Netweb Technologies: India's AI Proxy | The Spotlight ft. Vinay Menon

ELI5/TLDR

Netweb Technologies makes servers in India — the boxes that do the actual computing behind AI, supercomputers, and data centres. For 25 years this was a boring, low-margin business. Then AI arrived, server prices exploded, and Netweb’s revenue went up 15x in five years. The pitch from analyst Vinay Menon: it’s the cleanest listed Indian way to bet on the AI hardware boom, run by a frugal family that saw this coming. The catch he mostly waves away: it trades at roughly 130 times earnings.

The Full Story

What Netweb actually does

A server is a computer built to run non-stop — 365 days, no breaks — doing the heavy compute that sits behind everything from a bank’s records to a chatbot’s answers. Netweb has been building them in India since 1996, when founder Sanjay Lodha bet that “the future is compute” while everyone else was busy reselling desktops.

The company doesn’t make the brains. The chips come from Nvidia, AMD, and Intel; those are physically manufactured by TSMC in Taiwan; the underlying chip blueprint (the “architecture”) comes from ARM in the UK. Netweb’s job is to take those chips and design and assemble the finished server around them. Think of it as a high-end tailor who doesn’t weave the cloth but cuts a suit that fits one specific client perfectly.

That tailoring is the whole pitch. In raw volume Netweb is small — maybe 5-6% of the Indian market, dwarfed by HP, Dell, and Lenovo, who chase orders of ten or fifteen thousand identical machines. Netweb instead goes after the awkward jobs: a hospital running AI research on patient data, a high-frequency trading firm, a government lab. Custom requirements the giants can’t be bothered with.

Three engines, and the software trick

Revenue comes from three segments: HPC (high-performance computing — supercomputers for government and research labs, where Netweb started and built India’s fastest supercomputer, Airawat), cloud/enterprise (companies wanting to own their servers on-premises rather than rent from Amazon or Google), and AI (GPU-stuffed machines for training and running models). AI was under 10% of the mix two years ago; it’s now 25-30%.

A server is also designed as per that chip… if HP Dell can come out with servers on that chip, Netweb can also come at 3 months or sooner. And that is the capability.

The single most important thing that happened to the business was margins doubling from 5% to 10% around 2021. The cause was software. A normal server maker sells you the metal and walks away; you buy your software from someone else and glue it together yourself. Netweb decided to build its own software — five or six packages now — and bundle it in. One example is “GPU slicing,” which carves up an expensive graphics chip so a 40-person office shares it efficiently instead of some staff hogging it. Menon compares the ambition to Nvidia owning its whole stack (chip plus the CUDA software that runs on it). Software is a one-time R&D cost that then drops straight to profit, so every new package widens the margin.

Why the moat might be real

Two genuinely defensible things. First, the design partnerships with Nvidia, AMD, and Intel are decades old and slow to build — Menon reckons a deep-pocketed newcomer like Tata or Adani could fund a rival overnight but would still need 5-7 years to replicate the relationships and know-how. Second, server-building is unglamorous craft knowledge that AI can’t shortcut.

You have to actually have the knowledge of that 20-25 years to build those servers.

There’s also a tailwind from politics. “Make in India,” data-localisation rules (data generated in India increasingly must be stored in India), and defence sovereignty all push buyers toward a domestic maker. Menon notes India has one-tenth the server capacity of the US, and data-centre capacity is meant to grow from 1.5 to 5 gigawatts by 2030 — implying millions of new servers.

The financials have followed: 35-40% revenue growth (which the company reportedly beats), 35-40% return on capital, now debt-free with healthy cash flow, and — unusually — entirely family-funded, no venture capital or private equity ever. Menon frames that frugality as a virtue: a 2-5% margin business punishes anyone who overstretches, and this family never did.

The honest risks (mostly raised by the host)

To his credit, host Krish Kothari does the pushing. The supply risk: Netweb is at the mercy of Nvidia and AMD. Menon’s answer is that a chip shortage hits every server maker equally, so it’s an industry problem, not a Netweb-specific one — prices rise for all.

