A regulated-utility levered call on the Georgia data-center build-out — the cleanest large-cap way to own AI power demand, but priced as if the affordability politics and equity dilution won't bite; own the growth, respect the ~24x multiple.
| Date |
|---|
| Type |
|---|
| What happened |
|---|
| Source |
|---|
| 2026-08-10 | editorial note | Valuation figure revised: $93, → $91.33Valuation moved from $93, (deep-dive-2026-06-21.md) to $91.33 (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Capex figure revised: $12,737M → $6,639MCapex moved from $12,737M (deep-dive-2026-06-21.md) to $6,639M (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Revenue figure revised: $12,631M → $6,977MRevenue moved from $12,631M (deep-dive-2026-06-21.md) to $6,977M (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Verdict changed: A regulated-utility levered call on the Georgia data-center build-out — the cleanest large-cap way to own AI power demand, but priced as if the affordability politics and equity dilu…Before (deep-dive-2026-06-21.md): A regulated-utility levered call on the Georgia data-center build-out — the cleanest large-cap way to own AI power demand, but priced as if the affordability politics and equity dilution won't bite; own the growth, respect the ~24x multiple. After (deep-dive-2026-08-10.md): The two mechanical bear points got repaired — a $26.5B DOE loan at Treasury+37.5bp and an equity need cut to $1.1B — but the quarter that "beat" grew operating income 0.7% and only ~1.2 GW of 16-17 GW contracted load is actually flowing; the thesis is intact and almost entirely forward-dated to 2028-2032. | dossier |
The verdict
The two mechanical bear points got repaired — a $26.5B DOE loan at Treasury+37.5bp and an equity need cut to $1.1B — but the quarter that "beat" grew operating income 0.7% and only ~1.2 GW of 16-17 GW contracted load is actually flowing; the thesis is intact and almost entirely forward-dated to 2028-2032.
Primary sources
SEC filings
Earnings calls
Source documents — open to read in full
Not a delta, but a gap the new shelf closed: the $26.5B DOE loan guarantee (closed 2026-02-20 — $22.4B to Georgia Power, $4.1B to Alabama Power, FFB at Treasury + 0.375%, final maturity 2055-12-10, drawable through 2033-09-15; $1.0B drawn March 2026 at 5.041%) predates the prior dossier's date but was absent from it. It is the largest loan in DOE history and is projected to save customers ~$7B over the term.
The structural thesis stands, and one leg of the bear case broke. The legal monopoly, the Georgia concentration, the constructive-regulation dependency and the rate-base compounding mechanic are all unchanged. Demand is not the question it was: weather-normal retail sales rose 2.3% in H1 2026, the highest through-June growth in nearly two decades, with weather-adjusted commercial KWH +10.9% at Georgia Power in Q2 and data-center usage +55% YoY.
What broke is bear point #2 from the prior dive — "the financing model dilutes the thesis." The DOE facility replaces ~$26.5B of what would have been market debt with 30-year government-guaranteed paper at Treasury+37.5bp, and the residual equity need is now $1.1B rather than an open-ended ATM. That is a genuine improvement in the cost of the build, and it lands on customer bills too, which partly disarms bear point #1.
What got worse is earnings quality. Strip the below-the-line and the quarter is ordinary:
| Q2 | 2026 | 2025 | Δ |
|---|---|---|---|
| Operating income | $1,776M | $1,764M | +0.7% |
| Interest expense, net of capitalized | $(796)M | $(874)M | −8.9% |
| AFUDC — equity funds | $128M | $80M | +60% |
| Earnings from equity-method investments | $86M | $10M | +760% |
| Income taxes | $187M | $289M | −35% |
| Net income to Southern | $1,174M | $880M | +33.4% |
. The H1 effective tax rate was 14.1% vs 21.1%. Meanwhile cash interest paid rose 14.4% (H1 $1,472M vs $1,287M, with $100M capitalized vs $61M) while the income-statement interest line fell. None of this is improper — it is exactly how a regulated utility mid-build reports — but the growth investors are paying 22x for was, this quarter, produced by tax rate, AFUDC and capitalised interest, not by selling more power at a better spread.
