The company building underwater drones for $50,000
Summary
Ulysses’ base Mako can sell for as little as $50,000 to navies, versus roughly $5 million for the common U.S. Navy autonomous underwater vehicle; the work it targets is also often supported by ships costing a few hundred thousand dollars a day. Will O’Brien’s thesis is that lower-cost, persistent robotics can move ocean operations from occasional expeditions to infrastructure, unlocking new conservation, energy, and defense markets. “We need to be building millions of underwater drones.”
The system pairs a 2,000-nautical-mile autonomous surface craft, Leviathan, with modular Mako underwater vehicles. Leviathan transports, recovers, wirelessly charges, and connects the drones to Starlink; Mako’s modular battery packs, propulsion, stereo cameras, sonar, and robotic-manipulation capabilities adapt one platform to multiple jobs. O’Brien calls it dropping “the kids off to school.”
Underwater communications are the central technical constraint, forcing intelligence onto the vehicle. Acoustic links deliver only about 10 kilobytes per second over five kilometers, so Mako carries roughly 100 times more compute than other autonomous underwater vehicles, including a full NVIDIA 5090 and Intel CPU. Missions currently use relatively rudimentary local autonomy, such as generating a mapping route and avoiding rocks, but the architecture anticipates more complex behavior.
Manufacturing capacity, not customer demand, is now Ulysses’ limiting factor. The company is declining contracts, plans a factory next year that could rank among the world’s largest underwater-drone facilities by unit count, and hopes to produce several hundred vehicles—maybe 1,000—while one supply-chain supplier makes only 100–200 of a component annually. O’Brien is progressively bringing sonar and inertial navigation in-house because existing maritime supply chains cannot support his intended scale, lead time, and cost.
Seagrass restoration supplied the initial wedge and the reusable technical stack. Seagrass covers 2% of the seafloor yet holds 20% of ocean carbon, and about 20% of commercially important fish stocks rely on it. O’Brien links declining fish stocks to overfishing and habitat loss, while restoration occurs 100–1,000 times more slowly than ecosystems are lost. Restoration entails mapping habitat, modeling suitable sites, collecting seeds, and depositing them; Ulysses can then reuse the sensing, navigation, autonomy, and surface-vehicle platform elsewhere.
Early commercial traction spans mine countermeasures and offshore-wind maintenance. Ulysses won its first U.S. Navy and offshore-wind contracts in its second year; for floating wind, its robots can inspect power cables daily, assess structural integrity, detect discarded fishing lines, and eventually cut entangling lines away. The intended autonomous alternative would reduce dependence on a crewed, weather-constrained ship operation and support more continuous monitoring.
John Collison challenges Ulysses’ dependence on slow government and energy buyers, but O’Brien and Chris Sacca see procurement friction becoming a moat after award. Clearances, master service agreements, and contracting vehicles delay revenue, yet replacement is equally slow, making successful deployments “very sticky” and, in Sacca’s words, “very reliable and very bankable.” The larger strategic opportunity is persistent monitoring of fragile undersea cables and critical waterways rather than today’s episodic patrols.
Deep dive
1. Guinness gave O’Brien a model for industrial innovation
O’Brien’s favorite example is Guinness’ in-house R&D operation, which began as early as the 1890s, hired smart people, and effectively told them, “Come in here and find a problem to work on.” John Collison’s translation: Guinness had its own version of Xerox PARC.
William Sealy Gosset developed small-sample statistics because Guinness needed to estimate hop quality without testing hundreds of samples and losing a few percentage points on every batch. He published under the pseudonym “Student,” O’Brien says, because Guinness did not want Anheuser-Busch or Heineken learning about its research advantage.
Project ACORN—originally Project Dynamite until shipping parcels bearing that label between London and Dublin proved awkward—created the canned-Guinness widget as drinkers shifted toward consuming at home. O’Brien’s point is that a familiar pint conceals decades of technical, manufacturing, and marketing invention.
O’Brien also admires Guinness’ bar-quality system, the 7 Cs: crafted presentation, consistent freshness, counter visibility, correct gas, correct temperature, clean lines, and clean glassware.
That appreciation came from an entrepreneurial Irish family where “business was the talk of the table.” Guinness showed him that consequential technology need not look like software or arrive from Silicon Valley; a homegrown, high-volume business could treat research, process control, and brand as one operating system.
