> ## Content Index
> Fetch the complete content index at: https://www.moatmarginresearch.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# E2E Networks' EBITDA and Its Profit Tell Two Stories. Five Quarters Show the Margins Converging
- URL: https://www.moatmarginresearch.com/e2e-networks-gpu-cloud-machine-cost/
- Published: 2026-10-17T02:41:08.000Z
- Updated: 2026-10-17T02:41:08.000Z
- Description: At E2E Networks, a GPU cloud, the machine is paid up front and reaches profit only as depreciation. Five quarters show the two margins drawing closer.
- Author: A K Karthikeyan
- Tags: Moat Insights

## The idea in one sentence

In a GPU cloud the machine is paid for up front and reaches the profit and loss account only as depreciation, so E2E Networks' EBITDA margin (76.6% last quarter) and its profit tell different stories; five quarters of filings show the two margins converging, and point to what decides whether that continues: how long the machines last, how full they stay, at what price, and how they are paid for.

## The fact that makes a reader stop

E2E's press release for July to September 2026 leads with revenue of ₹223.8 crore, up 5.1 times, an EBITDA margin of 76.6% and a profit after tax of ₹72.7 crore, 32.5% of revenue; it also states depreciation of ₹74.0 crore \[PR\] \[RES\]. EBITDA is earnings before interest, tax, depreciation and amortisation: revenue less the running costs, before the cost of the machines. Here is that line, and what sits below it, for five quarters.

| ₹ crore                                       | Jul–Sep 2025 | Oct–Dec 2025 | Jan–Mar 2026 | Apr–Jun 2026 | Jul–Sep 2026 |
| --------------------------------------------- | ------------ | ------------ | ------------ | ------------ | ------------ |
| Revenue from operations                       | 43.8         | 70.0         | 95.6         | 156.8        | 223.8        |
| Running costs (services bought, staff, other) | 25.8         | 30.4         | 37.5         | 38.9         | 52.3         |
| EBITDA                                        | 18.0         | 39.6         | 58.1         | 117.9        | 171.5        |
| EBITDA margin                                 | 41.1%        | 56.6%        | 60.7%        | 75.2%        | 76.6%        |
| Add other income                              | 8.3          | 5.2          | 5.5          | 11.4         | 9.9          |
| Less depreciation and amortisation            | 42.8         | 47.6         | 51.3         | 60.6         | 74.0         |
| Less finance costs                            | 2.0          | 4.7          | 3.7          | 10.1         | 10.3         |
| Profit before tax                             | (18.5)       | (7.5)        | 8.6          | 58.6         | 97.1         |
| Depreciation as a share of EBITDA             | 238%         | 120%         | 88%          | 51%          | 43%          |

*Source: \[RES\] for Jul–Sep 2025, Apr–Jun 2026 and Jul–Sep 2026 (all unaudited, limited review); \[FY26\] for October to March (Oct–Dec 2025 is unaudited; Jan–Mar 2026 is the audited year less the unaudited nine months). Running costs, EBITDA and the last row are our arithmetic; EBITDA margins match the company's \[IP, p.15 and 19\].*

That depreciation was 43% of EBITDA. A year earlier the margin on profit before tax sat 83 percentage points below the EBITDA margin; now the gap is 33 points (our arithmetic). In FY26 as a whole it took more than all of it: ₹169.2 crore against EBITDA of ₹126.3 crore, 134% (our arithmetic) \[AR, p.18\]. For that year the EBITDA margin was 51.4% \[IP, p.20\] and the company reported a loss after tax of ₹15.6 crore \[AR, p.18\].

The loss also affected pay. The AGM notice states that, because of the FY26 loss, which it puts down to a "substantial increase in depreciation, a non-cash charge", the ₹3.20 crore paid to the three executive directors "had exceeded the managerial remuneration limits" of the Companies Act. The Act lets shareholders waive recovery by special resolution, and the board proposed a waiver, citing the directors' "significant contributions" \[AGM, p.21-22\].

## One machine, followed

Take the newest machines: the NVIDIA Blackwell B200 cluster. The company says it has about 5,100 GPUs live, "including 1024 B200s" \[IP, p.11\].

**The purchase.** At 31 March 2026 E2E held ₹533.45 crore of capital work in progress (machines bought but not yet running), all of it "Assets purchased in financial year 2025-26 however deployment is in progress" \[AR, note 3.1, p.150-151\]. The presentation says this was "CWIP of INR 5,334 Mn for GPU's, deployed in May 2026" \[IP, p.16\]. Our arithmetic: ₹533.45 crore for 1,024 GPUs is about ₹52 lakh per GPU, if the amount is the whole cluster; the filings do not say what else it includes.

