DARPA Wants to Print Concrete on the Seafloor Out of Mud and Seawater. Hoboken's $3M Direct-to-Phase-II SBIR Closes September 23.

There is a particular kind of DARPA topic that looks like science fiction in the headline and reads like a procurement document in the body. Hoboken is one of them.

The headline version: DARPA wants robots that 3D-print structural concrete on the seafloor, mixing the mud already down there with the seawater already around them, hauling almost no cement to the job site. The procurement version: DARPA has already funded the chemistry, watched it work in a tank at Cornell, and is now buying the four hardware and software subsystems that stand between a laboratory arch and a machine that can repair a pier.

DPA26BZ05-DV018 — Hoboken opened August 26, 2026 as part of DARPA's FY26 SBIR Release 5 and closes September 23, 2026 at 12:00 PM ET. At up to $3,000,000 over 18 months, it is the largest single award in the release — double the Semantically-Aware ISR topic, and more than the $1.5 million heavy-lift UAS topic that ran alongside it.

It is also Direct-to-Phase-II only. There is no Phase I. There will not be one.

What Trenton already proved, and why that matters to your proposal

You cannot read Hoboken correctly without reading Trenton first, because Trenton is the reason Hoboken is a Direct-to-Phase-II topic rather than a feasibility study.

Trenton was DARPA's program to answer a single question: is underwater 3D concrete printing physically possible using in-situ materials? The technical obstacle was never robotics. It was chemistry. Ordinary concrete placed underwater washes out — the cement paste disperses into the water column before it can set, and what remains is a weak, contaminated mess. The conventional workaround is anti-washout admixtures and enormous binder fractions, which defeats the entire point of an expeditionary system: if you have to ship the cement, you have not saved anything.

Trenton's answer was to use the seafloor itself as the aggregate and the fines, seawater as the mixing water, and hold binder content below 20 percent while still achieving self-supporting, printable strength. That is an aggressive number. Conventional structural concrete runs meaningfully higher on binder, and every point you shave off is a point of cement you do not have to carry to a contested shoreline.

In September 2025, a Cornell University team demonstrated progress toward DARPA's sediment target for visiting agency officials. The technical move that made it work is worth understanding, because it constrains the hardware Hoboken is now buying: rather than blending the accelerating admixture into the base material, the team injected the chemical directly at the printer nozzle. The mix stayed pumpable all the way through the delivery line, then hardened rapidly the instant it was deposited. Lab tests in a large water tank characterized arch formation — strength, geometry, surface texture — under continuous water exposure.

That two-stage, inject-at-the-nozzle architecture is not a footnote. It is why one of Hoboken's four tracks is specifically about in-line mixing at the nozzle with real-time quality monitoring. DARPA is not asking the field to invent an approach. It watched one work and is now paying to industrialize it.

The four tracks are a supplier list, not a menu

Hoboken splits into four subsystem tracks. Read them as the bill of materials for a machine DARPA intends to integrate later.

Two requirements cut across every track and disqualify a surprising number of otherwise-credible bidders:

All mixes must use seawater. Not simulated seawater in a beaker as an afterthought — seawater as the design assumption. Chloride content, ionic strength, and variable salinity are inputs to the chemistry, not contaminants to be engineered around.

Subsystems must be designed for future interoperability. DARPA is explicit that these four tracks will eventually have to bolt together. A Track 2 pumping system that only works with your own proprietary Track 1 printer is a weaker proposal than one with a documented, open mechanical and data interface — even if yours pumps better.

That interoperability clause is the tell. DARPA is running a supplier-qualification exercise. Four vendors get funded on four subsystems, and Phase III integrates them into one system for dual-use military and commercial deployment. If you propose as though you are building the whole machine yourself, you have misread the topic.

The entry bar: prior feasibility, and DARPA means it

Direct-to-Phase-II topics carry a statutory requirement that proposers demonstrate the Phase I feasibility work was already accomplished — through other funding, internal R&D, or prior contracts. Hoboken sharpens the general rule into something specific and unforgiving:

Proposers must demonstrate prior experience with low-binder sediment concrete mixes.

Not additive manufacturing generally. Not marine robotics generally. Low-binder sediment concrete, specifically.

This narrows the credible bidder pool dramatically, and it narrows it in a direction that is worth naming plainly: the people who can clear this bar are largely people who were already funded on or adjacent to Trenton, plus a small set of academic labs and marine-construction firms working the same chemistry independently. If your company's relevant experience is "we build ROVs" or "we do concrete 3D printing on land," Track 1 is probably not winnable in four weeks — but Track 4 might be, because the AI mix-optimization track rewards data science and formulation expertise more than it rewards a wet-testing history.

Track 4 is the underpriced track. It is the only one of the four where the deliverable is a database and a model rather than steel in the water, and it is the only one where a team without a hyperbaric test facility can compete on equal footing.

The Phase II structure is four gates in eighteen months

Hoboken's 18-month base period is organized in four stages, and the sequencing tells you where the risk is priced:

1. Feasibility and requirements analysis
2. System design and detailed specifications
3. Component fabrication and testing
4. Prototype demonstration and transition planning

Eighteen months for design, build, and wet demonstration of marinized hardware is not generous. Teams that arrive with an existing platform to marinize will be at Stage 3 while teams starting from a clean sheet are still writing Stage 2 specifications. Build your schedule against that reality rather than against the nominal stage boundaries.

Note also what Stage 4 asks for: prototype demonstration tied to real-world use cases. DARPA names port repair, cable protection, and artificial reefs. This is the dual-use hook, and it is where most defense-only proposals lose points. The commercial market for underwater concrete repair — piers, seawalls, bridge piles, offshore wind monopile scour protection, submarine cable armoring — is large, aging, and currently served by divers and precast units barged to site. A proposal that quantifies that market and names a commercial transition partner is materially stronger than one that treats reef-building as a rhetorical flourish.

Practical mechanics before September 23

• Submission portal: DSIP at dodsbirsttr.mil. If your SBC registration, SAM.gov entity registration, or SBIR Company Registry entry is stale, fix it this week — DSIP will not let you submit through an expired registration, and there is no grace period.
• Technical questions close September 16. DARPA stops answering inquiries in the final seven calendar days. Any ambiguity about track boundaries or the interoperability interface needs to be asked well before that.
• ITAR and EAR restricted. Foreign nationals on the technical team require documented compliance. Budget the time.
• VC-backed companies are eligible provided they are properly registered with SBA under the majority-VC-ownership provisions. That registration is not instantaneous.
• No Phase I option exists. If you cannot document prior feasibility, there is no smaller on-ramp to bid instead. Wait for a future release.

The strategic read

DARPA's monthly SBIR release cadence — topics pre-released the first Wednesday, opening three weeks later, closing four weeks after that — has turned what used to be a scramble into a plannable pipeline. Hoboken is a good illustration of what that cadence rewards.

The companies that will win DV018 did not start working on it on August 5. They have been running sediment mixes since Trenton made the approach public, and the pre-release period functioned as a starting gun for proposal writing, not for research. That is the structural lesson for anyone watching this space: on a Direct-to-Phase-II topic, the technical work has to precede the solicitation by a year or more. What the four-week window actually buys you is time to write.

For the 2027 releases, the actionable move is to identify which DARPA programs are currently in their Trenton phase — proving a chemistry or a physics — and start building the feasibility record now. By the time the Hoboken-equivalent posts, the bidder list is already effectively closed.

Underwater 3D concrete printing is going to be a real capability. Whether it is yours depends on work you did before this NOFO existed.