And the load is still almost entirely forward-dated. System data-center load is >1.2 GW today [transcripts/2026-q2] against 16-17 GW contracted. Roughly 7% of the contracted book is actually flowing. The OpenAI phases run 2028-2032. This is not a criticism of the contracts — it is the honest shape of the position.
Two additions to the moat inventory since the prior dive:
The bargaining-power read is better than the prior dive's. The OpenAI structure — customer pays full infrastructure cost, posts security, and supplies 1,000 MW of demand response back to the grid — is a utility dictating terms, not accommodating them. Womack's own framing on the call was that Southern is "not going to be next" on new large-scale nuclear, while arguing the country needs additional units operating by the mid-2030s [transcripts/2026-q2] — a discipline signal after the $14B→$35B Vogtle history.
Net income by registrant, Q2 and H1:
| Registrant | Q2'26 rev | Q2'25 rev | Q2'26 NI | Q2'25 NI | H1'26 NI | H1'25 NI |
|---|---|---|---|---|---|---|
| Georgia Power | $3,133M | $3,110M | $779M | $607M | $1,408M | $1,204M |
| Alabama Power | $1,963M | $1,968M | $437M | $381M | $862M | $755M |
| Southern Company Gas | $966M | $979M | $126M | $106M | $573M | $524M |
| Mississippi Power | $403M | $400M | $52M | $59M | $112M | $114M |
| Southern Power | $535M | $546M | $(25)M | $51M | $(22)M | $138M |
| Consolidated (to Southern) | $6,977M | $6,973M | $1,174M | $880M | $2,531M | $2,214M |
Three things worth naming:
Q2 2026 (10-Q, filed 2026-07-30):
H1 2026: revenue $15,374M vs $14,748M (+4.2%); operating income $3,794M vs $3,774M (+0.5%); net income to Southern $2,531M vs $2,214M (+14.3%); diluted EPS $2.23 vs $2.00. Adjusted H1 $2.46 [transcripts/2026-q2].
Guidance. FY2026 adjusted $4.50-$4.60, now expected "near or at the top." Q3 2026 adjusted estimate $1.50 per the call transcript [transcripts/2026-q2] — note that at least two web summaries report $1.65 for the same figure; the transcript body is primary and I use $1.50. Conflict surfaced, not resolved. Long-term guidance (8-9% through 2028, 5-7% raised toward 8% through 2030) is carried from the prior dive and was not re-verified this run.
Balance-sheet flags at 2026-06-30:
Market reaction: the news is landing and the stock is not. SO closed Q2 at $95.71 (2026-06-30) and trades at $91.33 on 2026-08-10 — −1.9% against the $93.09 price at the prior dossier, despite a beat, a guidance raise, a 3.2 GW OpenAI contract and the largest loan in DOE history. Consensus rating has slipped from Moderate Buy to Hold (6 Strong Buy / 1 Moderate Buy / 16 Hold / 2 Strong Sell) with a mean target of $102.02 and a high of $114. Trailing P/E 22.32, forward 20.00, yield 3.33% on a $3.04 dividend, beta 0.33. Peer multiples were not re-sourced this run — see Lens 7.
The prior dive had no transcript and synthesized sentiment from coverage. transcripts/2026-q2.md is now the first primary call document in the shelf, so this is a step up in evidence grade rather than a like-for-like sentiment comparison. Against the Q1 2026 and FY2025 commentary the prior dossier characterised:
No single ≥5% session was identified in the 2026-06-21 → 2026-08-10 window. What the window actually shows is more interesting than a big move:
| Event | Date | Stock response |
|---|---|---|
| Q2 end | 2026-06-30 | $95.71 |
| OpenAI 3.2 GW / 25-yr contract announced | 2026-07-23 | +0.67% that day (+1.4% the prior session) |
| Ratepayer Protection Pledge | 2026-07-23 | no discernible move |
| Q2 print: $0.14 beat, guidance to the top | 2026-07-30 | traded largely unchanged; <2% decline |
| Current | 2026-08-10 | $91.33 |
The pattern has changed since the prior dive. The prior dossier said "each upward load-forecast revision has re-rated the stock." That stopped being true in this window: the largest single-site utility AI contract in US history moved SO less than 1.5%, a $0.14 beat moved it not at all, and the shares are down ~2% while the analyst consensus slipped from Moderate Buy to Hold. Either the market has fully discounted the data-center story, or it is discounting the delivery risk between contract and cash flow — which, given that only ~1.2 GW of 17 GW is flowing, is the rational thing to discount.
