Bilicic and Wood/Mackenzie both find renewables cheapest but many data centers are choosing gas. Solar panels and batteries are in good supply worldwide. Installation takes about the same time. I’m trying to understand why they are the hyperscalers almost ubitous choice.
Some think that the cost isn’t that low everywhere. Some think acquiring the land and getting permits for renewable takes too long. (BAAJA in Congress would change that and may happen quickly.) Utilities don’t want competitive solar and may hold them back.
Or it may be that gas, in the U.S., is cheaper. Not in Europe or Africa, of course.
I’m still looking for more data. I’m happy to present Bilicic’s thought. He has been a top analyst to industry leaders for years. I hope he’s right. Musk is deploying some very dirty generators. Dave
Source: George Bilicic, Lazard’s global head of power, energy & infrastructure, in conversation with hosts Laurent Segalen and Gerard Reid on the Redefining Energy podcast, episode 252, “Lazard, LCOE and the real cost of power,” Oct. 26, 2026. Link
Lazard’s top power banker says the AI-driven building boom looks nothing like the speculative 1990s bust, because today’s plants come wrapped in long-term contracts — but renewables’ cost edge is narrowing, nuclear stays stubbornly expensive, and Darlington’s celebrated cheap-SMR number doesn’t survive contact with Lazard’s own data.
Every year, Lazard’s power bankers publish a report that functions as the industry’s shared scoreboard: the Levelized Cost of Energy+ (LCOE+), now in its 19th edition, which prices out what it actually costs to build and run a power plant. George Bilicic, who runs Lazard’s power, energy and infrastructure group, came on Redefining Energy to walk through this year’s numbers — and to explain why Wall Street isn’t panicking about the data center buildout the way it probably should be, if history were any guide.
Why this boom isn’t the last one
The last time the U.S. built power plants at this clip, it ended badly: the early-2000s cycle left developers holding plants exposed to open markets that never grew into the demand bankers had projected. Bilicic’s pitch is that this time is structurally different — new generation is backed by long-term contracted cash flows rather than a bet on spot prices, which shifts the risk from “will anyone buy this power” to a narrower question of terminal asset value. It’s a real distinction, though it’s also the same reassurance an investment bank has an obvious interest in giving just as its clients pour capital into exactly this kind of project.
Renewables still cheapest, but everything’s getting pricier
Lazard’s 2026 LCOE+ confirms what the report has said for years: unsubsidized wind and solar remain the cheapest new-build power in the U.S., with onshore wind as low as $37/MWh and utility-scale solar around $40/MWh — both still roughly in line with the marginal cost of running existing gas and coal plants. But the gap with everything else is narrowing. New-build gas combined-cycle plants now range $51–129/MWh, new coal $72–177/MWh, and new nuclear $175–255/MWh — the nuclear figure built almost entirely off public cost data from Georgia’s Vogtle 3 and 4, still the only recent U.S. nuclear benchmark Lazard has to work with. Bilicic’s own figure on the podcast — “$180 to $200” — sits comfortably inside that range. Wind and gas costs, he said, have been climbing faster than inflation for three or four years, driven by equipment and labor costs rather than any technology regression.
The Darlington number doesn’t check out
Pressed on Ontario Power Generation’s SMR project at Darlington — where officials have floated a cost near $100/MWh for the GE Hitachi-built reactors — Bilicic was blunt: Lazard’s own review of the data suggests the real number is “much higher.” He has a point. Ontario’s own grid operator, the IESO, has pegged the project’s average lifetime cost at 14.9 cents per kWh (roughly $149 CAD/MWh, or somewhat under $110 USD at current exchange rates), with a stated range running from 13.5 to 18.4 cents — a band that only grazes $100 USD/MWh at its very low end. That’s still well short of Lazard’s $175–255/MWh range for large-scale U.S. nuclear, but it undercuts the cleaner “$100” headline number now circulating, and it’s worth remembering OPG hasn’t broken ground yet on units two through four, where the real cost discipline will be tested.
Batteries are already reshaping gas
One trend that got less attention than it deserved: batteries are cutting into gas plants’ dispatch hours in Texas and California, pushing developers toward smaller peaking units (CTs) instead of big baseload combined-cycle plants (CCGTs) — except that data centers’ appetite for round-the-clock power cuts the other way, keeping baseload gas in demand even as batteries nibble at the margins. Bilicic also expects a slow drift back toward vertically integrated utilities in deregulated U.S. markets outside Texas, on the logic that it’s simply easier to finance big, complicated power plants when one company controls generation, transmission and distribution together — a structural swing worth watching given how much capital has been deployed on the opposite assumption.
What’s not coming soon
Fusion remains, in Bilicic’s words, roughly “20 years away” by the account of the academics Lazard consults — a number that’s been stable for a long time regardless of how much capital chases it. Geothermal gets credit as promising but is, in his view, a niche technology that can’t scale to meaningful grid share. And despite over a decade of talk about electrifying heat, Bilicic doesn’t see natural gas heating losing ground in cold U.S. regions any time soon; the economics, he said, just aren’t there. The upshot for the next decade: since nuclear can’t be built fast enough to meet near-term data center demand, gas, renewables and storage will carry the load — and U.S. emissions are set to rise before they fall again.
Other opinions on these numbers
Not everyone treats Lazard’s “renewables remain cheapest” headline as settled. Critics on the right — the Independent Women’s Forum and the Institute for Energy Research among them — argue the real story is that existing, already-paid-off nuclear and coal plants now beat new renewables on cost, and that pitting new wind and solar against new fossil and nuclear plants, as LCOE does, obscures that. A sharper critique, pushed by commentators like David Turver and the Energy News Beat newsletter, calls for scrapping LCOE altogether in favor of a “full system cost” metric that bakes in transmission, backup capacity, and grid-reliability costs they say LCOE ignores entirely. Lazard has partly conceded the point: this year’s report, as Heatmap News noted, acknowledges that gas plants carry value beyond their raw per-megawatt-hour cost because they can be dispatched exactly when the grid is under stress. Clean-energy outlets like CleanTechnica and RenewEconomy, by contrast, take the “renewables remain cheapest” finding largely at face value. And on the nuclear side, some advocates note Lazard’s entire new-nuclear range rests on a single data point — the over-budget Vogtle 3 and 4 — which may not represent what a mature, repeat-build U.S. nuclear program would actually cost.
Lots of claims for nuclear costs are based on new designs. Be very skeptical until they are proven. Dave
authored by Dave Burstein with close collaboration by Claude