**The cost, spread out.** The annual report gives "Computer Equipment" a useful life of 6 years, depreciated "on the straight line method", that is, the same amount each year \[AR, p.142\]. Over six years ₹533.45 crore comes to ₹88.9 crore a year, or ₹22.2 crore a quarter (our arithmetic, assuming no resale value). The cluster ran for about one month of April to June and all of July to September. Depreciation rose by ₹13.3 crore between the two quarters \[PR\]. Our reading: that is close to the extra two months of the cluster, about ₹14.8 crore on our arithmetic, so the six-year life appears to apply to it, though the filings do not say so directly. The managing director's letter speaks of capitalising "our B200 deployment" within FY26 \[AR, p.18\]; we follow the presentation and the asset note, which the filings do not reconcile with it.

**The customer.** On 31 August E2E told the exchanges it had "entered into the binding term sheet with a Sovereign AI company based in India" for "NVIDIA Blackwell cloud GPUs and allied services", with an "aggregate contract value of approximately Rs. 1,000 crore", before taxes, which "remains valid up to June 2029" \[TS\]. The customer is not named; the disclosure says it is not a related party \[TS\]. The CFO said it "transitions the NVIDIA Blackwell capacity we deployed in May end, 2026 from a pay as you go model to a longer term commitment" \[TSPR\]. A term sheet records the main agreed terms; the filings do not say a full contract has followed, though the 16 October press release also calls it "a multi-year cloud GPU contract" \[PR\]. Nor do they say what it covers: only the 1,024 B200s now running, or also the further "1024+ B200s planned" \[IP, p.11\].

**If the term sheet were for this cluster alone,** and ₹533.45 crore its whole cost, the sums would run like this (our arithmetic). By June 2029, about 37 months after deployment, the six-year schedule would have written off about ₹274 crore, leaving about ₹259 crore on the books for the three years after the term sheet ends. The ₹1,000 crore would be 1.9 times the cluster's cost, before power, space, staff, interest and the "allied services" the customer also gets.

## The mechanism

**The machine is the larger cost.** Running costs are small beside it: in the latest quarter services bought, staff and other expenses came to ₹52.3 crore, against ₹74.0 crore of depreciation (our arithmetic from \[RES\]). EBITDA leaves out the larger of the two. The annual report says the economics "depend on capacity utilisation, pricing, infrastructure and financing costs, and customer adoption" \[AR, p.93\].

**So profit after depreciation depends on how full the machines are, and at what price.** Once a machine is bought, its depreciation is the same whether it is rented all day or sits idle. So when more revenue comes from the same machines, profit after depreciation rises faster than revenue. The EBITDA margin rose for a different reason: running costs grew more slowly than revenue. In a year revenue rose 5.1 times, running costs 2.0 times and depreciation 1.7 times (our arithmetic from the table). The EBITDA margin went from 41.1% to 76.6%, and profit before tax from a loss of ₹18.5 crore to a profit of ₹97.1 crore. The company puts the latest quarter's growth down to "contribution from Blackwell GPUs, better utilisation and higher average realisation across the GPU fleet" \[PR\], with no figure for either.

**The company says it built ahead of demand.** Depreciation starts when a machine is put to use, sold or not. Machines bought in FY25 were all put to use during FY26 \[AR, note 3.1, p.151\], and in two of its quarters depreciation ran ahead of EBITDA, at 238% and then 120%. The managing director describes FY26 as building "ahead of demand instead of behind it" \[AR, p.18\]. Our reading: each new batch may repeat the cycle. In the latest quarter the fleet stayed at about 5,100 GPUs, the number given for 30 June \[AR, p.21\] and again in October \[IP, p.3\]. Our reading: much of the quarter's 42.7% rise in revenue came from having the B200s for all three months rather than one, the rest from the better utilisation and price the company mentions; the filings do not split it.

