The patent before the prototype, and the long shape of Indian deep tech.

Vishal opens with the diagnosis that frames everything else in the conversation. India is the largest exporter of services and one of the largest exporters of software talent. It has not, in any matching sense, learned to ship physical product that the rest of the world buys from it. Raghu doesn't argue with the diagnosis. He sharpens it. The gap is not capability. Indian engineers built the Mars Orbiter and the Chandrayaan landers. What is missing is an investor and procurement environment that funds a five-year build before a single revenue dollar.

The reframe matters: hardware is not waiting for a clever founder. It is waiting for an institutional patience that no Indian fund has yet supplied at scale. Anscer exists in the small window where that patience is finally beginning to appear, at funds like Speciale Invest and the new deep-tech vehicles inside Lightspeed and 3one4.

Raghu spends a careful minute pulling apart the two categories most people in the conversation will conflate. A passenger AV operates in an open world with unbounded edge cases: pedestrians, weather, regulation. An AMR operates in a constrained one: a warehouse floor, a hospital corridor, a factory aisle. The planning problem is different, the safety case is different, the regulatory path is different, and the unit economics are completely different. AMRs are profitable in 2026; passenger AVs are still subsidised research projects.

The deeper point: deep-tech founders in India should look for the constrained twin of every glamorous unsolved problem. The constrained problem has a buyer, a contract template, an insurance market, and a path to gross margin.

The most counter-intuitive habit Raghu describes. Most Indian hardware founders treat a patent as a victory lap, something to file once the device is working and a customer has bought one. Anscer files the provisional months before the soldering iron is plugged in. The cost is small. The benefit is enormous: by the time the prototype exists, the IP is already cited in a USPTO database, the priority date is locked, and the next conversation with a strategic investor or a foreign acquirer starts inside a defensible position rather than next to one.

There is a quieter consequence. The discipline of writing the claim forces clarity about what the invention actually is. A team that cannot describe its novelty in patentable language usually has not yet found its novelty. The filing is a thinking tool, not just a legal one.

Raghu walks through the bill of materials of an AMR with the calm of someone who has rebuilt the spreadsheet a dozen times. The LiDAR is a commodity from a half-dozen vendors in Shenzhen, Korea and Germany. The motor controllers are increasingly Indian. The chassis is local. The wheels are local. The compute is the obvious imported expense and the firmware on top of it is the value capture. The question is not "can we make this in India," it is "which line of the BOM do we want to own and which do we want to buy in volume."

The strategic frame: every Indian hardware founder should be able to point at the BOM and say which two or three lines hold the patent moat, which lines are pure commodity, and which lines are the firmware-on-silicon hybrid where the IP actually lives. If you can't draw that picture, you don't have a hardware company; you have an assembly line.

The most useful piece of go-to-market advice in the conversation, told sideways. Anscer chose to chase a small pilot in a European 3PL ahead of larger, easier contracts in India. The pilot lost money on a per-deal basis. It made the next year of deals possible. The principle Raghu names: a reference customer whose logo is recognised by your next ten prospects is worth more than three customers whose logos are not.

For Indian deep tech specifically, the asymmetry is severe. A German warehouse signing off on an Indian AMR is, in itself, a market signal that the rest of the Western buying universe can read. The reverse is not true. Ten warehouses in Hyderabad do not buy you a single conversation in Stuttgart.

The line Raghu returns to twice. A European AMR vendor has the firmware and the brand. They do not have a service organisation that can land a field engineer in Mumbai by the next morning. Anscer can. Once a Western 3PL has rolled out a fleet in an Indian or Southeast-Asian facility, the switching cost is the service relationship, not the firmware. That is a moat that compounds with every robot shipped.

The further implication: Indian deep-tech founders should treat the service organisation as a first-class product, not an after-sales cost centre. Spares stocking, remote diagnostics, retrofit firmware and training programs are what turn a one-time hardware sale into a fifteen-year customer.

Raghu is candid about the appeal of defence procurement (DRDO, the iDEX program, the Make-in-India defence corridor in Tamil Nadu) and equally candid about the cost. Defence contracts have long evaluation cycles, high evidentiary bars, and brutal payment terms. They pay eventually; the working capital required to wait is the kind of money a software company never needs. The right way to approach defence is as a slow, prestige-laden second business unit, not as a near-term cash engine.

The deeper observation is about dual-use. An AMR designed for warehouses can, with a different chassis and a hardened compute stack, become a logistics robot for a forward base. The IP travels. The Bureau of Indian Standards work travels. The customer doesn't.

