable for the masses, but personalisation is not what you think”, but ended up with the above title.

The old title reflects Jilo Health’s July learning - it is unbelievable that we can challenge the decade-old one-size-fits-all healthcare system with just basic healthcare data.

I have visited doctors (gynaecologist, General Physician, and psychiatrist) with my partner three times, in the past four months, and all these doctors have written the same lab tests. We were shocked when a psychiatrist wrote the CBC and Malaria lab tests, can you believe?

There are incentives for Healthcare professionals and Hospitals for writing unnecessary lab tests. And in the absence of basic Healthcare data, independent diagnosis companies are forced to offer one-size-fits-all health packages - full-body checkups.

There is some level of personalisation to improve customers’ experience, which is not the objective of the masses in any industry, including Healthcare. When a customer in Patna sees a full-body checkup ad, the first question that strikes their mind is Why do they need this full-body checkup? Let’s say, without knowing why, if they opt for a body check-up, the next question will be on what basis this package includes the lab tests and the biomarkers. Ultimately, the package is one-size-fits-all.

When we think of personalisation in Healthcare for the masses, it is so beautiful; it is all about outcome at a lower cost. This lower cost is not at all about the one component like treatment (of course, this is one), but it tocuches every aspect - discovering a chronic disease before the final stage, the identification of the best lab tests, as part of full-body checkups, out of 3000 lab tests, the identification of best generic drugs (which is being sold as a premium of up to 90%), the best recommendation of food, exercise, life style changes that is possible based on the the masses financial and geographics, the best recovery protocol that can prevent the masses to readmit in hospital etc. (the list goes longer).

To an extent, personalisation for the masses is also about minimising the non-medical expenses, which in some cases are up to 35% of medical expenses.

This is a complex problem to solve, but today is the best time in history to solve this for the masses.

Unfortunately, there is almost nothing we can do about doctors and hospitals writing unnecessary (one-size-fits-all) lab tests, but a full-body checkup can be personalised to an individual, at a marginal cost, to make it affordable with proper outcome.

Today, it is possible all thanks to AI and Data. I love the concept of infrastructure-driven AI - AI is not about foundational models, it is about building infrastructure that can lead us to use at a population scale for outcome-driven use-cases.

Again, there should be no confusion on infrastructure - data centres and GPU clusters. I think the companies that are developing foundational models will take care of that. We need to think of infrastructure from the use-cases POV that can bring quality outcomes at much lower cost.

The basic healthcare data that can make full-body checkups, and the rest of the Healthcare delivery system personalised, affordable, and evidence-based, would be one of the infrastructures that we need to build for the masses.

Therefore, Alibaba’s ex-CTO is correct with his assessment of thinking of AI as infrastructure rather than a product. What do you think?

Read the full essay: https://www.sumanjha.com/post/the-winners-in-ai-from-india-will-approach-ai-as-an-infrastructure-not-as-a-product-what-exactly-d

You can watch the interview

Doubling down on thinking of AI as infrastructure, not as a product

Date: 09-19-2025 at 11:06

Global consultancy companies spending billions of dollars on India’s BPM (Business Process Management) companies for the workflow data is a great example of how, in the coming time, most of the outbound customer conversations will be through agentic AI.

Indian BPMs have workflow data of global companies (therefore developed consumers) accumulated for the past 30 years. However, when we talk about the same scale for the Indian consumer, the scenario is different.

Let’s take the example of Healthcare:

  • The lowest insurance penetration
  • Less than 10% digitised health records
  • 90% of facilities are fully or partially offline
  • 20+ local languages
  • Less than 20% English penetration
  • Zero standardisation

In all of these scenarios, when we talk about building agentic ai to solve the above problem, the math, somehow, doesn’t make sense.

Now, imagine if you are thinking of this “AI as an infrastructure”

  • How do you build infrastructure (physical + digital) to record the subjective (symptoms) of billions of Indians speaking 20+ languages?
  • How do you build infrastructure (physical + digital) to create and save clinical data (vitals, labs, parameters etc.) for billions of Indians?

If you can build the above two, now you can use AI to do magic - I mean, real magic.

  • Train voice AI to identify potential diagnosis based on objective (vitals) and subjective (symptoms), and keep humans in the loop only to validate.
  • Automate most of the workflow related to insurance - personalised premium, claim approval, claims settlement, Medical Coding etc.
  • Hyper-personalised insurance
  • Identify demand and build health infrastructure
  • Leverage millions of semi-trained healthcare personnel to eliminate the health professionals supply deficiency.

Infrastructure + AI Agents + Scale = Affordable quality healthcare for billions of Humans!