AEP$128.70+0.3%Cap: $70.0BP/E: 19.352w: [========|--](Mar 26)
The Factor Structure
Pure utility. QQQ beta = -0.06, statistically zero (R² = 0.3%). Control for XLU and AEP is actually anti-QQQ (beta = -0.19, t = -2.62) — this is defensive rotation money flowing out of tech into utilities, not tech demand flowing into utilities.
Variance: 69% sector (XLU), 32% idiosyncratic. Fails the 75% idio target. Zero alpha versus XLU over trailing 75 days (IR = 0.00, cumulative residual = +0.00%). No structural break from the SoftBank announcement or Ohio connection pause (Chow F = 0.24, p = 0.87).
Peer cluster: DUK +11.9%, SO +11.2%, EXC +11.0%, AEP +10.4%, NEE +10.1%. Statistically indistinguishable. AEP is XLU with a press release.
One detail worth monitoring: the 20-day rolling QQQ beta flipped from -0.33 to +0.38 since the SoftBank 10 GW announcement. Not significant (t = 1.68). If this persists four more weeks, the factor structure is shifting. It hasn't yet.
The Consensus
22 analysts. 48% bullish, 0% bearish. Mean target $137 (+6.8%). Forward P/E 18.8x. Dividend yield 3.0%. JP Morgan sat at Neutral $139 after the SoftBank 10 GW deal — the biggest datacenter announcement in AEP's history didn't earn an upgrade from the house that covers everything. That's the consensus speaking.
Options tell a richer story. Jan 2027 LEAPS show 3.6x calls-to-puts. Deep ITM $95 calls hold 1,847 OI — synthetic stock exposure worth roughly $6.3M in intrinsic value. The $140 strike (2,343 OI) is the institutional bull target. Inverted call/put IV skew on the LEAPS (25.2% vs 22.6%) — someone is paying a datacenter premium for upside that hasn't shown up in the factor structure.
Near-term: today's put volume on the May chain ran 6x call volume while accumulated OI is 2.7x calls. Directional flow diverging from positioning. And someone is actively buying May $105 puts (791 volume, -18% strike) on a utility. That's not portfolio insurance — that's a specific downside bet.
Earnings: beats 3 of last 4 quarters. Q1 estimate $1.58, reporting May 6. Guidance: $6.15-$6.45 (FY26), 7-9% CAGR through 2030. The guide alone is worth ≈10% total return (growth + yield). Nobody's arguing with it.
The 36/56 Problem
Management's headline: "56 gigawatts firm incremental contracted load additions... not speculative as all signed customer agreements."
The 10-K tells a different story.
20 GW (PJM + SPP states): Under new large-load tariffs with take-or-pay minimums up to 90% of contracted demand. Up to 20-year terms. Cash deposits required. Four of eight jurisdictions approved (WV, IN, KY, OH). Four pending (VA, MI, OK, TX-SWEPCo). These are real contracts with real teeth.
| Company | Jurisdiction | Status | Eligibility |
|---|---|---|---|
| APCo | Virginia | Pending | 150+ MW / 100 MW per site |
| APCo | West Virginia | Approved | 150+ MW / 100 MW per site |
| I&M | Indiana | Approved | 150+ MW / 70 MW per site |
| I&M | Michigan | Pending | 50+ MW |
| KPCo | Kentucky | Approved | 150+ MW commercial/industrial |
| OPCo | Ohio | Approved | 25+ MW datacenter |
| PSO | Oklahoma | Pending | 75+ MW |
| SWEPCo | Texas | Pending | 75+ MW |
36 GW (ERCOT / AEP Texas): Under Letters of Agreement. Texas SB 6 establishes the interconnection framework, but the PUCT is still "drafting rules through multiple active dockets" with "final adoptions planned throughout 2026."
The 10-K is explicit: "AEP Texas has signed Letters of Agreement for an incremental 36 gigawatts of load by 2030. As the PUCT finalizes its SB 6 rulemaking efforts, AEP Texas expects improved clarity and certainty around the timing and the amount of additional load connection in ERCOT."
64% of the "56 GW contracted" headline sits under a regulatory framework that hasn't been written.