A quietly important observation buried halfway through the conversation. A decade ago, an Indian hardware founder could not, in a single trip across Bangalore, find a CNC house that did short runs, a sheet-metal shop, a PCB assembly line, a motor-controller vendor, and a firmware contractor who knew ROS. In 2026 they can. The stack is not yet at Shenzhen's density, but it is no longer the open question it was. The supplier ecosystem that ATL, Tonbo, Cyient, Bharat Forge and a hundred smaller shops created over fifteen years is what Anscer is drafting on top of.

The framing matters because every Indian deep-tech narrative now has to acknowledge what the previous generation built. The current cohort is not pioneering. It is the second wave on top of a supply base laid by the first.

Raghu is sober about the geopolitics. Yes, Western buyers are diversifying out of China. No, they are not doing it by ordering twice as many Indian robots. They are doing it by qualifying Indian, Vietnamese and Mexican vendors as second sources: small allocations, long qualification cycles, painful audits. The right reading is that the door is propped open, not flung open. The founders who walk through are the ones who can survive the eighteen months of qualification work before the first real purchase order arrives.

The corollary: any pitch deck that says "China-plus-one" without describing the qualification path is gambling on a tailwind it has not earned the right to use.

Raghu's reading of the Production Linked Incentive scheme is unsentimental. It is real money on the table, disbursed against actual manufacturing output, and it changes the BOM math at the margin. It does not, in any meaningful sense, decide which Indian hardware companies survive. A robotics business whose unit economics only work with the PLI rebate is a business whose unit economics don't work. The subsidy is a tailwind, not a thesis.

This is the thing he most wants the next cohort of founders to hear. Government schemes solve a financing problem, not a product problem. A company that gets confused about which is which will be in trouble the day the policy changes.

The single sentence in the conversation that should be quoted to every Indian SaaS investor who tries to fund a hardware company. Raghu says, plainly, that Anscer is a fifteen-year company. Not five. Not seven. He is not negotiating. The unit economics, the patent moats, the service organisation, and the certification stack each take three to five years to mature. A founder who privately believes they will be done in six is going to make decisions that destroy the company in year nine.

The frame travels. Founders should not adopt timelines because their investors prefer them. They should adopt the timeline the underlying physics demand and then find investors who can live with it.

One of the most useful pragmatic notes in the conversation. An Indian AMR cannot be sold into a German warehouse without CE marking, an ISO 9001 quality system, and a machinery-directive conformity assessment. None of this makes the product better. All of it costs eighteen months and a meaningful chunk of equity. The founders who underestimate this line item end up raising a "manufacturing" round that is in fact a "certification" round in disguise.

The mature reading: budget for certification the way a SaaS company budgets for SOC 2: a planned engineering project with a named owner and a real deadline. Treat the audit as a build, not a tax.

The most quietly humbling line in the episode. Raghu describes integrating an AMR fleet into a customer warehouse as a six-month project where the robot itself is two months and the four that remain go to the warehouse: the WMS, the racking layout, the safety procedures, the union conversations, the night-shift training. The product is technical. The deployment is sociological.

The structural implication for any robotics founder is that the team must contain people who have run a warehouse, not just people who have built one. Without that knowledge in the room, the company will design beautiful robots that the floor refuses to use.

Raghu separates the two phrases that most policy conversations conflate. Make-in-India is about the assembly line, the final-mile manufacturing step that adds five to ten percent of the value. Design-in-India is about the IP, and forty to sixty percent of the value sits there: the perception stack, the planner, the safety case, the proprietary firmware. India can be a Make-in country without being a Design-in country; many of its electronics factories already are. Anscer is built to be both.

The distinction matters because investors and policymakers measure each differently. PLI rewards the first. Patent filings and IP-backed revenue reward the second. A founder who confuses the two will optimise for the wrong incentive.

Closing his thoughts on capital, Raghu describes the test he applies to a prospective lead investor. Have they previously held a hardware company for nine years and made money on it? If not, if their wins are all SaaS, they will, with the best intentions, push for decisions that look like SaaS playbooks: aggressive marketing spend, headcount-led growth, multiple-expansion via narrative. Those moves kill hardware companies. The right fund has a track record of patience that no amount of conviction at the partner meeting can substitute for.

The advice cuts harder than it reads. Most Indian founders cannot afford to be picky about investors. Raghu's point is that, in deep tech, you cannot afford not to be picky.