Digi Power X: a $1.1 billion contract, a $200 million loan not yet signed and a hold at $2.38
Hold at $2.38, watchlist for new money
Digi Power X trades at about my Base value of $2.42, so the market already pays for the Cerebras data center contract being built on time and earning a normal margin. The Bear value of $1.09 is 54% below the price and the Bull value of $5.37 is 126% above it. With so little reported about the project's cost and margin, I require a 55% margin of safety, which puts the buy range at $1.01 to $1.14. This is a speculative, binary stock, not a bargain.
- Upside to base value
- +1.7%
- Maximum buy price
- $1.01 (range $1.01 to $1.14 by method; a conditional re-entry trigger, not an unconditional order)
- Required margin of safety
- 55%
- Asset-based cross-check
- $2.25 (5.5% below the price)
- Price to book value
- 0.95x on about $2.50 a share (my estimate for September)
- Market value and enterprise value
- $243 million and about $108 million after $135 million of cash
- Return the price implies
- an EBITDA margin of about 62% on the Cerebras contract (my Base 65%)
- Cash and debt
- about $135 million of cash on October 13, 2026, no debt, a $200 million project loan committed but not signed
- Revenue and loss, last twelve months
- $31.4 million of revenue, $36.0 million net loss
- Dividend
- none
- Business quality
- 38/100
- Valuation score
- 39/100
- Confidence
- 7/25, grade D
Digi Power X spent the first half of 2026 changing from a small bitcoin and colocation business into a builder of AI data centers. In May it signed a ten-year contract with Cerebras Systems for 40 megawatts in Columbiana, Alabama, with an initial value of about $1.1 billion. The shares have since fallen from $3.97 in mid-August to $2.38, and 61% over twelve months. Today the contract is worth only what the company can build, finance and operate it for, and the filings do not say what the building costs.
At $2.38 on October 29, 2026, the shares trade 1.7% below my Base value of $2.42. The Bear value of $1.09 is 54.2% below the price and the Bull value of $5.37 is 125.6% above it. I assign no probabilities to the cases. The required margin of safety of 55% gives a buy range of $1.01 to $1.14 and a conservative limit of $1.01. For existing holders this is a hold, with a position size that allows for a total loss. For new money it is a watchlist name.
Valuation route: Pre-profit infrastructure developer with one contracted project, valued on a project cash flow model plus net cash and other assets. Economic classification: not yet assessable, because the main business has no operating history. Primary method: discounted after-tax cash flow of the Cerebras contract, less the cost still to be spent, plus cash and other assets, less capitalized corporate costs. Cross-check: 0.9 times estimated book value. Values are present fair-value estimates, not 12-month price targets.
What Digi Power X is
The company has three activities. The first is the legacy business of colocation, bitcoin mining and energy sales, which earned about $5.5 million of revenue in the second quarter and is being wound down. The second is NeoCloudz, a rental business for NVIDIA graphics processors, which produced $1.1 million of revenue in about five weeks of operation. The third is the AI data center campus in Columbiana, Alabama, which is the reason the shares exist at this price.
Under the Cerebras agreement the company must build, equip and commission two phases: 15 megawatts targeted for December 15, 2026 and another 25 megawatts by the end of the first quarter of 2027. Cerebras pays a monthly fee based on the power delivered, part of it prepaid, and makes one-time payments tied to construction. The initial term is ten years from the later phase start. Cerebras can extend by one, three, five or seven years, and the company puts the total value with one seven-year extension at about $2.5 billion. The agreement for the second phase is conditioned on the company securing adequate financing.
The company also has plans for a 75 megawatt campus in North Carolina, sites in New York, a non-binding letter of intent for a 1.3 gigawatt power asset in West Virginia and a Silicon Valley laboratory. None has a contract, so I give them no value in the Base.
