They Resupplied a Moving Warship Without a Satellite Link. The Startup Has 9 Employees.

An 11-foot boat slipped into the flooded stern gate of an 844-foot amphibious assault ship underway off Hawaii. No radio link. No satellite feed. No joystick. The Typhoon USV, built by a nine-person startup barely two years old, had just resupplied a Navy warship autonomously — while the ship was moving.

That July demonstration at RIMPAC 2026, the world's largest maritime exercise, was the inflection point. Axios reported the same week that Splash Industries had also delivered parts to the carrier Theodore Roosevelt during the drill, and that the company had closed a seed round led by Ubiquity Ventures. The funding and the validation arrived together. They tell the same story: maritime autonomy's cost curve has bent, forcing traditional naval shipbuilders and maritime regulators to adapt to a future where swarming, unmanned vessels defend critical infrastructure.

Splash left stealth mode claiming its fleet had already logged 10,000 nautical miles of autonomous operation across three continents, roughly halfway around the world. The Typhoon, an 11-foot-3-inch hull that assembles in eight hours, carries a 600-nautical-mile range and sustains 50-plus knots, with 70 knots recorded in testing. The Defense Innovation Unit's SWAP-USV challenge, opened July 29, set the entry bar at 200 nautical miles and 10 knots. Splash's boat clears both thresholds by a factor of three to five.

Splash prices a Typhoon at roughly one-tenth the cost of rival systems and designs the electronics, software, and sensors in-house. Its proprietary C2 autonomy stack ran the Essex resupply without external commands — a detail that matters because the DIU solicitation explicitly warns that adversaries will jam first. A boat that finds a moving stern gate with its link cut can find a launch point with its link cut.

DIU attached prize money across three sprints, plus a separate allocation for follow-on procurement. Sprint 1 submissions closed August 10 with a 29-day turnaround from down-selection to a complete, integrated boat-and-drone system on a government test range. Splash entered with a strike module, the Airstrike Pack, already built and already exercised. Most competitors would be drawing one from scratch.

The Navy has already used cheap boats as weapons. Naval News reported July 13 that CENTCOM sent three Saronic Corsair uncrewed vessels carrying explosive packages against an Iranian submarine at Bandar Abbas — the first U.S. use of a USV in a strike role. Splash's Typhoon, by contrast, demonstrated the logistics mission: cargo delivery to a moving warship without a tether. The same hull, the company says, bolts on a sensor package for seabed mapping, a cargo pod for resupply, or the Airstrike Pack for one-way drone launch. Three mission profiles. One hull.

Avanesov has a larger hull coming, the Tempest, rated for 1,500 pounds of payload and 300 nautical miles, aimed at Navy testing. But the Typhoon is the proof article. It exists. It has run. It has resupplied a carrier strike group asset at speed, in open water, without a human in the loop. The funding buys the production line to build more of them, and the data to prove they work.

Why the Navy Needs Drone Boats Now

Splash Industries has turned autonomous patrol boats from concept into production reality. The Typhoon USV carries a configurable deck mount that swaps cargo pods, sensor suites, or, as of late August, the Airstrike Pack: a module that launches one-way attack drones from the deck at the press of a button.

Speed is the selling point for a drone launcher. A 50-knot boat carries its aircraft toward a target far faster than the aircraft would get there on its own battery, and it can leave before anyone shoots back. Small one-way attack drones typically trade range for cost, so a fast mothership widens the circle they can reach. A boat that can navigate a warship's wake without a link can also carry a drone launcher into contested water where the link is jammed — exactly where a one-way attack drone earns its keep.

Splash credits its in-house electronics and software for the module's ruggedness and cost. The company says the design gives it "a tenth of the price of our nearest competitor," though the announcement attaches no dollar figure to that claim and names no competitor. The same sentence has two blanks in it: which competitor, and at what price. The autonomy stack runs on Splash C2, the company's own command-and-control software. Operators trigger the Airstrike Pack launch from that interface. The company combines rapid hardware iteration (two complete boat builds in one month during development) with full in-house autonomy software, targeting the defense and maritime security market with small, fast, affordable patrol boats. That iteration speed is the structural difference. Traditional shipbuilders measure design cycles in years; Splash measures them in weeks.

The Navy's own drone-boat effort has had a rougher time. In August 2025, collisions, software stalls, and a paused L3Harris autonomy contract marked Navy tests off California. Ukraine settled the concept more than a year ago. The Magura V7, a 24-foot boat operated by Ukrainian military intelligence, was launching explosive quadcopters for coastal strikes alongside its air-to-air missiles by June 2025, and had shot down two Russian Su-30s. The Magura V7 proved a drone boat can serve as a mothership for aerial drones in combat. The question for Splash is not whether a robot boat can carry attack drones. It can. The question is whether an American startup can make one cheap enough to lose, and whether the Navy will buy it before it buys something bigger and slower.

