Constellation Energy Corp (CEG.NASDAQ) AI OPINIONS & ADVISOR ANALYSIS

The bull case activates when a named hyperscaler signs multi-gigawatt capacity at premium pricing while the ten-year yield falls back toward 4%. Contracted EBITDA visibility rises, short interest at 3.28% of float unwinds violently, and the buyback retires shares into a rising earnings base. A FERC co-location framework then legitimises behind-the-meter economics across the fleet, and any relaxation of PJM's capacity collar adds pure margin. Constellation re-rates toward regulated-utility multiples on merchant-scale earnings.

The bear case starts if AI capital discipline arrives before the contracts do. Hyperscaler load forecasts get cut, PJM and ERCOT forward curves sag, and the June 2027 lockup dumps seller shares into a market with no narrative to absorb them. Guidance trims follow the weaker ERCOT realizations already visible in Q2, the buyback becomes dilutive support rather than accretive arithmetic, and the stock reverts to a conventional merchant 10-12x EBITDA valuation near the 52-week low.

The consensus calls this a broken AI-power trade awaiting a catalyst: street mean near $367 against a $255 tape, with Morgan Stanley lifting to $369 on 18 September [15]. The crowd treats the Calpine equity overhang as mechanical and the data-center PPA drought as pure timing. The anchoring bias is the $412 peak, not the cash flows underneath it.

The crowd modestly underestimates these shares, though far less than the $367 street mean implies. Two pieces of evidence point the same way: management repurchased at $281 and $285 [20], and short interest rose 23% to 3.28% of float by 31 August [13], a crowded bearish lean into a name with a legislated PTC floor and 18.5-year contracted PPAs. The blind spot cuts both ways, however: bulls ignore that Henry Hub at $2.91 caps merchant upside, so fair value sits nearer $300-320 than $367.

The Q3 2026 print in early November is the verification event: H1 delivered $5.29 of adjusted EPS against a $12.00 midpoint, so roughly $6.71 must land in H2. Delivery plus a named data-center contract would force short covering; the first observable sign is forward multiple expansion back above 22x.