The latest quarter
| Second quarter 2026 | Result | Comment |
|---|---|---|
| Revenue | $6.63 million | down 18% from $8.11 million |
| Net loss | $14.36 million | $0.17 a share; includes a $5.0 million warrant loss and a $2.8 million crypto loss |
| Adjusted EBITDA | $3.3 million | a positive figure that excludes the items above |
| Colocation and legacy mining | about $3.6 million | winding down |
| Energy sales | $1.9 million | legacy |
| GPU rental | $1.1 million | about five weeks of operation |
| Cash | $142.4 million | June 30; about $150 million on August 14 |
| Total assets and equity | $279.6 million and $265.0 million | June 30, 2026 |
Revenue over the last twelve months was $31.4 million with a net loss of $36.0 million. Nothing in these figures shows what the Alabama campus will earn, because it is not yet operating. The next report, expected on November 12, will show the third quarter, the first without most of the legacy business, and any update on the capital cost.
What the earnings are made of
- Reported earnings are noise. The net loss includes warrant fair value changes and the revaluation of bitcoin holdings, which move with the share price and the crypto market.
- The legacy business is shrinking. Revenue fell 18% in the quarter and the company says the wind-down is planned.
- The contract is the value. At $1.1 billion over ten years, the monthly fee works out at about $110 million a year, or $2.75 million per megawatt a year, before any one-time construction payments. How much of it is profit is not disclosed.
- Power and operating costs are not disclosed. The filings do not say whether Cerebras pays for the electricity, so the margin on the contract is my estimate.
I therefore model the contract rather than the reported results. This is the largest source of uncertainty in the analysis and it is the reason Confidence is D.
Capital, debt and shares
| Item | Value |
|---|---|
| Cash, June 30, 2026 | $142.4 million |
| Cash, October 13, 2026 | about $135 million |
| Capital spent in 2026 so far | about $153 million (October 13), against about $110 million at the end of June |
| Debt | none reported |
| Alabama project loan | $200 million committed on October 13, arranged by Goldman Sachs, expected to close within 90 days, not yet signed |
| Shares outstanding (October 26) | about 102.1 million |
| Warrants and options | not disclosed in the material I reviewed |
| Share sales under the at-the-market program in Q3 | none |
The loan sits in a project subsidiary. The company gives an unsecured guarantee until the project is complete. Its rate, term and repayment schedule are not disclosed, and the commitment depends on documentation and lender checks. Cash rose from $73 million at the end of March to $142 million at the end of June because the company sold shares, and it has since fallen to about $135 million as capital spending continued.
An amended shelf registration was filed on September 11 and a preliminary proxy statement on October 16. I did not read either, and either could allow more shares to be issued. For the per-share values I use 107 million shares in the Base, which is the 102.1 million outstanding plus about 5% for warrants, options and equity issued before the project is complete. This is an estimate.
Primary valuation: the Cerebras contract
| Base case | Value | How I get it |
|---|---|---|
| Contract revenue | $110 million a year | $1.1 billion over ten years |
| Cash flow after tax and upkeep | $58.6 million a year | 65% EBITDA margin, 15% tax, upkeep spending of 2% of revenue |
| Present value of ten years of cash flow plus residual | $404 million | 12% discount rate, residual of four times the final year's cash flow |
| Less remaining capital spending | -$233 million | $360 million total at $9 million a megawatt, less $120 million spent, my estimate |
| Plus cash | $135 million | October 13, 2026 |
| Plus legacy business and GPUs | $35 million | $15 million and $20 million, my estimates |
| Plus other assets | $10 million | stake in US Data Center and the Silicon Valley property |
| Less capitalized corporate costs | -$80 million | my estimate of overhead for the growth team |
| Equity value | $272 million | $2.54 a share on 107 million shares |
The $200 million loan is not deducted separately. It finances part of the capital spending, and its cost is inside the 12% discount rate. Counting it again would charge the project twice. The one-time construction payments from Cerebras would reduce the capital the company must fund, but their size is not disclosed, so I give them no credit. Their size is one of the biggest unknowns, and it is an upside to the Base.
