Setup

Digi Power X (DGXX) is a $735M fully-diluted ex-Bitcoin miner pivoting to AI/HPC datacenter colocation. The stock is up 349% YTD on a single catalyst: a May 4, 2026 announcement of a "10-year, $1.1 billion" colocation agreement with Cerebras Systems. The Q1 2026 10-Q filed May 15 disclosed the contract structure. The headline number is doing a lot of work — what's actually firm is roughly a quarter of it.

What the filing says

10-Q Note 20 (Subsequent Events) and Exhibit 10.2 (confidential treatment requested):

• 40 MW total at Columbiana, Alabama
• Phase 1: 15 MW by December 15, 2026 — firm
• Phase 2: 25 MW, "conditioned on the Corporation securing adequate financing for Phase 2 operations"
• 10-year initial term, ≈$1.1B aggregate
• With one 7-year extension, ≈$2.5B
• Cerebras holds exclusive license to the facility for the duration
• Monthly colocation fee + prepaid component + one-time construction payments

Phase 2 represents ≈$700M of the $1.1B TCV. DGXX must source the construction capital before Cerebras's commitment activates.

Other Q1 2026 facts:

• AI/HPC segment revenue: $0 (Note 17 — "in the development phase")
• Total revenue: $6.79M vs $9.28M Q1 2025 (-27% YoY); colocation -40% YoY
• Cash: $57.8M at quarter-end → $125M post-quarter after $102.9M ATM tranche
• Fully diluted shares: 97.5M as of May 15 (up from 77.1M at start of year)
• ATM upsized to $175M
• Going concern language unchanged from 10-K
• Second AI customer: SubQ AI, 24-month $19.6M GPU rental

What the market thinks

Idio vol 123.8%. RSI 79. ATM IV 170% (92nd percentile). Put/call open interest: 0.00 — zero puts outstanding. Max pain $2.00. Sole analyst (HC Wainwright, recipient of $840K + 269K warrants in 2025 settlement) targets $12.

Fair value is sensitive to the bull-state revenue multiple. AI infra peers (IREN, CORZ) trade at 8-15x forward revenue with operating revenue. DGXX has $0 AI revenue today and execution risk. A ramp/uncertainty discount is warranted.

At a defensible 5-6x multiple for a pre-revenue ramp story, weighted FV is $5.73-$6.66 — current $7.54 is 12-30% above intrinsic. At a generous 8-10x (full peer parity assumed), FV is roughly current price. The market is implicitly underwriting the higher multiple.

This is a softer claim than a "39% gap." The gap is real but depends on whether you grant DGXX peer multiples it hasn't earned. We assume not.

Why the gap exists

1. Cerebras has not confirmed the deal in any of their own disclosures. Cerebras IPO'd May 14, 2026 — 10 days after the DGXX MSA was signed. The 424B4 prospectus lists datacenters as "California, Oklahoma, and Canada" — Alabama omitted. The S-1/A amendment filed May 11 (7 days post-deal) has the same omission. CEO Andrew Feldman's IPO-day CNBC interview named OpenAI and AWS as headline partnerships — no DGXX. The Cerebras blog posted May 6 and May 8 with no Alabama post. The cerebras.ai/cbrs IPO landing page features OpenAI, AlphaSense, and GSK as customer spotlights — no DGXX.

The bull's counter is the IPO quiet period: maybe Cerebras's legal team couldn't process a 10-day-old material agreement before lock-up. Two facts cut against this. First, the S-1/A amendment was itself filed 7 days post-deal — Cerebras did update their registration statement in the window, and chose to omit Alabama from it. Material developments require disclosure in registration amendments, not just at IPO. Second, the Cerebras blog continued posting through the period — May 6 and May 8 posts on other topics demonstrate operational capacity to publish.

