The cheapest line on the BOM got cheaper again last week, and it did not help. Wood Mackenzie and SEIA put residential module prices down 16% year-over-year at $0.37/W in Q2, and turnkey installed prices went up anyway in every segment except residential, because freight rose 15% across the board. That is the shape of the year: the commodity is deflating and everything wrapped around the commodity — trucking, entry paperwork, pile tolerance, soiling, headcount — is not.
The same week, Commerce made it materially harder to front-run the December floor prices, the E2 count finally put a number on last year's job losses, and two vendors shipped products that exist because installed cost now gets decided in the field rather than at the purchase order. If you buy, schedule or crew, last week was about the margin that lives outside the module.
Residential module prices fell 16%. Commercial installed cost still rose 5.6%
The pricing half of the Q2 Wood Mackenzie and SEIA analysis landed on 24 September, and it splits cleanly by segment. On their turnkey $/W basis:
- Residential: $3.36/W, down 1.4% year-over-year — the only segment that fell.
- Commercial: $1.77/W, up 5.6% — the steepest inflation of any segment.
- Utility-scale, fixed-tilt: $0.95/W, up 0.9%.
- Utility-scale, single-axis tracking: $1.06/W, up 2.0%.
The interesting number is not the installed price, it is who got the module discount. Residential module prices dropped 16% to $0.37/W; utility-scale module prices fell 2%, to $0.33/W. Wood Mackenzie and SEIA attribute the drop to the invalidation of the IEEPA tariffs earlier this year. So the segment with the least buying power got the biggest per-watt relief, and then handed most of it back: logistics and freight rose 15% on average across all segments, against roughly 50% year-over-year growth in oil and gas prices.
For anyone pricing commercial work, this is the quarter to stop quoting module savings to a customer. A 5.6% increase on $1.77/W is real money on a 2 MWdc rooftop, and it did not come from the panel. Redline your freight assumption before you redline your BOM.
Commerce caps new importers at 55 modules a week until 4 December
This is the most operationally concrete thing that happened all week and it got the least coverage. BIS published a temporary final rule — Docket No. 260915-0004, RIN 0694-AK57, 15 CFR Part 705 — effective 22 September through 3 December 2026. It exists to stop importers pulling volume forward ahead of the Section 232 minimum import prices, which are floor prices below which duties apply, not duty rates, and which bite on goods entered for consumption on or after 4 December.
Any importer of record that registered with CBP on or after 6 August 2026 — the date of Proclamation 11052 — is capped, absent a Commerce waiver, at these weekly volumes by HTSUS subheading:
- 2804.61.00 (polysilicon): 12 kg
- 3818.00.00 series (ingots and wafers): 7 kg
- 8541.42.00 (cells): 2,000 units
- 8541.43.00 (modules): 55 units
Fifty-five modules a week is not a supply channel, it is a rounding error — roughly a pallet. The rule is explicit that these caps were set from historic data so a new importer of record cannot exceed what an established one typically brings in, and that Commerce and CBP will act against anyone standing up multiple IORs to get around it. Existing importers are not exempt: Commerce is comparing each one's weekly average since 6 August against its January-to-August average and its 2025 average, and an importer found bringing in volumes "substantially greater than their historic averages" is barred from further entries until 4 December, waiver applications to [email protected].
Practical read for anyone with a signed EPC contract and modules not yet on the water: the safe-harbour-by-stockpiling play is closed, and your supplier's plan to spin up a fresh import entity is now a compliance risk rather than a workaround. Ask your supplier which IOR is clearing your containers and when that entity registered. If the answer is "a new one", your delivery date is a Commerce decision.
E2 counts 37,000 clean energy jobs lost in 2025, and 11,500 of them in generation
E2's Clean Jobs America 2026, published 22 September and built on Department of Energy data analysed by BW Research Partnership, puts 2025's clean energy job losses at 37,000 — the first annual decline since the pandemic, wiping out nearly 40% of the previous year's gains. Renewable energy generation, the category that carries solar, lost 11,500 jobs, a 2% decline. Energy storage and biofuels both posted slight increases. Total clean energy employment remains above 3.5 million.
Two things to keep straight. First, this is the E2/BW Research series, not the IREC National Solar Jobs Census — different methodology and occupational cuts, so do not splice the two into one trend line. Second, the direction inside the number is what matters: generation down, storage up. The migration visible in resi crews for two years is now in the national count.