A terminal share is not meaningful here, because the value rests on a ten-year contract and a residual that is 12% of the project value.
| EBITDA margin on the contract | Discount rate 10% | 12% | 15% |
|---|---|---|---|
| 55% | $2.27 | $1.93 | $1.52 |
| 65% | $2.94 | $2.54 | $2.04 |
| 75% | $3.61 | $3.14 | $2.57 |
| Total project cost | Margin 55% | 65% | 75% |
|---|---|---|---|
| $300 million ($7.5 million a megawatt) | $2.48 | $3.08 | $3.69 |
| $360 million ($9.0 million) | $1.93 | $2.54 | $3.14 |
| $420 million ($10.5 million) | $1.39 | $1.99 | $2.60 |
The value moves between $1.39 and $3.69 for reasonable changes in two inputs that the company does not disclose. The price of $2.38 sits in the middle of that range, which is why I call it fair and not cheap.
Secondary valuation: adjusted book value
Shareholders' equity was $265.0 million at June 30. I subtract about $10 million for the third-quarter loss and spending, my estimate,, which gives $255 million, or $2.50 a share on 102.1 million shares. I apply 0.9 times, which gives $2.25. The discount reflects the cash that the company is still spending, the risk that assets built for one tenant are worth less than their cost, and the dilution that has funded the balance sheet.
This method does not use the contract economics, but it is not fully independent of the first: both rely on the money spent being worth about what it cost. If the project is worth more than its cost, book value understates the equity, and if less, it overstates it. I give no extra Confidence for the agreement.
Historical cross-check
The shares fell 61% over twelve months, from about $6.03, according to a broker chart. The chart shows a spike toward $9 earlier in the year, followed by a decline to $2.38. The fall in mid-August followed the second-quarter report, which showed a wider loss, and the financing announcement on October 13 did not reverse it. I did not collect a multi-year history of the price to book value, so the historical component of the valuation score is neutral.
Valuation triangulation
- Primary (contract cash flow model): $2.54, weight 60%.
- Secondary (0.9 times estimated book value): $2.25, weight 40%.
- Divergence between the two in the Base: 12.1% of their midpoint, inside the 20% limit.
- Weighted Base: $2.42. The Bear and Bull values differ by much more than 30% between the methods ($0.82 against $1.50 in the Bear, $6.78 against $3.25 in the Bull), which flags the wide uncertainty of the cases.
I weight the contract model higher because it shows each assumption, and the asset-based check lower because it values cost, not future cash flow.
What is the market already pricing in?
| Reverse valuation | Result |
|---|---|
| EBITDA margin on the contract that equates the Base to $2.38 | 62% (Base 65%) |
| What that requires | both phases built for about $360 million, a 12% discount rate and a residual after ten years |
| Enterprise value the market pays | about $108 million, or $2.7 million for each of the 40 megawatts |
| Expected return if the Base comes true | about 12% a year, the discount rate, as the project is completed |
The enterprise value looks low next to a $1.1 billion contract, but it excludes about $233 million of capital that is still to be spent. A buyer at $2.38 is paying for completion and has no cushion if costs or margins are worse than my Base.
Bear, base and bull scenarios
| Case | Contract model | Book value check | Blended | Versus price |
|---|---|---|---|---|
| Bear | $0.82 | $1.50 | $1.09 | -54.2% |
| Base | $2.54 | $2.25 | $2.42 | +1.7% |
| Bull | $6.78 | $3.25 | $5.37 | +125.6% |
- Bear: the margin is 50%, the discount rate 15% and the project costs $420 million. Financing for the second phase fails, so only the 15 megawatts of the first phase are completed, with about $30 million recovered from the equipment already bought for the second. The company issues shares to fund the gap, and the share count rises to 125 million. The book value check uses 0.6 times.
- Base: as described above.
- Bull: the margin is 75%, the discount rate 10% and the project costs $320 million. The residual after ten years is worth eight times the final year's cash flow because Cerebras renews, and $150 million is added for further contracts at the other sites. The book value check uses 1.3 times.
The spread between the Bear and the Bull is five to one, and I think a large part of the outcome is binary: the loan closes and the buildings come on line, or they do not. The Bear is not a floor. If the project failed completely, the equity could be worth less.
