Sify Infinit’s 560 page DRHP is a masterclass on the Data Center (DC) market in India
Apart from covering the nuances of its own business, Sify has also included a primer on India’s 5GW ambition for FY30 and the challenges that lie ahead
After China, India has the lowest cost per MW of DC capacity - we plan to almost X3 our DC capacity in the next 5 years.
Sharing a few takeaways ⤵️
(1) Energy is 44% of total expenses
- This data point confirms what several industry reports have pointed out: Energy is the single largest OPEX component of a DC
- This % has been on a downward trend for Sify from ~47% to ~38% - driven by shifting to more Renewables & operational efficiency
PS: PUE (Power Usage Effectiveness) is the KPI for measuring energy efficiency; Sify doesn’t provide this figure in its DRHP
- However, Sify does shared that 27% improvement in PUE in FY24; driven by 20% reduction in energy leakages.
- Btw, Sify passes these costs on to its customer (this is part of the Colocation DC operating model - more on this below).
Imp nuance (can skip too): Industry avg. PUE in India is close to 1.5 - versus high end DCs in USA which are at 1.1 i.e. they are able to reduce energy wastage by ~40% (hence, can operate at Energy as ~30% of COGS).
(2) India is really DC capacity starved!
Don’t even bother zooming into the capacity utilization chart - the mature DCs operate at 98% to 100% capacity (!!)
- For context: Global average vacancy for mature DCs is ~6%; Sify is close to 2% for mature sites
- It takes anywhere from 3 to 24 months post operationalization for a DC to start seeing high capacity utilization
- On a overall basis, utilization is up from 84.7% to 86.2% DESPITE growing operational capacity from 77MW to 113 MW (i.e. ~50%) in past 2.5 years
It was one thing to read about this from an industry POV (Kotak MF and JM Financial) but very interesting to see it in a company’s filings 🙂
(3) Who are Sify’s top customers?
- While Sify has ~500 customers, there are probably 10-odd customers who contribute > 50% of the revenue.
- Hyperscalers are ~67% of revenue & Enterprise clients are ~33% of revenue; this has remained consistent in the past 36 months
- Interestingly, the consistency remains despite Sify doubling from 92MW capacity in FY23 to 188MW capacity in H2 FY26
- This data is VERY instructive: High customer concentration BUT balanced with stickiness over time & ACV growth from existing customers.
TL;DR while Sify has doubled capacity, its customers have also at least doubled their utilization (and therefore spending) at Sify 🤯
From a cashflow perspective, customers pay in arrears each month or quarter and typically commit to a 3 to 5 year lock-in (!!)
Btw, this is inherent in the business - Sify runs colocation DCs - in the Enterprise DC business; operators rent 100% of capacity from a single DC to a client
(4) Guess why Sify does NOT spend on software & systems as part of CAPEX?
- There are 4 types of DCs - Colocation, Cloud, Enterprise & Edge (for more; check below)
- The simple answer is: Sify operates co-location DCs i.e. it rents rack space primarily to hyperscalers (global cloud providers)
- Sify only provides the site, shell, electrical wiring & power inputs to allow its hyperscaler customers to plug & play their own servers & software
- Therefore, you can see that the CAPEX from Sify is primarily Building (i.e. land parcel & site) and Plant & Equipment (”MEP” in industry parlance)
💡Some industry context: CAPEX per colocation 1 MW is ~₹46.5 crore v/s ₹138 crore for a cloud 1 MW - the entire cost delta comes down to servers & software (source: JM Financial)
Therefore, if you think CAPEX incurred by Sify looks small - it is because nearly X2 Sify’s CAPEX is incurred by its own customers to set up the servers etc.
PS: I had explained the types of DCs in my prev. thread on the topic here: x.com/Rahul_J_Mathur…
(5) How much does Sify charge?
