E2E Networks Ltd Q4 FY2026 Concall Summary

Key Takeaways

• Profitability inflected, driven by utilization: A ₹12 crore sequential swing in profitability and 413 bps EBITDA margin expansion came from strong execution and high utilization, particularly in March, even as depreciation on GPU investments kept the full year in a PAT loss.
• Blackwell capacity is imminent but unsold: The first 1,024 B200 GPU cluster goes live mid-May, with a second cluster to follow, yet no deals have been closed on this capacity; management will announce a customer only after a deal is signed.
• Management is deliberately not guiding: Citing a rapidly changing future and a desire to avoid self-imposed limiters, the company declined to provide MRR, capex, or capacity targets beyond a minimum of ~6,000 GPUs by FY27 end.
• The strategy is shifting up the value stack: The company is focusing on customers who utilize higher-value tokens and generate higher ROI, moving away from price-sensitive, per-hour GPU cost comparisons, while investing in high-level talent to support this.
• What to watch next quarter: Whether the first B200 cluster goes live on time and secures a paying customer, the trajectory of MRR growth given the lumpy nature of the business, and any progress on asset-light financing models.

Blackwell Deployment & The Capacity Pipeline

• Imminent go-live: The first cluster of 1,024 B200 GPUs is expected to go live in mid-May, with a second cluster of 1,024 planned for deployment in a couple of months.
• No pre-sold capacity: Despite "a lot of interest" and active work with multiple customers, no deals have been closed on the Blackwell capacity being built. Management will announce a customer only after closing a deal.
• FY27 minimum capacity: Management committed to a minimum of around 6,000 GPUs by the end of FY27, explicitly stating they do not want to place a limiting number on the upside.
• Future chipset plans: The company is continuing to build its deployment plan around next-generation chipsets including B300, GB300, and Vera Rubin.
• Speculative build validated: Management characterized the company as one of the "haves" in a world divided between haves and have-nots, with access to speculative GPU capacity where capacity is running out everywhere.

Tarun Dua: "Where the world is divided between haves and have-nots, we are the haves in terms of having access to speculative GPU capacity where capacity is running out everywhere."

Financial Performance & Profitability Mechanics

• Profitability swing: A ₹12 crore sequential swing in profitability was driven by operating leverage and strong execution, with EBITDA margin expanding 413 bps sequentially.
• Depreciation burden: Depreciation increased to ₹51.3 crore in Q4, reflecting ongoing infrastructure investments. The full-year PAT loss of ₹15.6 crore was driven entirely by depreciation on GPU infrastructure.
• Core business health: The core business remains strongly cash positive and operationally profitable at the EBITDA level. As utilization ramps up, revenue will progressively outpace depreciation, improving reported profitability.
• Talent investment: Employee costs increased as the company invests in high-level talent to solve more interesting problems, positioning for opportunities 12-18 months or two years out.
• Revenue drivers: Revenue growth was driven by an increase in overall utilization, particularly in March, demonstrating the ability to scale AI infrastructure rapidly and convert capacity into revenue.

Demand & Pricing Environment

• Strengthening demand: The market has turned "somewhat much better" than a couple of quarters back for those with GPU infrastructure in place, with strong demand both in India and globally. The company prioritizes India first but is happy to support global infrastructure needs.
• Pricing stability with tailwinds: Management sees no negative pressure on GPU pricing; tailwinds are slowly inching prices up. The broad secular trend is that there are not enough GPUs in the world that people want to buy and deploy.
• Long-term asset utility: GPUs can be utilized over a 7-8 year period, and the shifting costs to new chipsets are significant, supporting pricing stability over extended periods. [12-13]
• Workload mix: The majority of revenue is still closer to training than to inference, though hard lines between the two are difficult to draw as the same GPUs can be used flexibly for both workloads.
• Higher-value token focus: The company is focusing on customers who utilize higher-value tokens, solving specific business problems that generate higher ROI, rather than price-sensitive customers focused on per-hour GPU cost. This is a broad industry trend. [18-19]

Tarun Dua: "Short-term of course we want to make more money and long-term of course we want to work with the kind of customers who will sustain and grow on our infrastructure over a long period of time."

Customer Contracts & Revenue Visibility

• Contract duration targets: For larger clusters, the company tries to secure at least 6 months to 1 year of visibility. For even larger customers, it looks for 2-3 years of visibility, aiming for a judicious mix of short-term, medium-term, and some long-term contracts.
• Lumpiness decreasing: Revenue lumpiness is inherent to the business but will decrease as the company builds more volume of GPUs under management. Current customers predominantly have sustainable long-term workloads, not short-term ones.
• No MRR split disclosed: Management declined to provide a specific split of MRR by contract duration, stating these numbers are subject to rapid change and do not add to understanding at the current scale.
• No inference/training split: Management will not provide a specific revenue mix between inference and training, saying these numbers would become clearer at a much larger scale.

Strategic Direction & Business Model

• Technology business framing: Management explicitly wants investors to view the company as a technology business, not an asset monetization business, and to move away from metrics like asset turns.
• Asset-light exploration: The company is exploring private credit asset-light models to bring rapid expansion to GPU capabilities. However, management said it is too early to comment on the margin impact of such structures.
• Partnerships: An exploratory-stage MOU with L&T aims to monetize GPU infrastructure L&T is building; this is not an exclusive arrangement. The company will continue to explore other partnerships of a similar or different nature.
• India AI Mission: The company continues to work closely with the AI Mission for coming generations of GPUs and is hopeful to continue collaboration, but is not trying to force anyone to offtake capacity and is happy to sell outside the scope of the Mission.
• Software stack capability: The TIR platform has demonstrated full-stack capability for bare-metal and container-based NVIDIA-powered GPU infrastructure at bigger scale, helping LLM teams operate trainings on large clusters. Major software improvements have been made to improve reliability, performance, and scalability.

Tarun Dua: "We want to get the framing away from things like asset turns. Essentially, don't look at us as an asset monetization business. Look at us as a technology business."

The Decision Not to Guide

• No MRR guidance: Management declined to give MRR guidance for one year, stating the future is changing very rapidly week on week and it is better to watch quarter-on-quarter than to predict four quarters in advance.
• No FY27 capex number: Management will not put a capex number for FY27, stating "a look back is more important than putting a number today" and that they continue to deploy capital under various business models.
• Self-imposed limiter: The refusal to guide is not due to macro worries but to avoid setting an underwhelming number that would act as a limiter on what the company is capable of doing over the next 52 weeks.
• Look-back preference: Across capacity, contract mix, and capex, management consistently prefers to report numbers on a look-back basis over time rather than providing forward targets at the current scale.

Tarun Dua: "We do not want to give an underwhelming number without first exploring like over week after week over next 52 weeks what we are capable of doing. So it will become sort of like a limiter for ourselves to say that, okay, we are only to meet this number."

Guidance & Commitments

Key Metrics