ESDS IPO: Is It Really an Infrastructure-as-a-Service Business?

Quick Summary: ESDS Software's ₹5,028 Cr IPO is built on a "landmark $1.25 billion AI contract." Read closely, that number is not revenue. It is a take-or-pay purchase commitment ESDS owes to a Nasdaq neocloud, Sharon AI. ESDS must on-sell about ₹2,366 Cr of GPU capacity a year, roughly 5 times its entire FY26 revenue, against an end customer it has not named. The ₹1,176 Cr "cash" on the balance sheet is that customer's prepayment, not earned money. About 45% of FY26 profit comes from a first-year subsidiary, Spochub, with a 63.5% margin the RHP never explains, and the margin itself is propped up by an under-depreciated base the IPO's own ₹576 Cr capex will correct.

The headline number is a purchase commitment, not a win

ESDS Software Solution Ltd. is one of only two Indian players offering the full spectrum of GPU-as-a-service, cloud, managed services, data-centre infrastructure and software, and the larger of the two by FY26 revenue at ₹472.21 Cr. It operates five data centres, serves 2,501 customers and employs 993 people. On the surface the financials look exceptional, and that is precisely the first thing to distrust.

The central number in the offering is a "landmark" US$1,250 million agreement with an "Australia-based neocloud AI compute service provider," total contract value ₹11,831.25 Cr (at the RHP's ₹94.65/USD rate). The issue was 28.31 times subscribed on the final day, and a grey-market premium implied a listing near ₹750 against the ₹429 band.

ESDS is the buyer on it, not the seller. Sharon AI's own SEC filing calls ESDS "the customer," with a customer option to extend. The ₹11,831 Cr is what ESDS has committed to pay, take-or-pay over five years, about US$250 M a year, monthly in advance, secured by US$140 M of letters of credit and bank guarantees, with no termination-for-convenience in the first 36 of 60 months.

CALCULATION: that commitment is about ₹2,366 Cr a year, roughly 5 times FY26 revenue (₹472.21 Cr). It is about 9.8 times the ~₹1,210 Cr "net cash" headline and about 25 times revenue. The reported financial strength of this IPO coexists with an off-balance-sheet purchase obligation nearly ten times its net cash.

Where the cash actually comes from

Strip the balance sheet down and the picture changes. Consolidated operating cash flow in FY26 was ₹1,367.71 Cr, and reported "free cash flow" was ₹1,241.82 Cr. These two numbers are the basis of the "cash machine" read.

CALCULATION: ₹1,176.64 Cr, or 86.0% of that operating cash flow, is one customer's advance. The balance-sheet footnote is explicit: cash jumped from ₹60.68 Cr to ₹1,253.39 Cr in FY26, and ₹1,187.58 Cr of that is a contract liability, an advance from a customer for a GPU-as-a-Service project, to be amortised only after go-live, with revenue recognition beginning only in Q3 FY27.

So the "net cash of about ₹1,210 Cr" is substantially not owned money. It is a liability against un-delivered service. Take the advance out and genuine net cash is about ₹33 Cr; earned free cash flow is roughly ₹65 Cr, not ₹1,241 Cr. The PAT-to-OCF ratio is inverted, ₹120.82 Cr of PAT against ₹1,367 Cr of OCF, a cash-conversion of about 9%, because a huge unearned liability swamped the P&L. The "cash exceeds PAT" observation is a classification artifact.

The ₹1,176 Cr advance: a near-perfect backing of what ESDS owes

The advance and the purchase commitment are two legs of a resale chain, and the money points in opposite directions. ESDS is simultaneously a customer of Sharon AI (paying for GPU capacity) and a supplier to its own end customer (selling GPU-as-a-service). The advance is what ESDS received from its own end customer, an "enterprise customer incorporated outside India," contracting through ESDS subsidiary Spochub. Sharon AI never paid ESDS; the flow on that leg is the reverse.