The Pentagon's Drone Dominance Program is buying toward an inventory of about 340,000 small one-way attack drones by early 2028. Every one of them needs a way to get to the target. A boat that carries them there for a tenth of the cost of the alternative, if the claim holds, is a bid for a slice of that program. But the Drone Dominance Program is buying drones by the hundred thousand. Whichever drone the launcher is built around decides how useful it is. Splash has not said which drone the Airstrike Pack integrates with, or whether it is a launch rail waiting for a customer to pick one.

Important details about the pack remain undisclosed, particularly its drone capacity and the performance of the aircraft it carries. Interesting Engineering's headline calls it a swarm. Nothing Splash has published supports that word, and nothing rules it out. The system should be viewed as a demonstrated mission module rather than evidence of a specific large-scale swarm capability. Splash was not among the vendors named in DIU's January 2024 solicitation for low-cost unmanned surface vessels with swarming capabilities that could go into production the following year. The global shipbuilding industry believes the technology for fully autonomous ships is now proven, but that the sector is not yet ready to deploy it, according to Roland Berger's Autonomous Shipping Industry Survey 2026. Autonomous vessel technology is rapidly transitioning from the conceptual to practical applications as the number and scope of unmanned vessel projects increase around the globe.

Shipyards Can't Keep Up

While DIU runs sprints, the Navy's fiscal 2022 shipbuilding plan never kicked off a major fleet transformation. Defense News reported in June 2021 that the Biden administration requested eight ships, down from the Trump administration's December 2020 plan for 12, and sea-power proponents watched the unmanned-surface-vessel line items vanish. "These programs aren't ready yet," was the Navy's verdict at the time. The service had one frigate builder, Fincantieri Marinette Marine, and no second yard lined up to expand capacity. The Virginia-class submarine line was already maxed at two boats per year alongside the Columbia-class ballistic-missile program. Shipyards simply could not expand production.

That bottleneck has not disappeared. Wes Hallman, senior vice president for strategy and policy at the National Defense Industrial Association, told Defense News his biggest concern was the workforce. "You see an aging workforce and decreasing access to skilled workers," he said. "By cutting back on shipbuilding, you're actually exacerbating that problem." The danger was twofold: industry lacked direction on where to invest capital, and national security faced a delay in preparing for a peer adversary. Bryan Clark, a senior fellow at the Hudson Institute, argued the Navy could have used the FY22 request to accelerate frigates, light amphibious warships, and unmanned surface and undersea vessels, putting them in the water within two to three years. Instead, the service punted.

The Pentagon has since sent a different signal. In July 2026, WorkBoat reported the Defense Department requested the total for drones and counter-drone technologies, including the DAWG amount, up from the 2025 baseline, potentially the largest single year-over-year boost of any defense program. The global autonomous-ships market is projected to reach the projected amount by 2034, driven by Navy investment in unmanned surface vessels. The message from HavocAI co-founder and CEO Paul Lwin was blunt: "The Department of War has sent a clear demand signal to the shipbuilding industry: we need more boats, faster, with more capabilities, for less money."

Traditional prime contractors are scrambling to answer. Huntington Ingalls Industries, the nation's largest military shipbuilder, made the Navy's new Medium Unmanned Surface Vessel (MUSV) shortlist alongside Leidos, Saronic Technologies, and Sea Machines, but Anduril Industries, which had invested tens of millions revamping the former Foss Shipyard in Seattle for the canceled Modular Attack Surface Craft (MASC) program, was left off. Anduril's Seattle shipyard footprint was "always small, a handful of employees," a spokesperson told GeekWire in August 2026. The company still employs 560 people across Bellevue and downtown Seattle offices, up from roughly 30 four years ago, and operates a multi-acre test facility in Carnation, Washington. It could grow to 1,000. The MASC cancellation and MUSV pivot illustrate how fast the goalposts move.

The Navy's new MUSV strategy replaces single-design competitions with a "marketplace" of competing prototypes. Companies that successfully deploy vessels at sea receive the award amount and become eligible for follow-on production. At-sea testing began in June 2026 and concluded in October. This approach aims to speed innovation and allow various vendors to produce ready-to-use systems on tight timelines; initial production vessels are targeted for fiscal 2027, an unusually fast schedule for a new naval platform class.

Partnerships are the bridge. Anduril allied with South Korea's HD Hyundai Heavy Industries and Edison Chouest Offshore to build a dual-use autonomous surface vessel class. Saildrone, expanding beyond intelligence-surveillance-reconnaissance with its 52-meter Spectre platform for anti-submarine warfare and strike, partnered with Lockheed Martin and Fincantieri Marinette Marine for the higher end of the MUSV market. Hanwha Defense USA, the only shipbuilder with an operational U.S. yard to joint-venture with an autonomy firm, teamed with HavocAI to jointly develop 200-foot autonomous surface vessels, with Hanwha Philly Shipyard eyed for first production. HavocAI had closed a round and confirmed dozens of vessels sold to the Department of Defense. Hanwha Defense USA CEO Michael Coulter called it "forging a partnership between an allied defense company with advanced manufacturing scale... with a software-first defense technology company."

