Open in interactive viewer → charts, metric popovers & call review

ERock Delivers the Quarter It Promised—Into a Market That Stopped Paying for It

First profitable quarter, record bookings, and a new Houston plant. The stock is still down 32% in six weeks, because the AI-power complex is being repriced, not rewarded.
EROC · Earnings Call · 2026-11-12

Delivered, and de-rated

ERock did the thing newly public companies rarely do: it hit the number. Third-quarter revenue came in at $153 million against $40 million in the second quarter, adjusted EBITDA swung to $12.6 million from negative $14 million, and net income of $8.9 million marked the company's first profitable quarter since listing. The full-year outlook of $435–465 million in revenue and $3–9 million of adjusted EBITDA was affirmed, implying a Q4 of $211–240 million in revenue that would be a company record. And the tape shrugged. Over the 32 trading days to the report, EROC fell 32%, closing 38.9% below its October 6 high of $13.64. That is not a verdict on the quarter. It is a verdict on the neighborhood. The global tape over the last 30 days shows the AI data center trade with 13 advancers against 57 decliners, and over 90 days the data center capacity basket is off roughly 30% with nine of nine constituents negative. Adjacent reads—high bandwidth memory, co-packaged optics, AI cloud services—are all in the same drawdown. ERock is a pure play on speed to power, and speed to power is precisely what the market has spent six weeks marking down. That gap is the story: a company executing on schedule sold as a proxy for a theme that is unwinding.

The backlog line nobody wanted to read first

“Contracted Power System sales backlog was approximately $1.6 billion at September 30 compared with $1.7 billion at June 30.” — R. Ian Blakely, CFO · 2026-11-12 A sequential backlog decline on the day the stock is already broken is the kind of data point that gets extrapolated into a thesis. Management's defense is credible and specific—backlog converts as Hyperion output and El Paso installation run through the income statement, and roughly $300 million of bookings signed after quarter-end, taking backlog to a record. What matters is the composition. The backlog is deliberately narrow: firm contracts only, deposits in hand, with the balance-of-plant and installation scope contracted separately, often three to nine months after the equipment purchase agreement while site design iterates. That means the reported number structurally lags the commercial reality—and also that it can look lumpy in either direction. The more interesting move is narrative. On the prior call, data center was the company's single highest-ranked keyword; this quarter it slid well down the list while El Paso, Bridge Power and service revenue took the top slots. John Carrington made the pivot explicit: “On our last call, we talked a lot about data centers. So I want to spend a moment on our commercial and industrial business.” — John Carrington, CEO · 2026-11-12 What follows is a reshoring pitch—Wistron, drone and robotics manufacturing, pharma, hospitals, 5-to-100-megawatt loads—where diesel displacement and multi-year interconnection waits create the same urgency without the headline risk of a single hyperscaler. Whether that is diversification or a hedge against the very theme the stock is priced on, both readings are defensible. But it is a deliberate editorial choice, and it is new.

Atlas, and the arithmetic of capital-light

The expansion is the real news. Atlas—650,000 square feet in Northeast Houston, 60% larger than Hyperion—should ramp in late 2027 and reach roughly 900 megawatts of annual capacity by the first half of 2028, expandable to 1.2 gigawatts, taking total capacity to about 2.1 gigawatts. It sits near gas supply for testing and near the Ship Channel for logistics. The problem is the price tag, and the analyst on the call went straight at it. On the August call, Carrington framed expansion as “a new facility in a 12-month timeline at 1.2 gigawatts at a cost of $15 million to $20 million.” — John Carrington, CEO · 2026-08-12 Atlas is $20–30 million for 900 megawatts. Divide it out and the cost per gigawatt has roughly doubled against the pre-IPO framing. Ian Blakely was candid about the drivers: “there's a little bit of general cost inflation,” — R. Ian Blakely, CFO · 2026-11-12 plus optional tooling and automation that would raise upfront spend to lower unit costs later. In absolute terms the capital light model still holds—a gigawatt of capacity for the price of a mid-size factory, self-funded from $549 million of unrestricted cash, zero debt and an undrawn $250 million revolver. In per-unit terms, the Atlas math is worse than the pitch. That is exactly the kind of small, checkable erosion that matters more than a headline beat.

Margin math: priced in 2025, banked in 2028

Adjusted gross margin expanded to 24.1% from 22.2%, but the composition is the story, and it is a rerun.

The projects we are delivering on in 2026 and 2027 generally have margins in the high teens to low 20s. However, we have seen a structural shift to margins in the mid-20s or higher for deals that we have priced more recently. That's a meaningful signal of our market position, but with so much near-term production already contracted, it's largely a 2028 and beyond story.

R. Ian Blakely, CFO · 2026-11-12
That is nearly verbatim the framing from August, when Blakely said the pricing lift was “more 2028 and beyond impact that we're seeing.” — John Carrington, CEO · 2026-08-12 Two quarters later, 2028 is still where the money is, and 2027 is largely locked because it was priced in 2025. The rolling deferral is not malfeasance—it is what sold-out capacity looks like—but investors should stop treating the mid-20s margin claim as an emerging fact. It is a promissory note, and its due date keeps sliding. The genuinely new value lever is duration. Bridge periods are stretching from a three-year assumption toward five to seven as interconnection queues lengthen, and ERock earns roughly three times the ongoing services revenue in bridge mode versus backup: “we're 3x in bridge versus outside of Bridge. So it's accretive to our model.” — John Carrington, CEO · 2026-11-12 A longer bridge with a use case that migrates from bridge to backup is a better annuity than the original model assumed—and it is an argument that grid dysfunction, the sector's problem, is ERock's pricing power.

What to watch

Two things. First, permitting. The El Paso air permit arrived in under 45 days—the air permit speed is the moat—and management says the December CCN is on track with no impact from Texas's debate over large-load interconnection, echoing last quarter's stance that Governor Abbott's Batch Zero commentary produced no impacts. Analysts are now asking about a potential moratorium under the next administration. That is a live, unquantified tail risk the company can only answer with confidence. Second, the Q4 print: $211–240 million of revenue, $17–23 million of EBITDA, and confirmation that the $300 million of post-quarter bookings actually lands in backlog. If they do, the current drawdown will look like a beta event in a de-rating theme rather than a credibility event. If backlog slips again and Atlas costs drift, the market will conclude the opposite—that the capital-light story was always a measure of the balance sheet, not the unit economics.