Acknowledge the partial confound: Cerebras's smaller WYFI 5 MW Quebec deal got buyer-side confirmation 13 months after signing, not 10 days. The strongest comparison is Cerebras's own S-1/A behavior — and there they affirmatively omitted the deal. The next decisive read is the Cerebras Q2 2026 10-Q (~August 14). At that point the IPO quiet period explanation expires; material customer disclosure obligations apply.

2. The conditional contract structure aligns with the silence. If Cerebras were firmly committed to all 40 MW, omitting Alabama from a public prospectus would be unusual. The cleanest reading consistent with the contract language: Phase 1 (15 MW, ≈$300M TCV) is binding; Phase 2 is an option Cerebras can wait to see DGXX exercise. Cerebras stays silent on the headline because they don't want to feature a partner whose 25 MW depends on capital DGXX doesn't currently have.

3. Prior instance on this ticker. In 2025, DGXX disclosed a "partnership" with Super Micro (SMCI) as part of the ARMS 200 build. SMCI never acknowledged the relationship in any 10-K, 10-Q, or earnings call across six transcripts. It turned out to be a standard purchase order ($8.5M equipment deposit) dressed up as strategic. This is N=1 not a pattern, but it's the same management team applying the same headline-vs-substance language.

Risks (ranked by impact)

1. Phase 1 actually delivers Dec 15, 2026 and Cerebras pays. If this happens, the silence becomes irrelevant — revenue prints, multiple expansion follows. The $45M YTD capex deployed at Columbiana shows real construction, not slide decks. Our scenario weights this state at 20%; we may be too bearish given the committed capital.

2. Cerebras confirms in their Q2 2026 10-Q (~August 14). Material customer relationships must be disclosed in periodic filings. Confirmation kills the silence signal entirely. The quiet-period defense expires here.

3. Squeeze risk. +349% YTD, 170% call IV, zero put OI, retail tape. Any positive 8-K (Phase 1 milestone, ATM at premium, Cerebras confirmation) could squeeze hard. Right thesis ≠ profitable trade if entered into momentum.

4. Phase 2 funded via non-dilutive debt. CEO mentioned "active discussions" on debt. A credit facility weakens both the dilution thesis and Phase 2 conditionality simultaneously.

5. Stale insider conviction signal. Feb 2025 coordinated insider buying ($2.5M, CEO + 3 directors). At the time, shares outstanding were 33M; today 90.4M. Dilution-adjusted, insiders bought at the equivalent of ≈$2.75 — stock is now $7.54, they're up roughly 2.7x. They have not added at current prices despite the Cerebras catalyst. The absence of fresh insider buying alongside a $1.1B headline deal is itself a signal — but the Feb 2025 conviction is real and 15 months stale, not nothing.

Catalysts

The decisive pair: Cerebras Q2 10-Q (Aug 14) and Phase 1 deadline (Dec 15). 210 days to a decision.

What would change our mind

• Cerebras explicitly names DGXX in their Q2 2026 10-Q → silence signal invalidated → raise P(bull) materially.
• DGXX 8-K announcing a non-dilutive debt facility specifically for Phase 2 capex → Phase 2 conditionality weakens.
• Phase 1 ready-for-service 8-K on or before Dec 15, 2026 → execution and contract reality both validated.
• Renewed coordinated insider buying at current prices ($7+) by C-suite and ≥2 directors → high-credibility signal that Phase 2 commitment is more durable than primary-source disclosure suggests.

What would NOT change our mind: stock price appreciation alone, analyst upgrades from conflicted firms, DGXX-side reaffirmation. The signal is about the counterparty's behavior, not the issuer's.

Evidence

Memo LR: 0.85. Bearish but partially priced. The signal is real (counterparty silence with prior instance, Phase 2 conditionality, dilution velocity), but three legitimate weaknesses temper magnitude: (1) IPO quiet-period partial confound on the silence claim — resolved Aug 14; (2) bull-state multiple uncertainty creates a sensitivity band, not a sharp gap; (3) stale-but-real insider conviction signal we're trading against. The case is sound, the magnitude is bounded.