A strong El Niño is making underperformance arguments unwinnable
The Solargis seasonal review published 25 September is nominally a resource-assessment story and is actually an O&M and warranty story. Global horizontal irradiance anomalies ran more than 20% below the long-term average across the Paraná Basin, Argentina, Uruguay and southern Brazil, and more than 20% above it in southern England, Wales and Normandy. In the continental US the monthly swings largely cancelled out across the summer, leaving a comparatively flat seasonal signal, with the sharpest anomalies through the central states where temperatures ran more than 2.5°C above the long-term average.
Variations in the solar resource may be due to a number of complicating factors that are not easily measured or accounted for with historical data and forecasting models. — Marcel Suri, CEO, Solargis
That is the problem stated politely. Stated as a work order: when the resource moves this far from the P50, separating weather from genuine asset underperformance gets much harder, and that argument is one an O&M contractor has to win against an availability guarantee. The detail worth writing down is Solcast's: soiling losses around Dallas hit 2% by late August for want of rain, the highest late-August figure in twenty years. A 2% soiling loss is inside the range where a plant looks like it has a string or tracker problem. Before you chase it with an IV-curve trace, check whether your wash cycle assumption survived the summer — and get the soiling number into the monthly report, because it is also the number that defends your availability position.
Amelia puts a five-second position check on micropile crews
Amelia, the Spanish firm behind the Hincator pile-installation device, added a Micropiling Mode on 22 September: a hardware-and-software change that measures pile position in real time and reports deviation on the X, Y and Z axes plus rotation and inclination. Amelia's claim is five seconds per measurement against fifteen for the standard mode, a complete initial-measurement-adjust-verify cycle in under a minute, and inspection of up to 350 piles in a day.
Take the throughput figure as a vendor number until your own crew reproduces it. The reason it matters is the foundation type. A driven pile that comes in out of tolerance can be pulled and redriven; a micropile set in grout gives you one attempt before the concrete decides for you, and the rework shows up as torque tube that will not reach and a tracker row that will not align months later. Anything that moves the position check to before the pour, rather than to the surveyor's report afterwards, comes out of the rework budget directly. If you run pile crews on terrain-following rack, this is the category of tooling to price into next season's bid.
Sunrun crews are about to be installing liquid-cooled GPUs
Span named Sunrun as installation partner for XFRA, its distributed data-centre product, on 23 September. XFRA puts compute nodes — liquid-cooled NVIDIA GPUs, orchestrated by Span — into residential and small commercial spaces. Sunrun supplies the PV and multi-battery storage that powers them, plus its smart panel, enrols the batteries in grid service programmes, and provides whole-home backup. The two companies are going through homebuilders on new construction, starting in Texas. No deployment volumes or node counts were disclosed.
Ignore the AI framing and look at the scope. A residential PV and storage crew is being asked to commission a permanent, high-density, liquid-cooled load inside a dwelling, sized around someone else's compute schedule rather than the household's. That raises questions a foreman should ask now rather than on the first job: who owns the cooling loop, what the load calculation looks like when the biggest circuit in the house is not the HVAC, which AHJ in Texas has seen this before, and what your general liability carrier says about water in a garage you also wired. New construction is the right place to trial it, because plan review happens once. It is still a scope your crews are not badged for.
Solar for All is back, and $6.95 billion of it never moved
Judge Mary S. McElroy of the US District Court for the District of Rhode Island granted summary judgment to the plaintiffs and vacated EPA's termination of the $7 billion Solar for All programme, holding that the agency "acted contrary to congressional intent and without statutory authority" and that Congress intended EPA to keep administering grants it had already obligated. pv magazine USA reported it on 21 September; Solar Power World the same day. Sixty prime award recipients are covered, and the programme was projected to reach more than 900,000 households.
The number that tells you what to do is $53 million. That is all that had been disbursed of the $7 billion before the August 2025 freeze, which means essentially the entire grant pipeline is unstarted rather than half-built. EPA says it is reviewing the ruling and weighing an appeal, so nobody should be hiring against this yet. But if you install low-income residential or community solar, find out where your state's prime recipient sits in its programme design: the gap between a programme that resumes and one that restarts is about a year of your pipeline, and the recipients who kept staff through the freeze will issue work first.
The through-line for the next ten weeks is that nothing on this list is decided at the module. Freight, an import entity's registration date, a soiling curve, a pile out of tolerance by an inch, a scope your crew has never been badged for — that is where the margin and the headcount now sit. The roles that survive a contracting installation market are the ones that own one of those variables. We keep the openings that match at career.solar/jobs.