2. Seagrass was the wedge into a much larger ocean platform
Ulysses began with the observation that autonomous systems would perform an increasing share of critical work. After comparing air, land, and sea—and then surface with undersea—the founders saw underwater jobs that were either not being performed, not performed at scale, or too expensive to attempt.
The first use case emerged when a cofounder met a marine biologist manually taking photographs, collecting water samples, and planting seeds. Having worked on aerial delivery drones at Manna, he asked why an underwater equivalent could not map ecosystems, collect seagrass seeds, and restore damaged habitat.
O’Brien calls seagrass “your rapper’s favorite rapper,” overshadowed by its “very good-looking friend,” coral. As stated in the episode, it covers 2% of the seafloor but holds 20% of ocean carbon, and roughly 20% of commercially important fish stocks depend on it. O’Brien attributes declining fish stocks to overfishing and habitat loss, while ecosystems are being restored 100–1,000 times more slowly than they are lost.
Restoration starts by mapping existing beds, then applying a habitat-suitability model using depth, water clarity, and other conditions. Ulysses identifies flowering plants, collects their seeds, and deposits them in suitable areas—a technical workflow it can now repeat without substantial new R&D, leaving field operations as the main scaling constraint.
3. Lowering access costs could turn the ocean into an economic frontier
O’Brien describes himself as a “pro-growth environmentalist” and “eco-pragmatist”: a flourishing future requires both a consistently growing economy and a biosphere capable of supporting it. His childhood beside the sea supplied the emotional pull; the lack of knowledge about most of the ocean supplied the frontier. He says more of Mars’ surface has been mapped than the depths of the ocean, that more than 75% of Earth is effectively unseen, and that fewer than 100 people think about this work daily.
His framework begins with expensive expeditions justified by one high-value use case: pelts in the American West or defense satellites in space. Railroads and reusable launch then lower access costs, increase persistence, and enable uses that were impossible to justify during the expedition phase.
Collison makes the SpaceX analogy explicit: cheaper access to orbit enabled businesses such as Starlink. O’Brien’s ocean equivalent is persistent robotic presence—systems that remain available, gather information, and perform work routinely instead of requiring a fresh ship campaign for every task.
Conservation still encounters “green tape.” O’Brien cites a potentially world-leading Australian seagrass-restoration project in an endangered location that Ulysses has pursued for a long time, only to find permitting the “single biggest barrier.”
4. Leviathan and Mako separate efficient transit from underwater work
Full XYZ access requires two machines because surface transit has less drag. The 2,000-nautical-mile Leviathan carries Mako to the worksite, launches it, calls it home after the mission, and recovers it—the surface craft “drops the kids off to school.”
Existing underwater vehicles were designed primarily to gather data, not physically manipulate their environment. Mako therefore uses a general-purpose, Lego-like architecture: add battery packs for another 150 nautical miles, change the rear-thrust module for speed, or fit stereo cameras, sonar, and task-specific implements.
Those early architectural decisions let Ulysses respond when the U.S. Navy or offshore-energy companies arrived with differently shaped problems. The common platform is the vehicle, sensing stack, navigation, autonomy, and surface support; customer-specific capability comes through modules rather than a new robot.
5. Ocean physics makes local autonomy and surface support mandatory
Corrosion can be mitigated through coatings and anodizing, while pressure defines the operating envelope. Ulysses currently operates to roughly 1,500 meters and has been crush-tested to 2,200; O’Brien calls 6,000 meters “full ocean depth,” although the deepest ocean reaches about 10,000 meters.
Communications are the harder constraint. Acoustic transmission resembles “a dial-up modem,” delivering about 10 kilobytes per second across five kilometers, so operators cannot continuously stream perception and decisions between the vehicle and shore.
Mako consequently runs the loop locally, carrying around 100 times more onboard compute than other autonomous underwater vehicles, including a full NVIDIA 5090 and Intel CPU. Present autonomy remains “relatively rudimentary”: an operator draws a polygon, the system generates a lawnmower-style route, and the vehicle navigates over or around obstacles.
Leviathan provides the missing infrastructure: wireless charging, Starlink connectivity, and live diagnostics after recovery. O’Brien’s longer-term design adds replacement modules and “vending machine style” repair capability, reducing dependence on a distant technician or port.
