Southern Company

The load-growth thesis converted from pipeline to signed contract at scale this quarter: contracted large load went from 11 GW to roughly 16-17 GW on ~6 GW of new deals including a 3.2 GW, 25-year OpenAI agreement, realised data-center usage accelerated to +55% y/y, and management moved FY2026 to the top of its range — all while the stock got cheaper.

• Roughly 6 GW of new large-load contracts were signed in a single quarter, taking the contracted total to ~16 GW in the 10-Q (management cited over 17 GW on the call) from 11 GW at Q1 — the fastest conversion of the pipeline the company has reported
• The OpenAI deal is the marquee validation: 3.2 GW over 25 years in Georgia, phased 2028-2032, with OpenAI paying the full cost of the electric infrastructure and service required, plus 1 GW of flexible demand response from 2028 — one of the largest single-facility demand-response commitments in the US
• Realised load is accelerating faster than the prior quarter, not decelerating: data-center usage grew 55% y/y in Q2 versus the 42% reported in Q1, with system-wide data-center load now above 1.2 GW and rising more than 500 MW a year
• The contracts are structurally de-risked in ways the filing spells out: ALL Large Load Contracts carry minimum-bill provisions, and ~13 GW worth also carry minimum duration, termination-payment requirements and collateral or guarantee posting scaled to customer credit — with ~$21B of aggregate collateral supporting an A- or better position. This directly attacks the stranded-asset objection
• Adjusted EPS of $1.13 beat management's own estimate by $0.13 and the full year moved to 'near or at the top' of the $4.50-$4.60 range, with a $1.50 Q3 estimate; GAAP net income rose 33% y/y
• Operating cash flow improved 24.7% to $4,280M in H1, and the equity programme is far more advanced than the capex plan implies — only $1.1B of remaining equity need by 2030 after $2.6B was raised in H1 and forwards were placed out to 2028
• The stock is cheaper than a quarter ago on better news: at $92.94, SO trades ~20.4x the $4.55 guidance midpoint (~20.2x at the $4.60 top) with a 3.27% yield

Strip out the below-the-line help and this quarter shows almost no growth: revenue was flat at +0.06% and operating income rose just 0.7%, while adjusted EPS grew 23% on a tax rate that halved, an absent prior-year debt charge, higher construction-period accruals and equity-method income — all funded by $2.1B of new shares in a single quarter and $2.36B of negative free cash flow.

• The quality of the 23% adjusted-EPS growth is poor. Pre-tax earnings rose $233M, but operating income contributed only $12M of that — the other $221M came from below the operating line: AFUDC equity +$48M, equity-method earnings +$76M, a $78M lower interest charge and $19M more other income. Then the effective tax rate fell from 25.3% to 13.6%, worth another ~$102M
• The apparent interest-expense relief is an artefact. The 10-Q states the $78M Q2 decline was 'primarily due to' a $129M reduction in debt-extinguishment losses plus $17M more capitalised interest, 'partially offset by' a $65M INCREASE from higher average outstanding borrowings. Underlying borrowing cost rose $65M in the quarter and $121M year-to-date, and cash interest paid rose 14.4% to $1,472M in H1
• Revenue was flat — +0.06% y/y — with retail electric revenues actually DOWN ($4,745M vs $4,758M). A load-growth story whose largest revenue line shrank in the quarter is not yet a revenue story, and the market said so by marking the shares down 1.69% on 2026-08-03
• Dilution is now material and accelerating: 23M shares issued in Q2 alone for $2,116M at roughly $92, with H1 common-stock proceeds of $2,596M against just $62M a year earlier. Average basic shares rose 3.3% y/y, mechanically taxing every dollar of per-share growth
• Free cash flow was approximately MINUS $2,359M in H1 as property additions of $6,639M (+21.7% y/y, 43.2% of revenue) outran a $4,280M operating cash flow. Long-term debt rose $3.1B in six months to $68,756M and net debt is ~$72.6B at ~1.78x total equity
• The adjusted-vs-GAAP gap is widening in one specific line: accelerated depreciation from wind repowering was $143M pre-tax in Q2 (vs $40M) and $296M in H1 (vs $65M). GAAP EPS of $1.03 versus adjusted $1.13 is a 10% haircut that is growing, not shrinking
• The OpenAI contract pushes Georgia beyond its recently approved capacity by roughly 1 GW on management's own account, so incremental generation approval — and the associated Georgia PSC cost-allocation risk — is now a live prerequisite rather than a hypothetical
• The 17 GW portfolio depends on ~$21B of aggregate customer collateral to sustain an A- or better credit position. That is enormous counterparty exposure concentrated in a handful of AI buyers whose own funding is not investment-grade in the traditional sense
• Management targets only 17% FFO-to-debt by 2029 — a thin cushion for a company adding this much debt, and one that leaves little room for a rate-case disappointment or a capex overrun