Nscale (NSCL) filed its S-1 on September 18 to list on the NYSE. I read the S-1, but I didn’t want to spend hours building the initial model framework myself. So I turned to Scout, Daloopa’s AI assistant for building and updating financial models in Excel. Nscale aims to become a fully vertically integrated AI provider. Its total contract value has grown from $100 million to $103 billion in two and a half years,mostly driven by Microsoft and Anthropic. If Nscale does not bring capacity online on time or meet uptime service levels, certain customers can terminate the related capacity.
The Model Build with Scout
I opened a blank workbook and entered the NSCL ticker into the Scout prompt to “build a model.” Scout pulled the statements, the revenue splits, the adjusted EBITDA bridge, the debt schedule and the KPIs. Values land in cells by fixed mapping rules, so each figure is placed the same way every time. Every fundamental data point links back to its source page in the filing.
What the Numbers Say
1. Gross margin is negative before depreciation
Cost of revenue excludes D&A. It still ran $189.6 million against $140.6 million of first-half revenue. That is a gross margin of -35% before depreciation. Including $174.0 million of D&A, the margin is -159%.
The old colocation business ran a 32.7% margin in FY2024. It turned negative in FY2025 as the GPU cloud scaled. Capacity is coming online ahead of billing. The model has to show when that flips.
Table 1. Income statement ($ millions)
| FY2024 | FY2025 | H1 2025 | H1 2026 | |
|---|---|---|---|---|
| Revenue | 19.1 | 33.0 | 10.4 | 140.6 |
| Cost of revenue (excl. D&A) | (12.8) | (45.6) | (7.8) | (189.6) |
| Gross profit before D&A¹ | 6.2 | (12.6) | 2.6 | (49.0) |
| Gross margin before D&A¹ | 32.7% | (38.2%) | 25.0% | (34.9%) |
| Depreciation and amortization | (5.1) | (40.2) | (3.9) | (174.0) |
| Operating loss | (16.2) | (169.8) | (24.7) | (492.0) |
| Loss on fair value adjustments | (40.2) | (527.8) | (348.9) | (457.1) |
| Net loss | (78.2) | (761.8) | (368.9) | (1,020.1) |
| Adjusted EBITDA margin | (48%) | (248%) | (147%) | (142%) |
¹ Computed: revenue less cost of revenue, which is reported excluding D&A. Linked values are Daloopa source cells. H1 figures are six-month periods tagged to Q2 in Daloopa. FY2024 values converted from thousands.
2. Nscale will mostly own its footprint
At scale, Nscale will run seven wholly owned sites, nine colocation sites, and one leased site. Today, most of the footprint is colocation and leased sites. By megawatts, owned sites will carry 1,005 MW of the 1,370 MW active and contracted, colocation will carry 165 MW, and the single leased site will carry 200 MW. Owning its sites gives Nscale direct control over power. It also explains why capex runs ahead of revenue. Nscale requires an upfront prepayment of around 23% of contract value at signing, which offsets part of that capex.
Table 2. Footprint and contract book (as of August 31, 2026)
| Sites | MW | |
|---|---|---|
| Wholly owned | 7 | 1,005 |
| Colocation | 9 | 165 |
| Leased | 1 | 200 |
| Total active + contracted | 17 | 1,370 |
| Owned share of MW² | n/a | 73.4% |
3. Two customers make up most of the book
Microsoft has statements of work worth up to $43.8 billion through 2033. Anthropic signed agreements worth up to $44.6 billion on August 25. Together, that is roughly 86% of total contract value. The S-1 says it plainly: Nscale expects Microsoft and Anthropic to be significant customers.
Table 3. Customers and fleet
| Value | |
|---|---|
| Microsoft statements of work, up to ($B) | 43.8 |
| Anthropic Services Agreements, up to ($B) | 44.6 |
| Total contract value, active + contracted ($B) | 103.0 |
| Microsoft + Anthropic share of contract value² | 85.5% |
| Active GPUs (thousands) | 25.0 |
| Active + contracted GPUs (thousands) | 461.0 |
| Active share of contracted fleet² | 5.4% |
² Computed. Contract value share uses the S-1 total of $103.4 billion. Active share of fleet equals active GPUs divided by active and contracted GPUs.