CALCULATION (at ₹94.65/USD): the ₹1,176.64 Cr advance is about US$124.3 M, and the US$140 M of letters of credit ESDS must post to Sharon AI is ₹1,325.1 Cr, so the customer's prepayment covers about 88.8% of the collateral ESDS owes. Whether designed or coincidental, the end customer's money almost exactly backs ESDS's purchase-leg obligation. That is the signature of a customer-funded pass-through: the customer's money sits in ESDS's bank, ESDS uses it to meet its obligations to Sharon AI, and it only becomes earned revenue once the cluster goes live.

This is simultaneously reassuring and sobering. Someone wired ₹1,176.64 Cr, about 2.5 times ESDS's entire FY26 revenue, to a company that had not yet delivered anything. That is not a paper contract; it is a genuine economic commitment. But it is not ESDS's money and not profit, and its refundability is undisclosed. The RHP says only that it "shall be amortised after the go-live date." What happens to ₹1,176 Cr if the deal fails to go live is not disclosed in any retrieved source.

The end customer: the one fact that decides everything, and is not disclosed

The single largest open question of the entire IPO is who ESDS resells this capacity to, at what price, for how long, and whether the onward contract itself is take-or-pay.

FACT: the onward contract's value, price, margin and duration are all undisclosed. The only figure is the advance. INFERENCE: the advance being about 9.9% of the purchase TCV is consistent with the onward deal being of a similar order of magnitude (a roughly 10% prepayment), but back-to-back matching cannot be confirmed. No figure larger than US$1,250 M, such as a commonly cited "about $1.95 billion" onward contract, appears in any retrieved primary source; the only AI contract value tied to ESDS anywhere in the RHP or Sharon AI's SEC filings is the US$1,250 M purchase agreement.

The gross margin between ESDS's resale price and its ₹11,831 Cr take-or-pay cost is invisible. No segment margin for the AI/GPU business is disclosed. Management describes the model as "using it to provide AI-computing services (not reselling)"; the spread is not stated. That is the load-bearing economic unknown.

Who bears the risk: a lopsided split

The utilisation risk is the sharpest asymmetry. ESDS must on-sell about ₹2,366 Cr a year of capacity, a 6.3 times jump from its roughly US$39.9 M FY25 revenue, with no publicly visible order book. That is a fundamentally harder and unproven proposition than the headline contract value suggests.

Spochub: about 45% of profit from a subsidiary the RHP never explains

The consolidated P&L hides the most concentrated part of the earnings story. ESDS consolidates three subsidiaries, including Spochub Solutions (India, 99% held). Spochub:

• Had zero revenue in both FY24 and FY25 and negligible losses.
• In FY26 posted revenue of ₹84.95 Cr and profit of ₹53.96 Cr, a 63.5% net margin.
• Contributed 44.66% of consolidated profit in FY26.
• Is the entity that received the ₹1,176.64 Cr GPUaaS advance.

CALCULATION: if Spochub's ₹53.96 Cr profit were non-recurring, reported PAT falls from ₹120.82 Cr to about ₹66.9 Cr and the IPO's P/E roughly doubles, from about 41.6 times to about 75 times. A 63.5% net margin is not a sustainable operating margin for any services or resale business; it is consistent with software licensing, a one-time gain, or revenue booked at near-zero cost, and the documents do not establish which. No standalone P&L, cost breakdown or customer detail for Spochub is in the retrieved corpus.

The margin is partly an artifact, and the IPO corrects it

The reported 49.6% EBITDA margin (₹234.22 Cr) is flattered by an under-depreciated asset base. D&A was remarkably stable, ₹52.55 Cr in FY24, ₹62.21 Cr in FY25, ₹63.46 Cr in FY26, even as revenue grew 30% and data-centre capex ran ₹79 Cr in FY26.

The IPO deploys ₹576 Cr (80% of proceeds) on cloud-computing equipment, servers, GPU servers, storage and networking, at the four existing Indian data centres (Airoli, Bengaluru, Mohali, Nashik). This funds ESDS's own Indian data centres, not the Australian cluster; that hardware sits with Sharon AI.