How the buy price is set
Confidence is D, which calls for a margin of safety of 50% or more. I use 55%, because the capital cost and margin are not disclosed, there is one tenant and one site, and the outcome is binary. The margin is applied once, to each method.
| Method | Value | Buy price at a 55% margin |
|---|---|---|
| Contract model | $2.54 | $1.14 |
| Book value check | $2.25 | $1.01 |
| Weighted Base | $2.42 | $1.09 |
The conservative limit is $1.01 and the range is $1.01 to $1.14. The price is more than twice that. If the shares ever reach this range, I would first check why: it would probably mean that the loan failed or the contract changed.
Dividend and buybacks
The company pays no dividend and does not buy back shares. It has funded itself by selling shares, and while no shares were sold under the at-the-market program in the third quarter, the shelf registration remains available. Management says the project loan is meant to limit dilution.
Reasons to own DGXX
- A signed ten-year contract worth about $1.1 billion for 40 megawatts, with a possible $2.5 billion including one extension.
- About $135 million of cash and no debt on the balance sheet.
- A $200 million loan commitment from a major lender, arranged by Goldman Sachs, which would cover most of the remaining spending.
- Further sites in North Carolina, New York and West Virginia, which I do not value but which could add to the Bull.
What could go wrong?
- The loan does not close, or closes on harsh terms, and the second phase, which depends on financing, is delayed or lost.
- The campus costs more than my $360 million estimate, or is late, which triggers penalties and service credits under the contract.
- All of the value rests on one customer, Cerebras, and the termination provisions are not public.
- More shares are sold at lower prices to fund the gap, as in the first half of the year.
- The legacy business continues to lose money while it winds down.
Management execution
- Management signed a large contract and raised cash during a strong market, and says that long-lead equipment for both phases is on site or on its way, some ahead of schedule.
- Phase 1 is targeted for December 15, 2026. It is the first real test, because the shares will be judged on delivery.
- The company has changed focus several times, from bitcoin to colocation, GPUs and data centers, and has funded itself by issuing shares. I give neutral credit until a campus runs.
- Disclosure is thin on the numbers that matter: project cost, margin, the construction payments and the loan terms.
Stock Analyza scorecard
| Business quality | Score |
|---|---|
| Moat and pricing power | 5/20 |
| Economic return over cost of capital | 3/20 |
| Balance sheet and resilience | 10/15 |
| Earnings quality | 3/15 |
| Growth and runway | 11/15 |
| Management and capital allocation | 5/10 |
| Dilution and governance | 1/5 |
| Total | 38/100 |
The valuation score is 39 of 100.
| Valuation score | Score |
|---|---|
| Discount to the weighted Base (+1.7%) | 14/40 |
| Protection against the Bear (-54.2%) | 3/20 |
| Agreement between methods (12.1% in the Base, wide in the Bear and Bull) | 10/15 |
| Historical valuation (no history collected, neutral) | 5/10 |
| Asymmetry of market expectations | 6/10 |
| Data and model penalty | 1/5 |
| Total | 39/100 |
Business quality and valuation are shown separately and are not averaged. The valuation score is low because there is no margin of safety against a Bear case that is 54% below the price, and because several critical inputs are my estimates.
Confidence score
| Confidence | Score |
|---|---|
| Data and sources (project cost, margin, loan terms, warrants missing) | 2/5 |
| Business predictability (one project, no operating history) | 1/5 |
| Valuation robustness (large sensitivity to cost and margin) | 1/5 |
| Accounting transparency (warrants and crypto revaluation) | 2/5 |
| Scenario dispersion ($1.09 to $5.37) | 1/5 |
| Total | 7/25, grade D |
The grade measures how well the model is identified, not the chance of a loss. A grade of D means that this analysis should be used to frame the questions, not to size a position with confidence.
What I would watch from here
Green
- The $200 million loan is signed, with a rate that leaves the project's cash flow intact, before about January 11, 2027.
