Record margins on AI supercycle, but peak sustainability unproven

Delivered ₹156.8 Cr. Growth rates match stated figures; Q4 implied ~₹95.6 Cr consistent with 64% QoQ expansion. YoY base unavailable in delivered-result context.

Delivered OPM 75.2% = EBITDA margin per transcript. Implies Q4 ~60.7%, consistent with 1,450 bps expansion claim.

No prior margin guidance to validate. 75% is exceptional for cloud infra; cost pressures (memory) acknowledged; July price hike required; sustainability not evidenced.

1,024 B200s live; 'maximal utilization' claimed without % data. Capacity fungible across training/inference/agentic makes true utilization opaque.

Management: 'very moderate' pricing impact. July price hike post-quarter confirms prior-period pricing neutral to modest.

Q4 FY26 implied 60.7% (75.2% - 1,450 bps). 1,024 B200s live at max utilization drove +1,450 bps. Exceptional expansion but likely peak-cycle.

Q4 FY26 ~40%; Q1 FY27 20-21%. Sharp drop unexplained by management. International grew to 37% offsetting. Concentration risk easing but IAIM growth stalled.

₹450 Cr loan drawn for B200 batch 1. Interest cost spiked Q1. More debt incoming with batch 2. Peak debt not disclosed.

July 2026 GPU/CPU price hike announced post-quarter, driven by memory cost inflation, not demand surge. Tests Q2 customer elasticity.

Expansion in both utilization and capacity additions majorly drove growth; pricing had very moderate impact

Yes, quite maximal utilization of capacity compared to any previous quarter

Yes, sustainable over medium and potentially long term due to increased utilization, robust demand, operating leverage, judicious mix of contracted/on-demand/spot/AI services

Company too small at 5,000 GPUs; these metrics not relevant today; focus on growing first

Hard to pin down fungibility; both run in parallel on same clusters; major volumes still training; no specific %

Day zero of AI, decadal theme, don't worry about news cycles. Broad org AI adoption early everywhere. Long runway ahead.

Honor all contracts; customers happy to renegotiate post-contract based on market; most understand market prices; value+support more important than price alone

Primarily cost increase (memory impact higher on CPU than GPU). Offered customers longer-term contracts (1-2yr) to lock price; customers agreed; enabled us to upgrade other capacity pricing.

Minimum 6-year life cycle for all GPU generations. Open-source model efficiency gains improve older GPU performance. Each generation finds own price/performance sweet spot. No active compression.

Quite sustainable; each generation finding sweet spot. Customers increasingly want 1-3 year contracts for price certainty. Building predictability and revenue stickiness.

₹450 Cr loan outstanding, will increase with next B200 batch. Peak debt: will not quantify.

Very early days; will announce when built and executed

Capex to hardware GPUs. All announced GPU funding backstopped by debt/accruals/prior equity. Future equity: will announce if and when it happens.

Arm's length. E2E buyer of L&T DC capacity; L&T buyer of E2E cloud. Joint go-to-market on platform possible.

16+ years in business; seen everything; cycles come and go; confident we'll grow every cycle

Management refuses MRR guidance as policy; 'look at past quarters' instead. No forward revenue range given.

No quantified margin target. Contradicted by acknowledged cost pressures (memory), July price hike required, hyperscaler competition accelerating. At peak likely unsustainable.

No capex plan quantified. Future funding opaque. Management refused to quantify peak debt or equity raise timeline.

75% EBITDA margin at peak-cycle; cloud infrastructure commodity. Hyperscalers scaling aggressively internal capacity. Customers harder on pricing. GPU lifecycles shortening (B300, Vera Rubin imminent). Memory cost inflation already forcing July price hike. Margin compression inevitable as supply normalizes.

India AI revenue fell sharply from ~40% (Q4 FY26) to 20-21% (Q1 FY27) with no explanation. Customer-type breakdown refused by management ('too small right now'). Concentration in unnamed large customers undisclosed. Loss of single large customer could cause severe revenue cliff.

B200 batch 2 expected 'next couple of months' with no firm date. B300 and Vera Rubin still in standing stage (no orders placed yet). Capex timing slippage, hardware cost inflation, or supply-chain delays could derail revenue ramp and margin profile.

₹450 Cr debt outstanding, rising with B200 batch 2. CFO refused to quantify peak debt ('will not be quantifying'). Future capex for B300/Vera Rubin unquantified. Equity raise plans vague ('if and when'). Funding trajectory opaque; risk of cash crunch if capex exceeds plan.

SovCloud launched to hold large-scale CPU clusters and enable funding arrangements; Delaware entity launched for US market expansion. Both at 'very early days' stage. SovCloud funding 'will announce when ready.' International revenue only 37%. Execution risk real on platform/geography diversification thesis.