CALCULATION: adding ₹576 Cr of equipment on the historical ~9.6% depreciation rate lifts D&A by about ₹55 Cr a year, cutting PAT (post-tax) to about ₹80 Cr and pushing P/E from about 41.6 times to about 63 times; at the five-year useful-life end of the policy, D&A rises about ₹115 Cr, PAT falls to about ₹36 Cr and P/E climbs to about 141 times. The FY26 margin is not the run-rate margin once the IPO money is spent.

The core business is flatter than the headline

The 30.7% FY26 revenue growth is real, but look at which line is growing:

The flagship IaaS cloud platform is flat (+1.7%); the SaaS "recurring platform" line shrank 14.2%; the entire growth is managed services (+157%), a different, typically lower-margin, more labour- and project-like economic animal than the platform ESDS markets itself as. No segment margins are disclosed, so whether the mix shift is margin-dilutive is itself a transparency gap.

The customer base is also demonstrably non-sticky at the top. Top-10 concentration was 45.36% of FY26 revenue; the FY25 top client, a Russian BFSI entity at 20.15%, collapsed to 2.80% in FY26 on sanctions. A single political and regulatory event erased a fifth of revenue in a year. Government share is declining, from 34.04% to 27.37%. This is a live demonstration of concentration risk, not a hypothetical.

What the IPO price is already pricing in

At the ₹429 upper band (market cap ₹5,028.35 Cr), FY26 multiples:

On an advance-adjusted EV/EBITDA of about 21 times, ESDS is priced at roughly double the large-cap Indian infra and cloud names, with Bharti Airtel at about 10 times and Tata Communications at about 11.6 times EV/EBITDA. The P/S of 10.6 times sits far above Tata Comm's roughly 1.9 times and Airtel's roughly 5.2 times. The one true listed peer, E2E Networks, the other full-spectrum player the RHP names, trades at a negative P/E on an FY26 loss, so there is no listed comp validating a 41 times P/E.

What the price embeds (CALCULATION): to trade at a still-premium 15 times EV/EBITDA by FY28, ESDS needs FY28 EBITDA of about ₹333 Cr on about ₹671 Cr revenue, a roughly 19% two-year CAGR on top of a 49.6% margin that is itself propped up by an under-depreciated base and a first-year subsidiary profit. The IPO price assumes both rapid continued growth and that the exceptional margin is sustainable, the two things this investigation most strongly questions.

The grey-market-implied listing near ₹750 (market cap about ₹8,790 Cr) prices in far more: P/E about 73 times, P/S about 18.6 times, advance-adjusted EV/EBITDA about 37 times. That embeds near-total delivery of the Sharon AI contract at high margin.

What the market may be missing

1. The flagship deal is a purchase commitment, not a customer win, and the whole valuation hinges on an undisclosed resale. The ₹11,831 Cr is take-or-pay cost ESDS owes Sharon AI, about ₹2,366 Cr a year or 5 times revenue, backed by a US$140 M letter of credit. Easy to miss because the RHP describes it as a "strategic AI agreement," and only Sharon AI's SEC filing states the direction. Why it matters: the market prices ESDS as earning US$1.25 billion; the actual question is whether it can resell ₹2,366 Cr a year of capacity at a positive margin to an end customer it has not named, at a margin it has not disclosed.

2. The "net cash ₹1,210 Cr" is mostly one customer's unearned advance, about 86% of FY26 operating cash flow. Cash jumped from ₹60.68 Cr to ₹1,253.39 Cr, of which ₹1,187.58 Cr is a contract liability against un-delivered service. Easy to miss because "cash" reads as owned strength and the footnote attribution is buried. Why it matters: it turns the honest EV/EBITDA from about 16 times to about 21 times, and the durability of that "strength" depends entirely on one contract going live. Refundability of the ₹1,176 Cr advance is undisclosed.

3. About 45% of consolidated PAT comes from Spochub, a first-year, 63.5%-margin, unexplained subsidiary. Zero revenue in FY24 and FY25, then ₹53.96 Cr of profit (44.66% of consolidated PAT) in FY26. Easy to miss because the consolidated P&L hides it. Why it matters: it is the difference between a roughly 41.6 times and a roughly 75 times P/E. The single most important number to audit after listing is Spochub's standalone P&L.

