Last week, we broke the story of ESDS Software Solutions committing $1.25 billion to lease GPU compute from Nasdaq-listed Sharon AI. The deal raised eyebrows. A company with ₹357 crore in annual revenue signing a contract worth ₹10,000+ crore over five years? That demands a closer look.
Today, we're doing what nobody else has done yet. We're opening up the calculator.
Because when you actually run the numbers on GPU economics, this deal starts looking very different from the headline.
Quick Recap
ESDS signed a 5-year agreement with Sharon AI (NASDAQ: SHAZ) to lease ~8,200 NVIDIA B300 GPUs in an Australian data center. Total contract value: $1.25 billion, roughly $250 million per year. ESDS is the customer. Sharon AI deploys and operates the infrastructure. ESDS pays monthly in advance and has put up $140 million in security.
(Read the full breakdown here)
Now let's get into the GPU economics.
What Is ESDS Actually Paying Per GPU?
Let's break the contract down to unit level.
Annual cost: $250 million GPUs: 8,200 Hours in a year: 8,760 (24 hours × 365 days)
ESDS's cost per GPU per hour = $250,000,000 ÷ 8,200 ÷ 8,760 = ~$3.48
ESDS is paying $3.48 per GPU per hour. Keep this number in mind.
What Does the Market Charge for NVIDIA B300 GPUs?
The NVIDIA B300 is the latest Blackwell Ultra generation GPU with 288GB HBM3e memory and 10 TB/s memory bandwidth. It's the most powerful commercially available GPU for AI training and inference right now.
Here's what cloud providers are charging as of April 2026:
| Provider / Tier | Rate per GPU per Hour |
|---|---|
| On-demand (CoreWeave, Lambda, Crusoe Energy) | $4.50 to $5.80 |
| Premium / high-availability listings | Up to $18.00 |
| Reserved (12-month contracts) | $3.40 to $5.65 |
| Spheron (bare-metal) | $3.50 |
| NeevCloud (pre-book) | $3.99 |
| Spot pricing (off-peak) | $2.90 to $3.50 |
Sources: getdeploying.com, tech-insider.org, Spheron, NeevCloud pricing pages. Data as of late March 2026.
ESDS is paying $3.48/hour. That's at the very bottom of the market. The on-demand rate that most enterprises pay is $4.50 to $6.00 per hour. The spread between ESDS's procurement cost and the market rate is where the opportunity lies.
The Revenue Opportunity: Two Scenarios
Given the sheer scale of 8,200 GPUs, which is far beyond what ESDS would need for its own internal operations, the most likely play is that ESDS will offer this compute capacity to its enterprise, government, and BFSI client base.
Let's run two scenarios at two different resale rates.
Scenario 1: 100% Utilization
With all 8,200 GPUs running 24/7 for a full year, that's 71.83 million GPU hours. At a conservative resale rate of $5/hour, ESDS generates $359 million in annual revenue against a $250 million cost, netting $109 million in gross profit per year ($545 million over 5 years).
Scenario 2: 80% Utilization (Conservative)
At 80% utilization, billable hours drop to 57.47 million. At the same $5/hour, the annual gross profit narrows to $37 million ($187 million over 5 years). Thin, but still positive.
Even in the most conservative scenario, ESDS is in the green. And remember, $5/hour is well below the on-demand market rate of $4.50 to $5.80. There's room to price higher.
In INR terms (at ₹93/$), the 5-year gross profit ranges from roughly ₹1,736 crore at 80% utilization to ₹5,073 crore at full capacity. For a company currently valued at ₹3,750 crore, even the conservative number is significant.
The key variable here is utilization. The higher ESDS can push utilization, the more profitable this deal becomes. Below 70%, margins start getting uncomfortable. Above 85%, this deal becomes extremely lucrative.
The Commitment Tells You Something About ESDS's Confidence
Let's put the commitment in perspective:
- Annual payment to Sharon AI (~₹2,325 crore) is 6.5x ESDS's FY2025 revenue of ₹357 crore
- Security deposit (~₹1,302 crore) is 3.6x the annual revenue
- 5-year total commitment (~₹11,625 crore) is over 3x the company's current market cap of ₹3,750 crore
These are not numbers a company throws around casually. The security deposit alone, ₹1,302 crore, is nearly four times what ESDS earns in an entire year.
This level of commitment strongly suggests that ESDS has significant visibility on demand. Whether through existing client conversations, pre-commitments, or pipeline deals, a company doesn't put up 3.6x its annual revenue as a bank guarantee on hope alone. Not after 21 years in business. Not when you're about to go public.
ESDS has 600+ enterprise clients, 200+ government institutions, and 400+ BFSI clients. These are exactly the kind of organizations that are beginning to invest in AI capabilities but may not want to deal with foreign cloud providers directly. The demand base exists. The question is how quickly ESDS can convert it into GPU compute contracts.
The Contract Structure Has Flexibility
A few structural points worth noting:
Monthly payments, not lump sum. ESDS pays monthly in advance, meaning cash outflows can be matched against incoming client revenue as the customer base builds. It's not writing a $1.25 billion cheque on Day 1.
36-month lock-in, not 60. ESDS can't terminate for convenience in the first 3 years. After that, there's more flexibility. If the market shifts unfavorably, ESDS isn't necessarily stuck for the full 5 years.
2-year extension option. If the deal works, ESDS can extend for 2 additional years. This gives it up to 7 years of GPU compute access without having to renegotiate from scratch.
Final Take
Here's what the numbers tell us.
At $3.48/hour, ESDS has locked in NVIDIA B300 GPU compute at a rate that sits at the very bottom of the current market. The on-demand market charges $4.50 to $6.00+ for the same hardware. The spread is real.
Even at conservative assumptions of 80% utilization and $5/hour resale, the 5-year gross profit potential is roughly ₹1,736 crore. At more optimistic but still reasonable assumptions, it's multiples of that.
But the most telling number isn't the profit potential. It's the commitment itself. A company doing ₹357 crore in revenue doesn't put up ₹1,302 crore as a security deposit and commit to ₹2,325 crore in annual payments without having a concrete plan and likely significant demand already in the pipeline.
ESDS has filed its DRHP to raise ₹600 crore via IPO. It has 1,200+ clients across enterprise, government, and BFSI segments. And it has just secured access to 8,200 of the most powerful GPUs available today at wholesale pricing.
Whether this bet pays off will depend entirely on execution: how fast ESDS can sign clients, how high it can push utilization, and how well it manages the operational complexity. But the math is on its side. And the confidence behind the commitment is hard to ignore.
Disclaimer: This article is for informational purposes only and should not be considered as investment advice. Investments in unlisted/pre-IPO shares carry risks including limited liquidity. Please consult a SEBI-registered investment advisor before making any investment decisions.