CPS Technologies Grew Its Margins, Not Its Revenue — and Pushed Its Factory a Year to the Right

CPS Technologies is a $67 million market cap maker of aluminum-silicon-carbide baseplates and hermetic packages — the unglamorous metal-and-ceramic guts that sit inside power modules on high-speed rail, grid gear, and increasingly, AI infrastructure. On November 4 it reported a quarter that, on the surface, looked like a coming-out party: gross margin of 20.9%, up from 14.8% in Q2 and 17.1% a year ago. The stock, however, went into the print down 30% over 90 trading days, sitting 85% below its February 2021 peak of $26.89.

That gap is the whole story. The margin is real and management-driven. The top line is not.

The record that isn't growing

Brian Mackey opened with the headline: revenue of $8.9 million, “a new quarterly record for the company for revenue as we continue to benefit from higher demand and increased production volumes.” — Brian Mackey, President and CEO · 2026-11-04 Three sentences later, CFO Chris Fraser described the same number as “revenue of $8.9 million for the period compared with $8.9 million for the third quarter of fiscal 2025.” — Chris Fraser, CFO or Senior Executive (likely CFO) · 2026-11-04 A record that is, by the CFO's own comparison, flat year over year. The filing history bears that out — quarterly revenue has oscillated between roughly $7M and $9M for six quarters.

The margin expansion is the genuine news, and Fraser attributes it to mix and discipline rather than volume: “gross margin expanded over 600 basis points from Q2, the best level in over three years.” — Chris Fraser, CFO or Senior Executive (likely CFO) · 2026-11-04 Some of that is pass-through — gold is billed at cost-plus per the CFO's answer to Chip Moore — but the rest is pricing and operational yield, exactly the lever that was broken through 2024 and 2025 as the company staffed a third shift. The gross margin series shows the round trip: 28.9% at the 2023 peak, a collapse to negative territory in early 2024, and now a climb back.

The move slipped, and the balance sheet is paying for it

The second story is a quiet schedule change with real cash consequences. CPS signed a 12-year-and-10-month new lease in August for 80,000 square feet in Attleboro, Massachusetts — double its current Norton footprint, three miles away — funded by a $9.6 million raise in May that netted $9 million. On this call, Mackey put a date on the build-out for the first time: “it's plausible that the build-out will continue into fourth quarter of 2027.” — Brian Mackey, President and CEO · 2026-11-04 Customers should note the permitting process is now the binding constraint — a new keyword for this company, and one that carries a distinctly different tone than the ebullience of prior quarters.

what we intend to do is outfit the new facility for our needs... and all of that would occur during calendar year 2026.

That was October 2025. The lease doesn't expire until February 2028, so nothing is at risk contractually — but a relocation originally framed as a 2026 event is now plausibly a late-2027 one. That has a measurable cost: inventory has ballooned to $10.3 million from $5.6 million at the start of the year, over $4.5 million of cash consumed to pre-build for the move. The free cash flow line has gone deeply negative, and management itself flagged the accounting wrinkle — building inventory under conservative standard costs is a margin headwind now that reverses as a tailwind during the shutdown.

The offset is a fortress balance sheet: $2.8 million cash plus $15.6 million in marketable securities, zero debt, plus a $0.5 million restricted security deposit. Effective net cash is around $22 million against a $67 million market cap — roughly a third of the equity value in liquid assets. That is the reason a 30% drawdown hasn't become a financing event.

Tungsten: the one genuinely new revenue line

The most interesting thing said on the call was buried mid-script. CPS won its first order from a military contractor for tungsten alloy components made by its QuickSet Injection Molding process, to be fulfilled before year-end, backed by a $1.15 million Army DEVCOM STTR award running through September 2027 for a controlled-fragmentation 40mm tungsten warhead. Mackey's pitch is speed: “there's not a huge iteration cycle of testing... it enables really a generally a rapid turnaround from interest by the customer to a potential PO to a potential shipment of product by us.” — Brian Mackey, President and CEO · 2026-11-04

This is a change in kind, not degree. In March, the same CEO was careful to set expectations: “That one specific opportunity will continue to play forward. That's not going to be significant revenue in 2026 or anything like that.” — Brian Mackey, President and CEO · 2026-03-03 Now there is an actual purchase order. The keyword tungsten is the single biggest new entrant in the company's language this quarter, and it isn't shared with any of the other 100+ companies in the recent reporting cohort — it's company-unique, which is rare.

Contrast that with HybridTech Armor, which has been the recurring tease for two years. Congressional funding for Navy destroyer ballistic shielding was approved "some time ago"; CPS and partner Kinetic Protection "remain hopeful that such contracts can be resolved in the coming months." That is the same phrasing pattern going back to 2025. It is a call option, not a business.

The anchor customer stopped compounding

The core Metal Matrix franchise matters most: it is, in Mackey's words, "the number one driver of our need for additional capacity." The annual contract with the largest semiconductor customer was re-signed — but the language is defensive. “The demand there remains consistent from last year,” — Brian Mackey, President and CEO · 2026-11-04 he told Chip Moore, who pegged the prior contract at roughly $15 million. Compare October 2025, when the message was "growth across the board. The large customers are ordering more." A flat anchor customer means the growth story now rests on hermetic packaging tied to AI-driven infrastructure build-out — and that is a thematic headwind at the moment. The market's AI hardware complex is in a sharp drawdown: the AI data center basket shows almost universal negative price action over the last 30 days across dozens of semis and infrastructure names. CPS's own 30% decline fits that tape, not its fundamentals.

Valuation reflects the ambivalence. At roughly 3.3x trailing revenue — more than double a year ago — the stock is not cheap for a company with flat sales. The bull case requires the Attleboro square footage, the tungsten funnel, and a hermetic-package AI tailwind to show up before the cash pile and the patience both run thin.

What changed this quarter is real: margins inflected, tungsten converted from R&D to an order, and the balance sheet buys time. What didn't change is the number that matters most — revenue, still stuck at the level it hit years ago, with the largest customer content to stay there.