The catalysts that remain live are unchanged in kind: Georgia PSC outcomes (the certification appeal, the 2028 cost-of-service study, post-2026 commissioner posture), the load ramp's actual delivery in 2028+, 10-year Treasury moves on a 0.33-beta bond proxy, and now the convertible conversion thresholds at $104.56 and $118.50 — which sit 14% and 30% above spot and become a dilution mechanic if the stock ever does re-rate.
Correction to the prior dossier: it named only Womack. The CFO is David P. Poroch, promoted from SVP/Comptroller/Chief Accounting Officer, effective 2025-07-31, succeeding Daniel S. Tucker (retired 2025-10-01). This predates both dossiers; it was a gap, not a change. Poroch is an accountant-CFO rather than a treasury/markets CFO — worth noting given that the single most consequential management action of the past six months has been a financing one.
Capital allocation, refreshed. The Q2 record is genuinely good execution on the thing that was most criticised: securing the DOE facility, cutting the residual equity need to $1.1B, pre-selling equity via forwards at $88-97 rather than dumping stock into weakness, and raising the dividend to $0.76/quarter ($3.04 annualised, 26th consecutive year). Against that, the $2.73B convert issued three days after the call is a financing choice that does not appear in the "$1.1B equity need" framing and deserves a question (Lens 14, new item).
Archetype and comp: unchanged — see prior dossier.
Regulatory findings re-run 2026-08-10: 0 SEC Litigation Releases, 0 AAERs naming Southern Company since 2021-08-10. No accounting-enforcement exposure.
Flags carried forward and now sharper:
Conclusion unchanged: no accounting or enforcement red flag. The genuine forensic concerns are (a) an earnings mix leaning heavily on construction-period credits and a low tax rate, (b) negative FCF funded by debt and equity, and (c) a share-count picture that is larger than the headline equity-need number. All three are consequences of building $19.5B of certified plant into a rate base, not of anything improper.
Model-workbook integrity flags (read these before any number below). model.xlsx was rebuilt this run from the four quarters now in our figures (2025-Q1 → 2026-Q2). our model reports: "No computed values" — 51 formulas, no cached results, so every valuation output is blank; the opening balance sheet is "partly sourced (8/10 lines)" (other current assets and other current liabilities missing); gross margin was seeded at a conventional 50% because a utility's CSV carries none, and revenue growth was seeded at 8% because fewer than eight quarters exist; no share price is set. I therefore cite nothing from the workbook. The path below is arithmetic off guidance, shown in full.
Anchors. FY2025 adjusted ≈ $4.26. FY2026 guidance $4.50-$4.60, "near or at the top." H1 2026 adjusted actual $2.46; Q3 estimate $1.50 → implied Q4 of $0.54-$0.64 to land the range. Long-term guidance 8-9% through 2028 (carried, not re-verified).
| Scenario | FY2026E | FY2027E | FY2028E | Logic |
|---|---|---|---|---|
| Base | $4.60 | $4.97 | $5.37 | Top of the guided range as management now signals; 8%/yr thereafter as certified plant enters rates and DOE-funded debt holds the interest line. |
| Bull | $4.62 | $5.08 | $5.59 | 10%/yr: the 2028 load ramp lands early, Alabama/Georgia RFP capex converts to rate base, DOE draws replace market debt faster, converts stay out of the money. |
| Bear | $4.52 | $4.79 | $5.03 | 6% then 5%: the tax rate normalises toward 21%, AFUDC and capitalised-interest credits roll off as plants enter service, the 2028 cost-of-service study shifts allocation, converts dilute. |
Implied valuation at $91.33: base FY2027E $4.97 → 18.4x forward; base FY2028E $5.37 → 17.0x. Trailing is 22.3x and forward (consensus) 20.0x. The stock is no longer priced at the ~24x the prior dive flagged — the multiple compressed roughly two turns while earnings rose, which is a de-rate the prior dossier's bear case anticipated and the prior dossier's bull case did not.