6. Scaling demands consumer-style supply chains and vertical integration
About 20 million aerial drones are built annually versus perhaps 1,000–2,000 underwater drones, O’Brien says. To reach millions, Ulysses must build on electric-vehicle, drone, and consumer-electronics supply chains that already operate at volume—not traditional maritime suppliers.
The first vehicle was roughly 80% off-the-shelf; Ulysses is now on generation five and expects to bring much more in-house over the coming year. Software was built almost entirely from scratch because the modular vehicle differs from conventional underwater robots.
Sonar and inertial navigation are next. Ulysses currently buys both, but is hiring an acoustics team and has internal navigation prototypes intended to replace existing suppliers. O’Brien says the initial navigation system is intended to cost a fraction of the expensive, long-lead-time supplier’s system while using a supply chain capable of scaling.
Collison presses that building an inertial-navigation system is unusually difficult. O’Brien declines to reveal the approach—mentioning fiber-optic gyros and MEMS only as possible methods—but says scale and lead time justify the work. One supply-chain supplier currently makes only 100–200 of a component per year.
Ulysses’ immediate problem is “building things,” not selling them: it is turning contracts away and plans a facility next year that could be among the world’s largest underwater-drone factories by unit count. The near-term ambition is several hundred vehicles, perhaps 1,000.
7. Navy and offshore wind extend the shared-use platform
In its second year, Ulysses secured its first U.S. Navy contract for mine countermeasures and its first offshore-wind contract. O’Brien jokes that it was “finding mines and hunting mines before it was cool,” but the signal is that one architecture crossed conservation, defense, and energy.
Floating offshore wind creates exposed power cables that can snag abandoned fishing nets and lines, potentially entangling dolphins and whales. Ulysses is building a surface-underwater system to map those cables daily, inspect structural integrity, detect fishing lines and entanglements, and cut fishing line away.
The incumbent workflow begins with a ship costing a few hundred thousand dollars per day, a tethered remotely operated vehicle, and rotating crews needed for 24-hour operation. Headcount expands into operators, managers, captain, mate, cook, and sometimes a doctor; weather and seasickness further constrain working conditions.
Mako’s base price can be as low as $50,000 when sold to navies, against roughly $5 million for the most common autonomous underwater vehicle the U.S. Navy buys today. Advanced sensing and navigation narrow the gap, but O’Brien still expects about an order-of-magnitude advantage; Leviathan costs less than $1 million and more than perhaps $500,000.
Ulysses generally sells vehicles to navies but provides offshore customers a service. The intended autonomous alternative would remove much of the labor burden, permit operation in worse conditions, and let customers substitute persistent capability for sporadic ship charters.
8. Slow procurement can become a moat around persistent infrastructure
Collison’s pushback is that “bad buyers kill companies”: governments and large energy firms impose clearances, safety reviews, master service agreements, and multiyear procurement cycles, potentially exhausting a startup before revenue arrives.
O’Brien accepts the delays but reverses the conclusion. Venture-backed companies can fund the R&D and waiting period; once a contract is won, they become difficult to displace because customers do not want to repeat a two-year qualification process. Sacca says that once companies reach the contracting phase, the work can become “very reliable and very bankable.”
The cultural motivation is broader than procurement. O’Brien contrasts “make something people want” with a newer hardware-founder impulse to “make something the world needs,” linking physical-world companies, faith, and post-COVID dissatisfaction only tentatively: “perhaps” founders forced deeper into digital life now want tangible impact.
Ulysses’ sharpest strategic case is undersea infrastructure. O’Brien says there are about 200 undersea cables today and identifies Ireland as a potential choke point because cables running from the U.S. to Europe largely cross there. He says cables are typically damaged by accidental anchor draggings, though some incidents are not accidental. Sacca separately warns that cables can also be snooped or tapped.
O’Brien’s prescription is persistent, 24/7 robotic monitoring around cables and critical waterways because manual patrols are prohibitively expensive. Unlike defense companies focused on making vehicles “go boom,” he wants to solve the harder general-purpose robotics problem. His stated endgame is to replace episodic ocean campaigns with permanent robotic presence—and help expand the ocean from a roughly $3 trillion economy into “a $10 trillion-plus place for commerce.”