Colocation DC pricing is always expressed in kW per month
- Sify charges anywhere from ~₹6,800 to ₹11,900 per kW per month; this is within the industry avg of ~₹7,700 per kW per month which JM Financial estimated in March ‘25
- The typical multi-year contract has a ~2% to 4% p.a. escalation clause (because its own cost base i.e. rent also escalates on a similar ~3% basis)
PS: Sify’s top 3 Enterprise customers spend ~ ₹140 crore per annum and utilize ~15 MW of power per month - this is equivalent to running ~10,000 ACs for 8 hours per day - eyewatering figure but nothing in comparison to Stargate and other projects in USA.
I would highly recommend reading the DRHP - especially the macro POV sections which explain the DC business in detail.
I’ve also written a short 101 on India’s DC industry - linked in Point (4) above; do check that out too if you found this interesting / if you had Qs.
You can also pair reading this DHRP with my prev. thread or with with JM Financial’s DC 101 report (March 2025) and Kotak MF’s DC report (Sept 2025) - both available on the Internet for free 😊
Factories were the mainstay of the industrial revolution - DCs are the mainstay of the digital revolution. Seeing how hyperscalers have committed to 3GW in India this year already - I think we’ll blow past the 5 GW target (colocation & cloud) easily!
Discl: Shared for informational purposes only. DYOR. This post is not an endorsement or paid promotion. DYOR. This is part of my own research to understand the DC space in India for investment purposes (will also record 2 Breakdown videos later this year on the topic)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
India will see its 1st pureplay data center IPO soon - Sify Infinit Spaces (SISL) has filed for a ₹3700 crore IPO last week
SISL was started in 2017 as a subsidiary of NASDAQ listed Sify Technologies - it operates 14 colocation data center sites in India.
For FY25 - it made ~₹1,400 crore revenue with ₹125 crore PAT; the EBITDA margin is 44%
Btw, SISL’s parent Sify Tech was affiliated to the infamous Satyam Computers at the time of founding until 2005 (full stake sale) - at the time of 2009 scam - it was no longer affiliated.
I’ve been studying the Data Center opportunity in India for a couple of weeks - sharing a few fun facts to begin your learning journey below ⤵️
(1) India consumes 20% of global data - but has only 3% of data center (DC) capacity
Our data consumption will only continue to grow:
(a) We are the 2nd largest consumption market for several AI co’s like OpenAI, Perplexity etc
(b) 5G roll-out & adoption is still WIP; has gone from 10% to 35% penetration last year - will go beyond 50% in the next 18 months
(c) Our per capita data consumption is ~25GB per month, expected to grow to ~40GB per month over the next 5 yrs
(2) Regulatory tailwinds in this sector are VERY strong
- The central Govt has the Data Center Policy 2020 which gives a single window clearance
- Each state (as shown below) has further subsidies & incentives to attract hyperscalers
- The dual catalyst is Data Localization norms & DPDP (2023)
Meesho clocked ₹30,000 crore of Sales in FY25 - this is approx. 30 lakh orders per day fulfilled by 5 lakh sellers!
They aim to raise ~₹6,500 crore in their upcoming IPO; and it is the first horizontal e-commerce platform to go public in India (Flipkart started 8 years prior btw!)
Meesho’s updated 667 page DRHP is a must read for anyone who follows E-Commerce in India
Here are my 5 key takeaways from the filings⤵️
(1) Meesho has 21 crore active shoppers who place avg. 9+ orders per year 🤯
Few observations v/s their 2024 annual report:
(a) LTM order frequency is up from 9 to 9.5
(b) ATUs (active shoppers) is also up from 18.7 crore to 21.3 crore
(c) Order growth (50% YOY) is outstripping NMV growth (36% YOY) - this is interesting, Meesho is already a value commerce platform - the AOV is declining even further (by design) - to grow the customer base!
AOV has reduced from ₹336 (FY 23) to ~₹270 (Q1 FY26) - worth tracking this for the future! Valmo (check pt 3) is critical to making this AOV decline unit economic sustainable.