The single number that governs this projection is not EPS growth — it is the ratio of flowing to contracted load. At 1.2 GW flowing against 16-17 GW contracted, ~93% of the thesis is a 2028-2032 event. A scoreable base call would be: "SO FY2026 adjusted diluted EPS ≥ $4.55, p=0.80, resolves 2027-02-28". Not logged — no forecast is created in an unattended run.
Bull case (stronger than at the prior dive). The demand is no longer a forecast — it is ~16 GW of PSC-reviewed and agreed contracts, all with minimum bills, 13 GW with termination payments and posted collateral, capped by a 25-year OpenAI agreement for 3.2 GW where the customer funds its own infrastructure and hands back 1,000 MW of demand response. The financing objection has been answered with the largest loan in DOE history — $26.5B at Treasury+37.5bp to 2055 — which simultaneously cuts customer bills ~$7B and defuses the affordability politics that were the thesis-killer. Residual equity need through 2030 is $1.1B. Retail sales just grew 2.3%, the fastest first half in ~20 years. Hatch is relicensed to 2054/2058. And the multiple has already compressed to ~18.4x FY2027E base on a 3.33% yield with 0.33 beta. You are being paid to wait for a ramp that is under contract.
Bear case (permanent-impairment risks, reordered).
Pre-mortem (18 months out, thesis broke). It is early 2028. The load ramp slipped a year — permitting, turbines, or a hyperscaler capex pause — while the certified plant entered service on schedule and began depreciating. The tax rate normalised toward 21% as construction credits rolled off, AFUDC fell as CWIP converted to plant, and capitalised interest stopped flattering the P&L. Adjusted EPS growth printed 4%, not 8%. The 2028 Georgia cost-of-service study landed in an election year and allocated more to the utility than to the data centers. The stock, already de-rated from 24x to 18x, went to 15x. Nothing was fraudulent; the sequencing was simply wrong, and the market had been telling you so since July 2026 by refusing to pay for the OpenAI contract.
Contrarian view (what the market refuses to see) — updated. The prior dive argued the market underweighted the concentration risk. Six weeks later the market appears to have fully absorbed it: a landmark AI contract moved the stock 0.67%, the consensus went to Hold, and the multiple compressed two turns into a beat-and-raise. The refusal has flipped sides. What is now under-appreciated is that the DOE facility structurally changed the return math on $26.5B of the build — 30-year money at a 37.5bp spread, in a sector where every competitor is financing the same race at market rates, is a durable edge that shows up in ROIC for three decades, and the market spent zero re-rating on it.
Dismantling the refreshed bull case:
the previous dossier)n/a.Every dossier we have written on Southern Company, newest first, including where a later one corrected an earlier one.
The two mechanical bear points got repaired
CorrectionCorrection to the prior dossier: it named only Womack.
A regulated-utility levered call on the Georgia data-center build-out
Covered in the Knowledge Base
Energy & Power
| Industry | Energy |
| Size | Public Company |
Where Southern Company sits against the other names we cover on this beat. Each line is that company’s verdict, not a summary of it.
The subsidy thesis just got a second leg — a $0.38/W federal module price floor from the Aug 6 Section 232 proclamation now sits ABOVE First Solar's o…
Cash $1.7B
The franchise got stronger and the price got cheaper
Cash $1.2B
The order line decelerated and the backlog shrank
Cash $256M
A best-in-class, AI-demand-levered renewable platform run through an externally-managed structure that skims ~$223M/yr in base fees plus IDRs off the…
A $1.1B-cap company sitting on a $3.5B storage backlog and 83 GWh dev pipeline, but the equity is a levered residual claim junior to $2.2B non-recours…
Cash $1.4B