**The three levers.** Profit over a machine's life depends on:

- *Useful life.* E2E uses six years. Management writes that "each generation of hardware can typically sustain a 6-8 year useful life as it finds its own price-performance niche" \[AR, p.21\]. The life is reviewed "at each financial year end" \[AR, p.142\]. How much rides on it: in April to September depreciation of property, plant and equipment was ₹123.1 crore and profit before tax ₹155.8 crore \[RES, cash flow\]. On a four-year life profit before tax would have been about ₹94.2 crore; on eight years about ₹186.5 crore (our arithmetic, a rough illustration that treats all of it as six-year straight-line, though some asset classes have lives of 3 to 10 years; the four-year case is ours, not management's).
- *Fill and price.* Not in the filings we read, as above; the earnings call was not heard.
- *Funding.* Borrowings rose from ₹103.2 crore at 31 March to ₹450.4 crore at 30 September \[RES, balance sheet\]. The FY26 term loans were "secured by an exclusive charge on the fixed assets acquired against such loans and by pari passu charge over the Company's current assets" \[AR, note 14, p.159\]. Since 2024 the company has raised ₹1,591.94 crore of equity in three issues (our arithmetic) \[AR, p.80\]. For the ₹1,079.28 crore preferential issue and the ₹107.00 crore QIP, the monitoring agency CARE Ratings reports "Deviation from the objects: NIL" for the quarter, with the standard note that it "does not perform an audit" \[MA\]. Shareholders were asked to let the board raise up to ₹1,500 crore more and borrow up to ₹10,000 crore over its capital and reserves \[AGM, items 7 and 8\].

Running costs grow with the fleet too. E2E buys data-centre space, bandwidth and related services from Larsen & Toubro (L&T), which holds 18.45% of its shares and also buys services from E2E (₹2.24 crore in FY26) \[AR, p.175\]. It paid L&T ₹13.42 crore in FY26 and ₹7.55 crore in April to June 2026, and asked shareholders to raise the annual limit from ₹30 crore to ₹100 crore, citing "increased deployment of cloud/GPU infrastructure services" \[AGM, p.29-32\]. Rent on leased assets appears as depreciation and interest, also below EBITDA \[AR, note 4, p.151\].

## Where this breaks

**Management's range is 6 to 8 years, and one asset sale made a profit.** Six years is the low end of management's own range \[AR, p.21\], and the filings give no resale evidence for GPUs. In April to September the company sold fixed assets for ₹9.97 crore at a profit of ₹5.18 crore, which means their book value was ₹4.78 crore (our arithmetic) \[RES, cash flow\]. We do not know what they were, and one sale proves nothing about GPUs in general.

**If the term sheet is for this cluster alone.** On that assumption, and if the ₹1,000 crore is earned as stated, it is nearly twice the cluster's cost before June 2029 (our arithmetic). Our reading: the risk would then lie less in this cluster than in the machines bought next, and in what this one earns after 2029.

**Depreciation's share of EBITDA has fallen for four quarters running.** It went from 238% to 43%. In that year revenue for each rupee of depreciation rose from about 1.0 to about 3.0 across the whole fleet (our arithmetic). The filings do not split this by GPU generation, so they do not show whether older machines hold their price as newer ones arrive.

**Cash was larger than borrowings at 30 September.** Cash and bank balances were ₹734.4 crore, against borrowings of ₹450.4 crore and lease liabilities of ₹46.8 crore, and equity was ₹1,804.1 crore (our arithmetic from \[RES\]). ₹102.4 crore of the QIP money, raised mainly for capital spending, was still unspent \[IP, p.16\] \[AR, p.81\]. In April to September cash from operations was ₹396.5 crore; ₹393.6 crore went on fixed assets and ₹26.0 crore on advances for them \[RES, cash flow\]. Operating cash also includes a ₹129.8 crore rise in other current liabilities, which the filing does not break down; at March that line, ₹17.34 crore, was mainly unearned revenue (₹13.77 crore) and customer advances (₹1.68 crore) \[RES, cash flow\] \[AR, note 19, p.161\].

**36.0% of FY26 revenue came from outside India.** In FY26, ₹88.49 crore of revenue, 36.0% of the total, came from customers outside India, up from 33.7% in FY25 (our arithmetic for the shares) \[AR, note 20, p.162\]. The filings do not say why those customers buy, and these figures predate the term sheet with an Indian customer. Our reading: the share from Indian customers is one way to watch whether the "sovereign" pitch shows up in revenue.

**Our sample is one company in one fast-changing year,** with interim figures under limited review and no peer's filing read.

## What follows from the documents

Things a reader can track each quarter:

- **Depreciation as a share of EBITDA** (the table's last row), and whether revenue rises between deployments or only when new machines arrive.
- **The useful life** in the next annual report, and any gain or loss on old machines sold.
- **The term sheet:** whether a contract follows, what it covers, when billing begins, and what share of revenue one customer becomes.
- **Funding:** borrowings, finance costs and any use of the ₹1,500 crore sought, while B300 / Vera Rubin is at "planning underway" \[IP, p.11\].
- **Revenue from Indian customers** as a share of the total.