The old model — expensive, highly specialized, slow to field, difficult to upgrade, dependent on decade-long acquisition cycles — aligned with an era of predictable threats. Modern conflicts evolve in the blink of an eye. Software updates, sensor requirements, communications needs, and mission profiles can all change in real time.

That WorkBoat assessment from July 2026 captures the core tension. Traditional naval systems reflect decades of engineering excellence but carry structural constraints. Adaptability has become a primary performance metric. Modular open systems architecture (MOSA) now sits at the center of naval strategy: autonomous platforms must integrate sensors, AI software, communications, aerial assets, and future payloads without complete redesign. Interoperability across air, land, sea, space, and cyber domains is a primary concern — isolated platforms cannot create decisive advantage in a distributed, data-dependent battlespace.

Manufacturing speed shapes battlefield outcomes. Advanced methods (additive manufacturing, modular assembly, digital engineering) enable faster iteration, reduced cost, and scalable deployment. Domestic manufacturing resilience has moved to the center of policy discussions; supply-chain independence, production capacity, and workforce readiness now influence national security outcomes as directly as platform performance. Attritable systems (lower-cost, designed for repeated deployment at scale) expand operational options in contested zones without placing the same financial or human risk on each mission.

The evaluation criteria for maritime capability are changing in real time. The question is no longer whether autonomous USVs belong in the future fleet. It is whether they can be made modular, interoperable, attritable, and scalable quickly enough to meet the pace of modern conflict, and whether the traditional industrial base can retool fast enough to stay relevant.

The Rules Haven't Caught Up

The Coast Guard has not written a new regulation for autonomous systems since 1988. That gap — three and a half decades — frames every conversation about where a Typhoon USV can legally operate today. Instead of new rules, the service applies the same regulatory framework built for crewed ships, leaning on "equivalents" provisions that let operators propose alternative designs if they can prove equivalent safety. The mechanism exists. The precedent does not. Coast Guard officials have said they need sufficient data and examples to evaluate autonomous technologies, and they have not seen enough. Companies guard proprietary design standards. Statutory minimum crew requirements add cost that discourages testing. The result is a loop: no regulation without data, no data without regulatory clarity.

The clearest statutory barrier is the crew requirement itself. Federal law prescribes minimum officer counts based on gross tonnage and mandates a credentialed master aboard every vessel. The Coast Guard has limited authority to reduce those numbers. Officials have acknowledged that any statute written on the assumption of human presence becomes a problem as autonomy removes humans from the bridge. A pilot program for autonomous at-sea rocket recovery (congressionally directed and still early) gives the service waiver authority for certain crew requirements, but it is narrow. Policy Letter 22-01 (Change 1) offers guidance to Captains of the Port for human-supervised testing of remote-controlled and autonomous systems. As of June 2026, local Captains of the Port had fielded 48 requests involving autonomous ship technology. They managed those operations and associated risks at the port level. Industry stakeholders have raised concerns about consistency across ports.

Collision regulations present a different friction. The International Regulations for Preventing Collisions at Sea require every vessel to "maintain a proper look-out by sight and hearing as well as by all available means appropriate." The IMO has determined that autonomous ships must comply with the look-out rule and that the rule itself does not need amendment. Each flag state must decide how compliance looks. Coast Guard officials said that how autonomous technologies satisfy the look-out requirement domestically has not been established. The Norwegian Maritime Authority allowed one operator to cut crew from five to three after successful tests and no safety incidents. The UK took a broader step: in December 2023 it modified regulations for small commercial ships to permit remote operation under 24 meters and exempted certain logbook requirements. In August 2024 the UK Maritime and Coastguard Agency issued a general exemption letting autonomous ships under 2.5 meters operate without agency certification. Canada's Transport Canada can grant exemptions from onboard lookout requirements for highly autonomous small ships. The U.S. has not matched those moves.

The IMO timeline adds pressure. The organization expects a non-mandatory autonomous ship framework in 2026, a mandatory framework adopted in 2030, effective 2032, though the then-chair of the Maritime Safety Committee warned the mandatory date is uncertain given complex issues. The Coast Guard leads the U.S. delegation and has driven work on crew competence, collision prevention, and the electrical, engine machinery, and lifesaving appliances sections of the coming framework. Officials have said they will not amend domestic regulations or issue guidance ahead of the IMO framework because harmonization will be more effective once the international baseline exists. That means the current patchwork (equivalents, exemptions, port-level discretion) persists for years.