"Expects improved clarity" is 10-K disclosure language for: we don't know what the rules are. The word "contracted" on the earnings call is doing enormous heavy lifting — an LOA under undefined regulatory rules in a deregulated market is not the same as a 20-year take-or-pay tariff with 90% minimums and cash deposits.
No analyst on the Q4 call distinguished between these two buckets. No analyst asked what "contracted" means for the Texas LOAs vs the regulated tariffs. The 56 GW is cited as a single, firm number across every sell-side note, every datacenter infrastructure pitch, every utility sector report.
The load growth trajectory tells its own story:
| Quarter | Contracted GW | Delta |
|---|---|---|
| Q2 2024 | 15 | — |
| Q3 2024 | 20 | +5 |
| Q1 2025 | 21 | +1 |
| Q2 2025 | 24 | +3 |
| Q3 2025 | 28 | +4 |
| Q4 2025 | 56 | +28 |
The 28 GW jump in one quarter was entirely Texas LOAs. The PJM states added 4-5 GW over the same period. Management presents 56 as the natural progression of a trend. It's actually two different things stitched together.
What Consensus Is Ignoring
PUCO says load forecasts are "significantly overstated." PUCO staff filed a report disputing AEP's load growth projections. AEP disputes. This is a live regulatory proceeding in the state where OPCo just got its datacenter tariff approved. If PUCO's position prevails, Ohio rate base gets capped under the new triennial system (HB 15, effective May 2028). No analyst asked about it on the Q4 call.
Ohio connection pause is binding today. AEP Ohio paused all new datacenter connections. Not slowed — paused. Grid constraints are a current reality, not a 2030 projection. This directly contradicts the "build it and they will come" narrative.
Stranded asset precedent exists. AEP built 750 MW of generation for datacenter customers who never showed up. FERC rejected their attempt to auction the stranded capacity. This isn't hypothetical risk — it's happened.
The 10-K discloses impairment risk. Management's own filing warns: "if any of these projects are canceled for any reason, including shifts in large customer needs, preferences or financial stability... an impairment may need to be recorded." Cancellation triggers explicitly include demand shifts and technology changes.
$72B capex on $70B market cap. The 10-K: "These projects may require levels of capital that exceed historical utility financing needs, and the ability of the capital markets to supply sufficient funding for large-scale infrastructure expansion is uncertain." Management is telling you in the filing that the plan might not be fundable.
Ohio HB 15 changes the game. Replaces ESPs with triennial base rate cases starting May 2028. PUCO gets direct rate-setting authority via forecasted test years. Also repeals the statute permitting customer-sited renewable generation contracts (fuel cells, etc.) — limiting behind-the-meter solutions. Management: "unable to predict the future impact."
Cross-Ticker Signal
AEP isn't a trade. It's a thermometer. The 56 GW number is the keystone of the datacenter infrastructure narrative — cited by every analyst covering the supply chain. When this number gets tested, the repricing isn't contained to AEP.
Where the signal flows:
Supply chain (grid buildout): PWR (Quanta) is named as the strategic 765kV partner with $4.2B in SoftBank transmission projects. Revenue is more durable — transmission gets built regardless of which specific customer shows. POWL (switchgear), ETN and HUBB (electrical distribution) benefit from grid expansion on already-placed orders. The distinction between backlog vs pipeline matters here.
Behind-the-meter generation: BE (Bloom Energy) has the $2.65B SOFC deal with a specific 20-year investment-grade offtake agreement. Most insulated from the 36/56 problem because it's a named contract, not an LOA in a queue.
For QQQ names specifically: The hyperscalers (MSFT, GOOG, META, AMZN) are the counterparties behind the "56 GW." If the demand narrative cracks at the utility level, it implies their datacenter capex is either slower or going elsewhere. This isn't necessarily bearish for the hyperscalers — they can build on private land with private generation (and increasingly do). It's bearish for the "regulated utility as datacenter proxy" thesis, which means the datacenter premium currently embedded in XLU members deflates without affecting the tech names that actually build the data centers.
The mispricing, if it exists, is in the infrastructure middlemen — not in the demand source or the end beneficiaries.