The cyclicality risk is the more interesting exchange. Menon openly concedes hardware prices have run to “unusual” levels and “have to come down,” which would dent demand for a year or two.

The valuations are unrealistic and the hardware prices have really dampened that kind of demand momentum.

His comfort blanket is diversification: if AI cools, the localisation theme and a future government supercomputing mission (NSM-2) keep the other segments humming. He invokes the dot-com bust — software valuations crashed, but the technology itself won — and bets AI plays out the same way.

Valuation, gently sidestepped

Then the elephant. Kothari notes the stock trades around 130x earnings and asks, essentially, are you saying it’s still cheap? Menon doesn’t defend the multiple head-on. He pivots to growth: you’re buying a 35-40% grower with optionality toward 55-60%, best-in-class returns, so “any kind of dip is where you should really look at it.” He recalls recommending it at ₹1,800 and 80x trailing earnings when people called it expensive — and they were wrong. The long-term story, he says, is intact. He thinks it can be a Nifty 50 company within a decade.

Key Takeaways

Netweb makes servers in India; it designs and assembles around chips from Nvidia, AMD, and Intel but makes no silicon itself.
Revenue grew roughly 15x between FY21 and FY26; the company guides 35-40% CAGR and reportedly exceeds it.
Three segments: HPC/supercomputing, cloud/enterprise, and AI. AI jumped from under 10% to 25-30% of revenue in two years.
Margins doubled from ~5% to ~10% around 2021, driven by bundling in-house software (e.g. GPU slicing) — a one-time R&D cost that lifts margin permanently.
Return on capital is ~35-40%; the company is now debt-free and entirely family-funded, never took VC or PE money.
Volume share in India is only ~5-6%; the strategy is custom solutions, not competing with HP/Dell/Lenovo on bulk orders.
Government is ~45-50% of revenue (supercomputing, ISRO, DRDO, defence, IITs) and likely stays that high.
Disclosed private clients include TCS and Infosys; BFSI and IT are the strongest private verticals.
R&D is only 3-5% of revenue because the heavy chip R&D is done upstream by Nvidia/AMD/TSMC — Netweb only does the software and integration layer.
Moat = decades-old chip design partnerships plus scarce server-building expertise; a new entrant would need 5-7 years even with capital.
Post-IPO, Netweb brought manufacturing in-house (its own SMT plant, Nvidia-certified) and doubled headcount; the plant is far from full utilisation.
Future optionality cited: space data centres, quantum computing (needs HPC server clusters), and homegrown AI startups like Sarvam.
Stock traded around 130x earnings at recording; management/analyst frame any dip as a buying opportunity.

Claude’s Take

This reads as a promotional pitch, and it’s worth flagging plainly. Vinay Menon is a sell-side analyst whose firm published a buy note on the stock; the host asks good adversarial questions, but every answer bends back toward “and that’s another reason to own it.” The space-and-quantum-computing detour is a tell — when an analyst starts pricing in data centres in orbit, the valuation conversation has left the building.

That said, the business itself sounds genuinely good, and the bull case isn’t fabricated. The hardware-to-software margin story, the 35-40% returns on capital, the debt-free family ownership, the real switching costs in mission-critical servers — these are substantive, not spin. The single most honest moment is Menon conceding that hardware prices are at “unrealistic,” “unusual” levels that “have to come down,” which is an admission that a chunk of recent growth is cyclical sugar.

The unresolved question is the one Kothari asks and Menon never really answers: 130x earnings. Even a flawless 40% grower has to grow into that number for years before a buyer at today’s price makes money, and the analyst’s own cyclicality warning suggests a one-to-two-year air pocket is plausible. “Buy the dip” is doing a lot of load-bearing work here.

Score 6: a clear, well-structured education on a real and interesting Indian business, dragged down by being a one-sided bull pitch on a stock priced for perfection. Useful for understanding the company; not a substitute for primary research on the valuation.