4. Take-or-pay, subject to delivery terms
Nscale sells on long-term take-or-pay contracts. Both the Microsoft and Anthropic contracts include termination rights that apply before take-or-pay obligations begin.
Anthropic ($44.6 billion).
- Payments are conditional. The $44.6 billion is “up to” and subject to delivery and service availability requirements. Each tranche’s term starts on Anthropic’s acceptance.
- Anthropic can exit late tranches. Miss a delivery window and Anthropic can terminate that tranche without liability.
- Supply-delay relief is limited. Delays are excused only for industry-wide GPU shortages and only if Nscale placed binding orders on time. Past a threshold, Anthropic can still terminate.
- Uptime is a trigger. Service levels are more stringent than in Nscale’s other contracts. Sustained low uptime lets Anthropic exit.
- Financing is not yet committed. As of the filing, Nscale had no binding commitments for the financing needed to perform.
Microsoft ($43.8 billion).
- Late delivery costs money from day one. Nscale credits Microsoft 100% of a daily bill for every day late, netted against the first invoices.
- Sixty days late is an exit. If a GPU service will be more than 60 days late with no equivalent capacity, Microsoft can terminate it or reset the date. Miss the reset date and the service is deemed terminated. Nscale refunds the matching share of the upfront payment.
- The NVIDIA carve-out is narrow. Credits are waived for NVIDIA delays only if Nscale placed binding orders within 60 days of the SOW and gave timely notice.
- Convenience is the open question. Microsoft’s purchase order terms, attached to the SOW, say Microsoft may terminate “with or without cause“, paying only for services delivered and taking back prepaid unused fees. The S-1 summary describes only termination for cause. How those two documents interact is my first diligence question.
So most of the book becomes take-or-pay only after Nscale finances, builds and delivers on a fixed schedule. For the model, that means breaking out the backlog by customer, tranche and delivery date.
5. Customers are funding the build
First-half operating cash flow was $1,686.1 million against $3,230.1 million of capex. Customer prepayments drove that operating cash flow: deferred revenue contributed $3,811.5 million. Strip out prepayments, and operating cash flow was negative $2,125.4 million. Debt on the June 30 balance sheet was $137.4 million. Nscale has added debt since then, which the model will reflect once September-quarter results are reported.
Prepayments are cheap capital. They also hand leverage to the customer. The Anthropic build is expected to rely on external financing and cash flows from operations, which puts the capital markets back in the picture.
Table 4. Who is funding the build ($ millions)
| FY2025 | H1 2026 | |
|---|---|---|
| Net cash from operating activities | 1,376.4 | 1,686.1 |
| of which: change in deferred revenue | 2,036.2 | 3,811.5 |
| Operating cash flow excluding prepayments³ | (659.8) | (2,125.4) |
| Purchases of property and equipment | (621.4) | (3,230.1) |
| Deferred revenue, total (period end)³ | 2,038.7 | 6,492.6 |
| Debt, total (period end)³ | 78.9 | 137.4 |
³ Computed. Operating cash flow excluding prepayments equals net cash from operating activities less the change in deferred revenue. Totals sum the linked current and noncurrent balances.
One more thing: $457.1 million of the first-half net loss came from non-cash fair-value marks on pre-IPO convertibles, warrants and SAFEs. Look through it. The operating line matters.
Bull/Bear Cases
- Bull. Nscale delivers to Microsoft and Anthropic on schedule, take-or-pay protection kicks in and margins flip as the fleet fills. Only 5.4% of contracted GPUs are live today. The other 94.6% is still to be delivered.
- Bear. A delivery slip lets Anthropic or Microsoft walk from a tranche, financing costs rise and negative gross margins persist longer than the prepayments last
Final Thought
Scout handled the data pull and model setup, so I could focus on the questions that price an IPO: the gross margin trajectory, the delivery schedule, and the fine print in the two largest customer contracts.
Data sourced from Daloopa. This is research commentary, not investment advice.