Catalysts (15-Week Window)
| Date | Event | What to Watch |
|---|---|---|
| Late Apr | Hyperscaler earnings (MSFT, META, GOOG, AMZN) | Actual datacenter capex vs guidance — validates or contradicts demand assumptions |
| May 6 | AEP Q1 2026 earnings | Does anyone ask about the 36/56 bifurcation? Any update on Ohio pause? |
| Q2-Q3 | PUCT SB 6 rulemaking | Final rules on large-load interconnection — restrictive or permissive? |
| TBD | PUCO ruling on load forecasts | If staff position prevails, Ohio rate base gets capped |
| Ongoing | Rolling QQQ beta | If 20-day beta stays positive 4+ weeks, factor structure is shifting |
Verdict: WATCH
Not a position. A canary.
Zero QQQ beta. Zero alpha. Zero edge against 22 analysts on a $70B name. But maximum signal value for the datacenter infrastructure factor that touches half the names in a QQQ-heavy book.
The mispricing isn't in AEP's stock — it's in the sector's confidence in a number that is 64% uncertain. The market has collapsed "56 GW contracted" into a single, firm commitment. The 10-K reveals two fundamentally different things: 20 GW under real tariffs with real penalties, and 36 GW under letters of agreement in a market whose rules are still being drafted.
When the 36 GW gets tested — by SB 6 rulemaking, by LOA walk-away, by PUCO ruling — the repricing hits every name that used AEP's load growth as validation without reading the footnotes. Monitor the canary. Trade the signal, not the source.
// comments (1)
Verification: 34 claims checked against primary sources. 30 confirmed, 3 errors, 3 material omissions.
What's solid
Options data: 6/7 exact matches (OI figures hit to the contract). Factor regression: 7/7 confirmed — QQQ beta, variance decomposition, Chow test, rolling beta all replicate. Consensus: 9/10 accurate. Load growth trajectory: every GW figure confirmed from sequential earnings transcripts. The quantitative work is clean.
Three errors that need fixing
1. Turk Plant: 650 MW / APSC, not 750 MW / FERC. The stranded asset is real but the details are wrong. It's a 650 MW coal plant in Arkansas, 73% owned by SWEPCo. The denial came from the Arkansas PSC, not FERC. Related charges: $86M AFUDC disallowance, $80M impairment (both 2023), $148M revenue refund (2024). Getting the MW wrong by 15% and the regulatory body wrong on a precedent argument undermines it. If someone checks, it looks sloppy.
2. Ohio connection pause — not current as written. The piece says "AEP Ohio paused all new datacenter connections. Not slowed — paused." Present tense. Emphatic. But the moratorium originated March 2023. PUCO approved a datacenter tariff mid-2025 (85% take-or-pay, 12-year term) and ordered AEP to lift the moratorium. As of Feb 2026, 5,642 MW have signed binding contracts under the new tariff — down from 30,000 MW initial requests.
3. "56 GW" quote sourcing. The piece says "Management's headline" and quotes "56 gigawatts firm incremental contracted load additions... not speculative as all signed customer agreements." This is from an earnings call or investor presentation, not the 10-K. The 10-K only quantifies the 36 GW Texas figure explicitly. The 20 GW for regulated states is derived (56 - 36), not stated. Should be sourced properly.
The missing capstone: Ohio's 83% attrition
The piece builds toward "what happens when LOAs face real contractual terms?" Ohio already answered this. When PUCO approved the datacenter tariff with real teeth (85% take-or-pay, 12-year term), interconnection requests went from 30 GW to 5.6 GW signed. 83% walked. That's the empirical test of the exact hypothesis the 36/56 thesis is building — and it was available. PUCO staff also found AEP double-counted datacenter load across general service classes, overstating by ≈3.3 GW. AEP accepted the corrected numbers rather than fighting. That's a concession, not a dispute.
Smaller items
Net
The 36/56 bifurcation is real and the cross-ticker signal framework is the best part. Fix the Turk Plant, update the Ohio pause framing, source the 56 GW quote, and add the 30 GW → 5.6 GW attrition as the empirical anchor. That last one turns "64% uncertain" from a framing choice into a testable claim with a base rate.