Vicor's Fab Expansion Is Real; Its Earnings Now Sit in the Patents

The market read Vicor's September 11 announcement as confirmation that the AI power build-out is real, and it said so with its feet: the shares rose more than 11% the day the power-components maker said it was buying two New Hampshire sites for a second and third ChiP fab. The physical facts are straightforward. A 334,000-square-foot building on 66 acres in Merrimack, plus 54 acres in Hooksett, will support Fab-2 and Fab-3 with a combined footprint of nearly a million square feet. Vicor's existing ChiP Fab-1 in Andover totals 320,000 square feet on 16 acres and is approaching full utilization, so the new sites roughly triple the company's manufacturing envelope. Fab-2 carries about a one-year lead time to initial deployment; Fab-3 has no disclosed timetable. Purchase prices were not disclosed.

The press release attributes the expansion to growing power-delivery requirements from advanced AI systems, hyperscalers, and OEMs — a demand story. Read through a supply-first lens, the capacity decision is the real tell, because it confirms where the bottleneck has migrated. The constraint in AI infrastructure no longer stops at GPUs or advanced packaging; it now reaches the power electronics between the rack and the silicon. Vicor's Vertical Power Delivery architecture, a 48V direct-to-load network that cuts distribution losses, sits exactly at that point, and Fab-1 at full utilization is the manufacturing signal that demand has met capacity. The fabs are the supply-side answer.

But there is a split in this business that the day's reaction risks blurring. The force lifting Vicor's earnings today is not the fabs. It is the licensing.

The constraint that pays is intellectual, not physical

In February 2025 ruled two Vicor patents valid and infringed, and barred unlicensed computing systems containing the infringing power modules from import, with cease-and-desist orders against Delta Electronics, Quanta, and their affiliates. The suppliers, and the OEMs and hyperscalers that built around them, needed a path back into the U.S. market, and Vicor supplied one: an all-inclusive license covering its converter topologies, control systems, and distribution architectures. The company expects this licensing practice to contribute nearly $300 million to revenues through 2026. In late May it raised its second-quarter revenue guidance to $142 million on royalties from an additional OEM licensee, then reported $143.35 million for the quarter, up 49% year over year, with EPS of $1.04 against a $0.65 consensus.

Licensing revenue is not module revenue. Most of it falls through to the bottom line — a single 2025 quarter of licensing exceeded the company's entire quarterly R&D spend — and it is why Vicor's blended gross margin sits near 59%. The economics have migrated from manufacturing power chips to owning the patents on them. Value creation now lives in the intellectual-property leg, and the physical fabs are a fulfillment channel for the product leg that license revenue does not feed.

What the capex actually buys

The new fabs buy product capacity, geographic redundancy, and U.S. sourcing that hedges the import risk Vicor itself helped create. They are not free. Vicor has spent roughly $33 million on capital expenditures over the trailing twelve months, and the fab build-out is a step-change from that baseline — on a balance sheet holding about $454 million in cash and no net debt, enough to fund the start, but with the depreciation and fixed cost landing on the product leg's margins.

That is the tension the pop papers over. The shares carry the expectation: a market capitalization near $9.1 billion against trailing earnings that put the stock at roughly 63 times trailing EPS and about 80 times EBITDA. The equity has roughly quadrupled off its 52-week low of $47.56 — and still sits about half its 52-week high near $383, a reminder of how hard this name moves. Much of the run is the licensing surprise, and licensing is, by construction, a legal-enforcement event with a finite patent lifetime. The fabs do not make that stream repeat; they only make the module business larger.

The operative question is not whether the fabs fill — Fab-1 at full utilization says they will. It is whether the licensing leg reoccurs after 2026 and whether product margins can carry a multiple built on royalty economics. If licensing rolls into recurring renewals and product gross margins climb as volume rises, the expansion is the rational supply build of a cyclical winner. If the settlement stream was the earnings story and the fabs add only volume and depreciation, the market will eventually pay 63 times something that looks mid-cycle. Watch gross margin on the product leg and whether the licensing line reprints past the 2026 figure — not whether the site work in New Hampshire stays on schedule.