4. The reported margin is partly an artifact of an under-depreciated base the IPO itself will correct. D&A is stable at about ₹63 Cr against ₹576 Cr of incoming equipment capex. Easy to miss because margins are reported before the depreciation wave. Why it matters: it adds about ₹55 to 115 Cr a year of D&A, potentially compressing PAT by 30 to 70% and pushing the P/E toward 63 to 141 times.

5. The revenue base is demonstrably non-sticky at the top. The number-one FY25 customer, a Russian BFSI entity at 20.15%, collapsed to 2.80% on sanctions; top-10 concentration is about 45%. Easy to miss because the FY26 top-client figure (15.93%) looks improved. Why it matters: it proves a single macro or regulatory shock can erase a fifth of revenue in a year, a live, demonstrated risk, not a footnote.

What remains uncertain. The end customer's identity, the onward contract's value, margin and take-or-pay terms, Spochub's profit driver, the refundability of the ₹1,176 Cr advance, and which party bears technology-obsolescence risk. None of these is established by the retrieved documents.

What to watch

(1) Does revenue recognition begin on schedule in Q3 FY27, and does the advance begin to amortise? (2) Does Spochub's 63.5% margin recur in its standalone Q1 and Q2 FY27 numbers? (3) The net-cash figure excluding the advance; is the balance sheet's strength real once the prepayment is stripped out? (4) Any disclosure of the onward customer contract and its margin, the one line that decides whether the resale model earns a spread or merely passes capacity through at cost. (5) Q1 and Q2 FY27 depreciation; does the ₹576 Cr capex start hitting the P&L, compressing the reported margin toward its true run-rate?

Frequently asked questions

Is the $1.25 billion AI contract revenue for ESDS?
No. It is a five-year take-or-pay purchase commitment ESDS owes to Sharon AI, about US$250 M a year paid monthly in advance. It is what ESDS must spend to buy GPU capacity, not what it earns. ESDS earns whatever margin it can achieve reselling that capacity to its own, undisclosed, end customer.

Why does ESDS have ₹1,176 Cr in advances if it is the customer?
The advance is on the other leg of the chain. It is money ESDS received from its own end customer as a prepayment for GPU-as-a-service, held as a contract liability until go-live. Sharon AI never pays ESDS; on the purchase leg, ESDS pays Sharon AI.

Who is ESDS's end customer?
Not disclosed. The RHP describes only an "enterprise customer incorporated outside India," contracting through subsidiary Spochub. Its identity, and the contract's value, price and margin, are all undisclosed.

Is the deal back-to-back matched?
Partially. The two legs are clearly linked, with the ₹1,176.64 Cr advance at about 9.9% of the ₹11,831 Cr purchase TCV, but the onward contract's value and terms are undisclosed, so a matched, margin-earning structure cannot be confirmed.

Why does the reported 49.6% EBITDA margin matter?
It is partly an artifact. Depreciation has been stable at about ₹63 Cr against an incoming ₹576 Cr capex wave from the IPO itself. Once spent, extra depreciation of about ₹55 to 115 Cr a year could compress PAT by 30 to 70%, pushing the P/E from about 41.6 times toward 63 to 141 times.

Data note

All figures are from the ESDS RHP/DRHP, ESDS and Sharon AI SEC filings, and IPO coverage retrieved for this article. FY26 figures are as reported. Multiples and derived figures (P/E, EV/EBITDA, CAGR, the advance-to-letter-of-credit ratio and the depreciation scenarios) are computed from reported data at the ₹429 upper band and the RHP's ₹94.65/USD rate. Segment margins, the onward contract's value and margin, Spochub's standalone P&L, and the advance's refundability were not disclosed in any retrieved source and are flagged as such. Grey-market premium and listing figures are market estimates from web sources, not official.

This is factual, descriptive analysis for educational purposes, not investment advice, and not prepared by a SEBI-registered Research Analyst.

Not investment advice.