Hadrian builds automated precision-manufacturing capacity for aerospace and defense customers. In October 2026, Axios reported that the company would install production cells inside Lockheed Martin's expanded Grand Prairie, Texas plant, with Lockheed employees overseeing them. That is more consequential than another faster machine. The factory builder is moving inside the customer's boundary while leaving ownership and operating roles deliberately split.
Capacity is the deliverable
Founder Chris Power originally described the factory itself as the product. TechCrunch reported in 2024 that Hadrian automated programming, scheduling, paperwork and inspection around CNC machining, while leaving harder work to people. Power said one person could run four to six machines and reported more than US$20 million in revenue after twelve months of customer operations. Those are attributed company claims.
Hadrian's current site now offers three arrangements: parts from its own factories, a dedicated facility built around sustained customer volume, or lines deployed in a customer-owned plant. The Lockheed arrangement makes the third model concrete. Axios says the cells will contribute to Patriot, GMLRS and Precision Strike Missile work inside more than 100,000 square feet of expansion.
The deployment avoids a simplistic buy-versus-outsource choice. Lockheed retains employees and the host plant. Hadrian supplies production capability and a broader system. The commercial question becomes how much capacity is committed, how acceptance is measured, and who absorbs idle time or rework.
These records mix company statements, founder recollections and independent reporting. Company numbers are attributed rather than presented as audited results. The mechanism is useful because it identifies a decision to test, not because one visible company guarantees another's outcome.
Utilization can make a fast cell expensive
Take an illustrative cell costing US$4 million a year across equipment, software, support and site overhead. At 6,000 accepted machine-hours, capacity costs about US$667 per accepted hour. At 3,000 hours it costs US$1,333. A claimed cycle-time improvement does not rescue poor loading. If a part needs two machine-hours and US$300 of material and direct handling, its simplified cost is US$1,634 at the higher utilization and US$2,966 at the lower one.
This arithmetic is illustrative. Its invented inputs are not the subject company's costs, margins or forecast. Replace every input with a dated quote, invoice or observed time before using the result for a commitment.
A decision rule worth carrying forward
Before agreeing to dedicated capacity, define the denominator as accepted output hours, not scheduled hours. Put changeovers, inspection failures, maintenance and customer-caused waiting into named buckets, then assign who pays for each.
The rule forces attention onto the denominator behind the attractive headline. Orders are not accepted deliveries. A technical specification is not a validated workflow. Enterprise value is not founder proceeds. Traffic is not contribution. Naming the conversion between those states is the beginning of useful operating evidence.
Build an evidence ladder before scaling
Start with the weakest evidence and refuse to give it a stronger name. A comment, waiting-list entry or sample request shows attention. A deposit or paid order shows willingness to buy under stated terms. An accepted delivery shows that the operation fulfilled the promise. A reorder, referral or renewed contract shows that value lasted beyond novelty. For Hadrian and the shift from selling machined parts to deploying manufacturing capacity inside a customer's plant, those stages should be recorded separately because each one authorizes a different size of commitment.
Next, attach time to every observation. A strong launch week can hide a slow replenishment cycle, seasonal use or a queue created by publicity. Record the date of purchase, date cash becomes available, date production begins, date the customer accepts the result and date any return or support cost appears. A contribution estimate assembled from events in different months can look healthy while the bank balance is under strain.
Then give variance the same attention as the average. One fast machine, cooperative bakery, cheap parcel or enthusiastic interview does not describe a repeatable system. Keep the best, median and worst observed case. Investigate the worst before buying capacity for the average. If the downside is a safety failure, unusable batch or loss of an essential customer, an average result is the wrong decision tool.
Finally, write who owns every handoff. The supplier may finish a component while the founder still owns inspection. A carrier may deliver the box while the business still owns setup. A retailer may place an order while the manufacturer still owns sell-through risk through markdowns or returns. A buyer may acquire the company while the founder remains exposed through deferred consideration. Ownership of the next action is part of the cost.
This ladder prevents a common analytical shortcut: using one vivid public milestone to fill several private cells. It is acceptable for cells to remain blank. In fact, a blank can be the most decision-useful result because it identifies the next small test. The goal is not to make the story look complete. It is to stop the next commitment from depending on an unsupported conversion.
Keep the first version deliberately small. One batch, one customer segment, one configuration or one channel makes a failure interpretable. Changing product, price, audience and fulfillment at the same time may produce a better headline, but it destroys the comparison. Record the baseline before intervention, change one important variable, and choose the review date in advance. If the result misses the threshold, shrink or stop the commitment without rewriting the threshold afterward. A disciplined no is useful evidence and preserves cash for a better test.
Run the one-hour version
Choose one constrained manufacturing step and draw its boundary from released drawing to accepted part. Mark every queue, approval and handoff. Ask whether the proposed automation removes the largest interval or merely speeds the visible cutting step.
- Write the next irreversible commitment.
- Choose one observed denominator connected to cash or accepted output.
- Separate company claims, independent evidence and your assumptions.
- Set the result that would shrink or stop the commitment.
- Keep the record and compare the forecast with the eventual outcome.
Two adjacent Builtplain cases offer useful contrast: Opentrons adoption brief and Sublime Cement adoption system. They use different products but apply the same discipline of counting the work around the visible milestone.
What the public record cannot answer
Public sources do not reveal complete private margins, cohort retention, contracts, ownership waterfalls or defect ledgers. No inference here fills those gaps. Survivorship bias also applies: visible companies leave better records than similar attempts that failed. Use the case to design a smaller test, not to claim a guaranteed result.
Sources checked on November 17, 2026
- Current company description of owned factories, dedicated facilities and customer-site deployment
- Founder testimony describing automation and claimed performance, treated as self-report
- Independent October 2026 report on the Lockheed deployment and its staffing model
- Earlier reported operating model, revenue self-report and machine staffing claims
Putting a factory inside another factory makes the promise tangible. It also removes the excuse that the customer's queues and the supplier's queues can be measured separately.
Discussion
Sign in with Google or just a name. No email link, no password to remember.