Insurance is tracking the same uncertainty. Marine insurers are still pricing the risk for autonomous vessels. Cybersecurity compounds it. GAO has identified cybersecurity as a government-wide high-risk area for more than 25 years. As of June 2026, 37 federal agencies had issued 117 cybersecurity regulations across nine critical infrastructure sectors. Transportation systems alone faced 22 regulations, 11 with incident reporting requirements, seven requiring cybersecurity plans, and four mandating reviews or audits. Duplicative reporting burdens are documented. For an autonomous vessel operator, the compliance surface spans Coast Guard safety rules, IMO cyber guidelines, CISA incident reporting, and sector-specific mandates, each with different definitions, thresholds, and timelines.

The regulatory picture is not static. The Coast Guard's AutoPoCo, chartered in June 2021, meets regularly to develop guidance for Captains of the Port, identify legal gaps, and engage industry. The service briefed congressional staff in February 2024 on statutory barriers and pilot progress, with quarterly updates planned. Officials maintain they can meet their safety mission with current authorities. They also say they will develop policy guidance and design standards when the technology reaches a critical mass, conditions not yet met. For a startup building boats at development velocity, the message is clear: the water is open, but the rules are being written in real time, and the cost of guessing wrong is a vessel that cannot legally leave the dock.

Nine People, Two Boats in a Month

Splash Industries runs on nine people. That headcount, listed on the company's Y Combinator profile (W25 batch), belies the breadth of what the El Segundo shop has already shipped: two complete patrol-boat iterations in a single month during development, a hundred-plus autonomous miles in San Francisco Bay, and a RIMPAC 2026 demonstration where a Typhoon USV resupplied a Navy warship without a crew aboard. The investment is meant to turn that velocity into production volume, and the open engineering and product roles are where the next bottleneck lives.

The founders' résumés set the hiring bar. Ivan Avanesov spent six years at UCLA's RoMeLa, where he worked on Artemis, the humanoid that won RoboCup 2024, and before that founded Vosentech, which designed and manufactured more than 10,000 airflow-test generators for customers including SpaceX, Apple, NASA, and Los Alamos National Laboratory. Marcell Veszpremi came from Verkada, where he wrote embedded software and battery-management firmware for enterprise security alarms, and from the Ozcan Research Lab, where his work on low-cost, point-of-care medical devices appeared in multiple publications. Together they spent a decade building a hardware company that shipped tens of thousands of units across dozens of SKUs. They hire for that same fluency across mechanical, electrical, and firmware domains — people who can spec a hull layup in the morning and debug a CAN bus in the afternoon.

The public robotics-software-engineer posting makes the stack explicit: integrate heterogeneous sensor data (lidar, radar, optical, AIS) for localization and perception; optimize autonomy pipelines for low-latency execution on edge compute; build deployment infrastructure that keeps vehicles running continuously in salt spray and intermittent connectivity; scale collaborative autonomy from today's small coordinated teams to large, heterogeneous fleets. That is sensor fusion, real-time Linux, ROS 2, and containerized CI/CD — all on a platform that does 70 mph in a ten-foot jet boat. The job description's closing line — "What could possibly be more fun than building a 10ft-long AI-powered jet-boat that goes 70mph? Well… possibly a 100ft one" — is half recruiting pitch, half roadmap.

Operational roles sit beside the autonomy stack. The Coast Guard still treats these ASVs as vessels, and commercial marine liability insurance is required for operations. The person who drives the chase boat during autonomy trials needs to know the rules of the road and the limits of the autonomy cold.

The interdisciplinary profile is the filter. A typical defense prime would split naval architecture, autonomy, embedded firmware, and test operations across separate directorates with months of requirements reviews between them. Splash's rapid iteration cadence forces every engineer to touch salt water. The modular payload bay (designed for side-scan sonar, deployable UAVs, or the Airstrike Pack's one-way attack drones) means mechanical interfaces and software APIs change together. When the team sank a prototype and rebuilt it in weeks, the same people handled the mechanical, electrical, and software rework. Hiring for that blend is the war the founders cite. Big Tech competes for pure software talent; Splash competes on mission, iteration speed, and the rare chance to own a full stack from carbon fiber to fleet autonomy. The nine-person team today is the seed of the "Navy of the future" phrasing on their YC page — not a metaphor, a staffing plan.

What This Story Leaves Out

This piece follows the surface-USV thread: a nine-person team, a seed, a Typhoon that resupplied a warship and now launches drones, and the regulatory thicket the Coast Guard is still writing. The subsea, space, and energy stories are real, and they're the next ones.

DIU's first production awards target fiscal 2027. By then, the stern gate that opened for a nine-person startup's boat at RIMPAC will be a routine berth if the shipyards and the rulebooks can move as fast as the boats.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.