Making DERs work for load growth

The premise LaFarge starts from is that the load growth consensus formed unusually fast and that its scale is now being understated rather than exaggerated. Electric vehicles were the largest driver of electrification growth two years earlier and are now, in his description, a distant third behind manufacturing and data centers. He raises the obvious skeptical response himself and calls it sober: the internet and personal computing were once forecast to create a legendary amount of new demand, and efficiency gains swallowed it. His answer is a robustness claim rather than a forecast. Haircut the data center and manufacturing projections by 80% or 90% and you still double the grid, which he argues makes doubling a degraded baseline rather than a high-side case. The reason to look at distributed resources in that world is not that they are cheap or clean but that they are fast: it is easier to build many small things quickly than a few large things slowly, and permitting, interconnection queues, environmental review and capital formation all bite hardest on the large things.

Distributed capacity procurement is his name for doing the ordinary thing with unusual assets. Strip the first word and it is what utilities have done for a century: work out how much capacity you need and where, buy it, and run it. Keeping the first word changes the asset rather than the logic. He argues most of that capacity ends up on customer property for a pragmatic reason rather than an ideological one, which is that in the parts of the grid where capacity is actually needed, downtown Atlanta or Minneapolis being his examples, there is nowhere to put batteries except on and around buildings. The commercial form is a host agreement: the utility rents the space and pays the building owner a long-term annuity for the asset’s life, with no debt, no financing and no maintenance obligation on the host, often plus first call on backup power or a guaranteed share of the battery during an outage. Set against a virtual power plant, he calls this a utility-led version. The assets are paid for by the utility as grid infrastructure and dispatched by the utility, possibly through an aggregator’s control layer, and the starting point is solar and batteries with thermostats, water heaters and vehicles as later additions.

The economics are where he concedes most and recovers it with a value stack. Distributed resources are more expensive per megawatt, and their accredited capacity is lower. His first illustration is a comparison Kann rejects, a gas plant accredited near 80% against standalone solar at 8%, and he accepts the correction that the fair comparison is rooftop solar against utility-scale solar, leaving a residual of construction cost spread over fewer megawatts plus, as Kann adds, the soft costs that get worse as projects get smaller. Pairing solar with storage pulls accreditation back into the high fifties or low sixties, and his rule of thumb is that about 1.6 gigawatts of nameplate solar and storage buys a gigawatt of accredited capacity. The premium becomes payable only if you count everything the sited asset displaces: compare the accredited capacity against a peaker plant, add congestion value where the asset sits on a constrained part of the network, subtract transformers that no longer need replacing, which matters because transformers are expensive and supply constrained, and then subtract substation rebuilds, feeder upgrades and even transmission. He says that math looks good so far while stressing that it is early and that utilities and regulators are only beginning to do the modeling that would locate the value. The rest of the premium, on his account, gets paid by large new loads, not because they are rich but because they value time to power, which he ties to a 1947 analogy about grid investment unlocking the postwar manufacturing economy.

The limit gets put on the table by both of them, which is the most useful part of the conversation. LaFarge’s thought experiment is that Microsoft’s 800 megawatt contract for the restarted Three Mile Island unit delivers power onto the grid rather than next to the data center, so in principle 800 buildings each hosting 1.6 megawatts of solar and storage could substitute for it. Kann’s pushback is that nuclear runs around the clock and solar plus storage does not, so the substitution holds for one project but not at system level, where you would still need more storage, more generation, or gas backup. LaFarge does not resist. He says he does not know whether the accreditation arithmetic breaks down at system scale for physical reasons such as inertia, load management and conductor temperature, suggests accreditation may be more a contractual and market construct than a physical one, and invites listeners to correct him. On scale limits generally his answer is that nobody knows, because it has not been tried at this scale. Both of them land in the same place: distributed belongs in the mix on speed grounds, and for every public conversation about a nuclear restart there should be one about a feeder full of buildings.