- Phase 1 of 15 megawatts is delivered close to December 15, 2026, and the company reports the first monthly fees.
- The company discloses the total project cost and the construction payments from Cerebras.
Yellow
- The third-quarter report on November 12 shows cash below the $135 million reported in October.
- The proxy statement or shelf registration asks for a large increase in the number of authorized shares.
- Phase 1 slips by more than a month.
Red
- The loan commitment lapses or is cut, or the second phase is delayed for lack of financing.
- A new equity offering at a price well below $2.38.
- Cerebras raises a dispute, claims a penalty or uses a termination right.
The three most important thesis breakers
- The project loan fails to close, which would leave the company to pay for the second phase with shares.
- The campus costs about $420 million or more, which cuts the Base value to about $2.00 at a 65% margin and to $1.39 at 55%.
- Cerebras's contract turns out to carry weak payment protection, so that the $1.1 billion is not a reliable cash flow.
Stress test and independent check
I recomputed the values with a separate script and solved the reverse valuation for the margin. That checks the arithmetic, not the forecasts. Several of the most important inputs are my estimates, as the table shows.
| Input | Status |
|---|---|
| Total project cost ($360 million) | Estimate at $9 million a megawatt; not disclosed |
| Capital already spent on Alabama ($120 million) | Estimate from $153 million of total spending in 2026 |
| EBITDA margin (65%) | Estimate; power cost treatment is not disclosed |
| Contract revenue ($110 million a year) | Calculated from the $1.1 billion total over ten years |
| Construction payments from Cerebras | Not disclosed; no credit given |
| Share count (107 million) | 102.1 million outstanding plus an estimated 5% |
| Cash ($135 million) | Company statement of October 13 |
The weakest point of the model is the combination of project cost and margin, which management has not disclosed and which together decide whether the contract creates or destroys value. The strongest bull argument is that AI computing demand makes a ten-year contract with the tenant a scarce asset, that the construction payments lower the capital needed, and that the other sites are worth far more than I credit. The strongest bear argument is that a company that has sold shares at every stage has not yet shown that it can finance and build on time, and that the market has been marking the shares down for that reason.
Final verdict: hold at $2.38, watchlist for new money
Digi Power X is an option on one contract. At $2.38 the market pays about what I think the contract is worth after the capital still to be spent, so there is no margin of safety, and the range of outcomes is very wide.
Verdict: HOLD at $2.38, watchlist for new money. Weighted Base $2.42, 1.7% above the price, cross-check $2.25, Confidence D (7/25). Required margin of safety 55%, buy range $1.01 to $1.14, conservative buy limit $1.01.
For an existing holder, the original cost of the shares is irrelevant. I would size the position so that a total loss is acceptable, and review it on November 12, when the third quarter is reported, and again when the loan is signed or lapses. New money waits for a price at or below $1.14, or for the company to disclose the project cost, the margin and the loan terms.
Sources
Primary sources are Digi Power X's Form 8-K on the Cerebras agreement, the second quarter 2026 filing and the October 13, 2026 financing announcement.
The reference market price is $2.38, the October 29, 2026 regular session close from a public price history, checked against a broker quote of $2.34 on October 30. The share count of about 102.1 million is from a public statistics page dated October 26, 2026. Balance sheet figures are as of June 30, 2026 unless stated.
Disclaimer. This analysis is provided solely for research and educational purposes. It is not personalized investment, financial, legal or tax advice. Values are estimates, not forecasts or trading instructions. Digi Power X is exposed to project financing and construction risk, dependence on one customer, dilution, power and operating costs, crypto and GPU market risk and valuation risk. Investors can lose part or all of their invested capital.
Method: Master Investment Analysis Framework v4.2. Approach: project cash flow model of the Cerebras contract plus net cash and other assets, cross-checked by 0.9x estimated book value. Confidence: D/7. Data checks: passed with warnings. Open data gaps: total project cost, contract margin and power cost treatment, construction payments, loan terms, warrant and option counts, third quarter 2026 results. Conditional research only.