PS - I had covered Meesho’s 2024 Annual Report a few months back (bookmark for later):
(2) Scale economies are somewhat questionable….. (my only bear case on this business)
(a) The cost per order is down to ~₹37 in Q1 FY26 from ~₹50 in FY23 👍 Awesome, right?
(b) Well, not really - because AOV has also come down!
(c) As a % of AOV - cost per order is ~13.7% v/s 15% in FY23 - not much has changed…
This actually flows into the Contribution Margin - it remains at ~4.5% in Q1 FY26 v/s the peak ~5.6% in FY24.
Now, I’d also highlight that the bull case here is - AOV decline will plateau and as Meesho moves from 60% Valmo orchestration to ~80%+ orchestration (plus some further tweaks) - this should work just fine (if it doesn’t, don’t ask me).
This year, ICONIQ led Anthropic’s $5bn Series F, ElevenLabs $180M Series C, Legora’s $80M Series B and TinyFish’s $47M Series A..
They are a media shy Wealth Mgt firm with $90bn+ of AUM who put their money where their pen is (quite literally)
Earlier this yr, they published their “State of AI 2025” report - now they’ve topped it up with several updates in a 73 page analysis of their portfolio companies.
Sharing a few takeaways ⤵️
(1) “AI” has changed what “best in class” means for Software
(a) OPEX leverage: ARR per FTE has increased 65% from ~$200K Pre-AI (2020) to $331K now; touching $390K for co’s with $500M+ ARR
(b) The goalpost “Rule of 40%” is now the “Rule of 60%” - BUT, there is ONE nuance: FCF margins are taking a hit - but growth rates are MUCH higher (even for “AI adjacent” SaaS co’s)
(c) NDRs (net dollar retention) also takes a hit - gold standard was ~135% in 2020; it is now down to ~115%. Elevated churn (experimental ARR, hype etc), competition & pricing pressure is hurting.
The change doesn’t impact Seed investors (like me) - but is VERY relevant for Growth & Public market investors.
Btw, ICONIQ’s Enterprise Five - ARR, NDR, Rule of 60%, Net Magic Number & ARR per FTE provides an excellent scorecard for Growth investors!
(2) The mere presence of “AI” revives growth (not joking)
(a) The one pint comment is that every VC has an “AI heavy” portfolio because all SaaS co’s have pivoted into AI co’s
(b) There is some data to support why this might help - it reinvigorates innovation (EPD teams), outbound (Sales morale) & inbound demand (FOMO)
(c) I’ve seen at least one of our portfolio co’s come back from extinction to growth mode because of AI - even SMBs can now be targeted profitably (CS cost is down, Sales can be automated & outcomes can be sold)
Example from ICONIQ: Palantir went from 15% YoY growth in Q2 ‘23 to ~35% growth in Q4 ‘24 - driven by “AI”
Doug Leone has been a Partner at Sequoia for 32 years now - he doesn’t do a lot of interviews - but his track record is incredible - ServiceNow ($187bn), Nubank ($73bn), Medallia (acq. $6.4bn) and he continues to serve as a board member to many companies.
Doug did a rare interview on the Training Data podcast where he answered some key Qs about investing in AI today; sharing a few notes from the same:
(1) Where will value accrue in the AI stack? Applications!
“Values always accrues up at the application layer - near the customer, money & the business user.”
(2) Can you sit out of AI as an investor? Nope
”You are at the front end of a cycle, which doesn’t mean you have to invest in everything but you have to invest.”
(3) What to do about round prices in AI today?
“You see a small company with very good momentum in a front end of the market; you lean in and you hold your nose on price.”
(Arguably, Sequoia did this with Google back in the day; a $100M post money valuation for a Series A)
(4) Did Sequoia foresee the AI wave? Nope!
“In March 2022, I had a slide that talked about all the waves - the next box was a question mark. We (Sequoia partnership) did not see the wave coming; this wave has been a tsunami and I don’t think there’s there’s any end in sight.”