## Back to the two stories

E2E's latest EBITDA margin was 76.6%; its profit after tax was 32.5% of revenue. Most of the gap is the machines, booked as ₹74.0 crore of depreciation. A year earlier that charge was larger than EBITDA; now it is 43% of it, because revenue, and with it EBITDA, grew faster than depreciation. The six-year life behind it is at the low end of management's own range, and if the term sheet is for the B200 cluster alone, its stated value is close to twice that cluster's cost (our arithmetic). The filings do not yet show how full the machines are, at what price, or what happens to the gap when the next batch is bought. Those are the numbers to watch.

Related: [STL Signs a \~$1.2 Billion Hyperscaler Supply Deal. Here's What Is Actually Committed.](https://www.moatmarginresearch.com/stl-hyperscaler-1-2-billion-supply-agreement/), on reading what a headline contract value commits, and [Jai Balaji Owns 30% of East India's DI-Pipe Capacity. Last Year Its DI Plants Ran at 30%.](https://www.moatmarginresearch.com/jai-balaji-di-pipe-market-share-utilisation/), on how utilisation, not share, sets the return on fixed capacity.

## Sources

- **\[RES\]** [E2E Networks, unaudited results for the quarter and half year ended 30 September 2026, with limited review reports](https://nsearchives.nseindia.com/corporate/E2E%5F16102026134952%5FFinal%5FOutcome%5Fof%5FBoard%5FMeeting.pdf) (NSE)
- **\[PR\]** [E2E Networks, press release, 16 October 2026](https://nsearchives.nseindia.com/corporate/E2E%5F16102026150410%5FSignedPressRelease16102026.pdf) (NSE)
- **\[IP\]** [E2E Networks, investor presentation Q2 FY27, 16 October 2026](https://nsearchives.nseindia.com/corporate/E2E%5F16102026151325%5FSigned%5FE2E%5FNetworks%5FIP%5FQ2%5FFY27%5FFinal.pdf) (NSE)
- **\[MA\]** [CARE Ratings, monitoring agency reports for the quarter ended 30 September 2026](https://nsearchives.nseindia.com/corporate/E2E%5F16102026181632%5FSigned%5FMonitoring%5FAgency%5FReport%5F16102026.pdf) (NSE)
- **\[FY26\]** [E2E Networks, audited results for the quarter and year ended 31 March 2026](https://e2e-mainsite-ui.objectstore.e2enetworks.net/Financial%5FCompliance/Financial%5FInformation/Financial%5FResults/20252026/Audit%5Ffinancial%5FResults%5Ffor%5Fthe%5FFinancial%5Fyear%5Fended%5FMarch%5F31%5F03%5F2026.pdf) (company website)
- **\[AR\]** [E2E Networks, Annual Report 2025-26](https://e2e-mainsite-ui.objectstore.e2enetworks.net/Financial%5FCompliance/Financial%5FInformation/Annual%5FReports/FY%5F2025%5F2026%5F2025%5F2026.pdf) (company website)
- **\[TS\]** [E2E Networks, disclosure of a binding term sheet, 31 August 2026](https://nsearchives.nseindia.com/corporate/E2E%5F31082026171929%5FOrder%5FDisclosureReg30%5F.pdf) (NSE)
- **\[TSPR\]** [E2E Networks, press release on the term sheet, 31 August 2026](https://nsearchives.nseindia.com/corporate/E2E%5F31082026232711%5FPress%5FRelease%5F31082026.pdf) (NSE)
- **\[AGM\]** [E2E Networks, notice of the 17th Annual General Meeting](https://nsearchives.nseindia.com/corporate/E2E%5F05092026150615%5FNotice%5F05092026.pdf) (NSE)
- **\[REC\]** [E2E Networks, intimation of the Q2 FY27 earnings call recording](https://nsearchives.nseindia.com/corporate/E2E%5F16102026195347%5FSESignedRecordingweblink.pdf) (NSE)

---

> Educational research, not investment advice. Every figure above is drawn from E2E Networks' own filings and annual report and from CARE Ratings' monitoring agency reports, read from the primary documents on 17 October 2026; sums, differences, shares, ratios and the useful-life illustration marked "our arithmetic" are ours. The September 2026 quarter figures are unaudited and were subject to a limited review with an unmodified conclusion. The term sheet's value and terms are as the company states them; we read no contract. Nothing here says what the company is worth or what its useful-life estimate should be. No buy/sell recommendations, no price targets, no share prices. Moat & Margin is not a SEBI-registered Research Analyst or Investment Adviser.