PS: That March ‘22 investor interaction was Doug’s last one as an active partner at the firm (he has since stopped doing new investments & has more of an advisory role)
(5) Is AI any different from the previous waves? Maybe
"The AI wave is bigger than the Industrial Revolution. I had never imagined that there would be something bigger than connectivity (Internet) and the mobility (Mobile) era."
(6) Does company building change in a world of AI? Nothing has changed.
“The business fundamentals are very similar - you need a terrific founder, world class engineers, salespeople that are not administrator etc”
GFF is an annual circus show for Financial Services professionals, founders & entertainers. Hence, I chose to wear my customary Papa Pump outfit (IYKYK, comment below).
I spent most of my two days in meetings (managed to get 17 done!) and a bit of time at the exhibition booths; sharing a few notes from these meetings:
(1) AI first collections agencies are emerging 🗣️
- Lazy VCs wrote this space off when Credgenics raised a $50M Series B from WestBridge in August 2023
- Collections is evergreen - stress emerges in each pocket at different times (e.g. unsecured PL, MFI etc)
- Due to competition now, Collections has moved from Software + Service to Outcome i.e. pricing based on ROR uplift delivered; this is a great pricing model for AI services
- The prev. gen of Collections companies were either offline agencies or human call centers or a hybrid of both
- The modern Collections agencies are building AI enabled Credit Engines i.e. a full funnel process to manage cases from pre-due to NPA; Voice AI to call || agents to email, WhatsApp etc || route optimization to direct field force etc
(2) Lots of activity in MFD roll-up & tech enablement 🏦
- Everyone’s uncle wants to build the new age Prudent (NSE:PRUDENT)
- The thesis is “priced in” - the next gen of MFDs are digitally savvy and/or are based in non-metro cities; they need training & then tenure to build their book of biz.
- Several VCs have started to take positions (e.g. $6M Series A in AssetPlus by Eight Roads, $3.5M Seed in ZFunds by Elevation)
- There are several companies still stepping out of woodwork; primarily ex AMC execs or seasoned BFSI professionals who have spent 3+ yrs building their book while bootstrapping & are now starting to raise $$
At a private event, the CEO of a leading Indian bank highlighted that over the next 3 yrs, they will spend ~10% of their IT budget on AI.
For context: The IT budget (headcount, Tech, outsourcing etc) at this size of bank would be ~₹2,500 to ₹3,000 crore per year
Single bank spend on AI is therefore ~₹250 crore per year (there are approx. 25 odd banks of this size or larger).
Now, there are 2 ways to interpret this ~₹6,000 crore ($750M) per year AI spending ⤵️
This depends on whether you are an optimistic or critic:
(1) Optimistic view: In 3 yrs, the actual spend would be X2 or X3 higher i.e. > ₹15,000 crore because some of the AI deployments will eat into the labor budget (check the example in this Thread below)
Indian banks are lighthouse customers (for SEA banks & rest of Indian Enterprise); therefore it would catalyze X5 further spending on AI - HUGE (~₹50,000 crore p.a.) opportunity within India itself - great for startups!
(2) Critical view: ~₹6,000 crore in AI spending is not sufficient to catalyze a new set of FinTech Infra vendors (centered around AI); maybe all of the incremental spend will go to incumbent service & Tech vendors
But, for the critics, I have 2 interesting examples from BFSI:
Navana AI x Bajaj Finance
Mr Sanjiv Bajaj (MD @ Bajaj Finserv) highlighted ~6 months ago that ~₹150 crore of loan disbursal per month is powered by Navana AI (indic voice AI)
The most recent public figure is that Navana AI is doing ~₹1,000 crore of loan disbursal per month for Bajaj Finance - rapid scaleup!
For those who track Bajaj Finance closely, you would know about their “FinAI” company ambition & the goal of saving ₹150 crore in OPEX for FY26 using GenAI; I had covered this in a 20 minute Breakdown video earlier this year: