I keep coming back to a number that does not behave like a navy budget. Roughly 1,320 small attack craft, none of them carrying a crew, priced in the neighborhood of NT$29.6 billion, or about $930 million. That is not pocket change. It is also not the cost of a single modern destroyer program once you count sensors, missiles, and a decade of upkeep. If you have ever watched a smaller force try to hold a narrow stretch of water against a much larger one, you already know why that arithmetic is getting attention.
Taiwan is moving fast toward a fleet of unmanned vessels meant to make a blockade, or a crossing of the strait, messier and more expensive. The idea is not to outbuild a continental shipbuilding machine ship for ship. It is to flood the problem with cheaper platforms that watch, point, protect cables and pipelines, and, in some cases, carry weapons. I have found that planners reach for this logic only after the conventional options start looking mathematically hopeless. This looks like one of those moments.
Why a Cheap Fleet Can Still Change the Math
The strait is not an open ocean problem. It is a congested, politically radioactive corridor where commercial traffic, fishing boats, coast guard cutters, and warships already share the same weather. A force that tries to seal that corridor has to sort friend from decoy, keep its own logistics moving, and stay ready for something sudden under the surface. Unmanned boats do not solve that puzzle by themselves. They do change who pays for every hour of sorting.
Think of it as a toll booth that never quite closes. A large navy can still push through. It just has to spend missiles, attention, and time on targets that cost a fraction of the weapon used to kill them. Ukraine’s Black Sea fight did not invent that idea, but it made the lesson hard to ignore. Unmanned systems, used with patience, can impose real cost on a conventionally stronger fleet. Perhaps the most interesting part is not the spectacle. It is the accounting.
Taiwan’s navy is weighing that 1,320-boat figure for the water. The coast guard is chasing unmanned hulls of its own. Put those tracks together and you get something closer to a market than a one-off purchase. Shipyards, electronics firms, and software teams all see a buyer who might stay around. That is a different political animal from a single foreign arms package that arrives, gets photographed, and then sits in a hangar.
What the Money Is Actually Buying
The headline number is easy to misread. Nine hundred thirty million dollars sounds like a fleet. In practice it is a down payment on hulls, integration, and the unglamorous work of teaching boats to survive without a person on the console. A useful sea drone is not a speedboat with a camera taped to the bow. It is a platform that has to navigate, sense, report, and sometimes strike while someone else is trying to blind it.
Break the spend into jobs and the shape gets clearer.
- Hulls and propulsion that local yards already know how to bend, weld, and launch
- Sensors that can pick out a contact in cluttered coastal water
- Secure links that still work when satellites and shore radios are under stress
- Autonomy software that does not freeze the moment the link drops
- Command tools that can aim dozens or hundreds of boats without a human babysitting each one
Taiwan does not need to invent shipbuilding. Yards already design and produce the physical platforms. The electronics base is deep. What is thinner is the layer that turns a boat into a military system: navigation that holds in a jammed environment, software that coordinates a crowd, and networks that assume the command post might not be there tomorrow morning.
A Split Between Hardware and the Nervous System
The emerging model is blunt, and I think that bluntness is a feature. American autonomy software. Taiwan-built platforms. A Taiwanese workforce doing the manufacturing. One executive in the autonomy business put it almost that cleanly. The line is less a slogan than a division of labor. Yards supply steel, composites, and production rhythm. Outside firms supply much of the software and the command nervous system.
That split will annoy purists who want a fully domestic stack. It also matches how the island already works. Semiconductors, machine tools, and contract manufacturing grew by pairing local process discipline with outside design know-how. Defense is pricklier, because the buyer is the state and the failure mode is not a late shipment. Still, the pattern is familiar enough that industry people are not treating it as science fiction.
The boats are the easy part to photograph. The hard part is the software that keeps a crowd of them useful after the first network node goes dark.
Government-backed shipbuilder CSBC has already paired with Rhode Island-based HavocAI to work on autonomous surface vessels, mixing local manufacturing with outside autonomy tools. Other American firms in the unmanned and defense-software world, including Saronic, Anduril, MARTAC, VATN Systems, Albacore, Shield AI, and Auterion, have built relationships with Taiwanese organizations, among them the state-owned National Chung-Shan Institute of Science and Technology. None of that is a signed fleet. It is a map of who is already in the room.
Who Wants the Contracts
Domestic yards are not waiting for a white paper to finish. CSBC has shown an unmanned boat called the Endeavor Manta and has talked about capacity around forty hulls a year. Jong Shyn and Lungteh, established shipbuilders, are in the competition. Firms better known for drones and chips, Thunder Tiger and Myson Century among them, are edging into unmanned maritime systems. That mix is telling. The bid is not only from people who paint hull numbers. It is from people who already live in electronics.
Forty boats a year is not 1,320. Anyone who has watched a prototype become a production line knows the gap between a demo day and a cadence. Tooling, suppliers, trained welders, test ranges, and a customer who does not rewrite the spec every quarter all have to line up. If the navy actually proceeds, the program becomes an anchor order. Anchor orders are how niche industries stop being niche. Without one, you get a handful of impressive boats and a slide deck.
Jobs These Boats Are Meant to Do
Not every unmanned hull is a kamikaze. Lumping them together is how budgets get misunderstood and how opponents write easy critiques. A serious coastal force would want several types, even if they share engines and radios.
- Persistent surveillance in lanes where manned patrols are expensive and politically loud
- Targeting support that feeds a shore battery or a manned ship without exposing the shooter
- Watch on undersea cables and pipelines that a blockade does not need to cut with a headline
- Armed one-way craft held back for a narrow set of contingencies
- Decoys and clutter generators that force a larger fleet to waste sorting time
Undersea infrastructure deserves its own sentence. A modern economy runs on cables most people never see. Interfering with them is cheaper, quieter, and easier to deny than a missile barrage. Boats that can loiter, inspect, and raise an alarm are not glamorous. They are the sort of tool a coast guard actually uses on a Tuesday. Folding that mission into the same industrial push as attack craft is smart, because the peacetime customer keeps the line warm.
I am skeptical of any plan that treats weapons carriage as the default. Arming a small boat raises rules of engagement, identification problems, and the chance that a fishing skiff gets misread. The attraction of the concept survives that caution. A platform that can be armed is not the same as a platform that must be armed on every sortie.
Pressure That Does Not Always Look Like a Fleet
Maritime pressure around Taiwan has been expanding in ways that are easy to film and hard to answer with a single dramatic counter. More ships close to the island. More coast guard and maritime militia activity mixed with navy gray hulls. More drills that rehearse the geometry of a closure without declaring one. The point of that pattern, if you take it at face value, is to normalize presence and to train the other side to treat abnormal days as routine.
A manned response to every probe burns fuel, crew hours, and political attention. Unmanned pickets do not remove the need for people. They stretch the calendar. A cutter can stay pier-side while a small boat holds a box of water and pipes back a track. That is not deterrence in the classical sense. It is stamina. Stamina matters when the other side is trying to exhaust you without crossing a formal line.
There is a trap here, and it is worth saying out loud. If every probe gets a robotic answer, the probe becomes cheaper for the side that sends it. The side under pressure still has to decide which contacts deserve a human. Software can rank them. It cannot own the decision. Anyone selling full autonomy as a moral shortcut is selling past the actual problem.
The Ukraine Lesson, Stripped of Myth
Commentators love the Black Sea comparison, and they are not wrong to reach for it. Unmanned surface craft have damaged and sunk ships that, on paper, should have been untouchable by a country without a traditional blue-water fleet. The useful lesson is narrower than the viral clips. Cheap attackers work when the defender’s sensors are strained, when the attacker accepts losses, and when the operational area is constrained enough that a small boat can reach a valuable target.
The strait is constrained. It is also watched, crowded, and close to a defender with a huge magazine. Copying a tactic without copying the context is how concept papers fail. What travels is the cost-imposition idea. A stronger fleet can still win the exchange and lose the week, if every exchange consumes interceptors and attention that were meant for something else.
Recent operational experience, read carefully, also shows the boring prerequisites. You need bases that are not all in one harbor. You need crews, even for unmanned systems, who can maintain, arm, and retask boats under disruption. You need a legal and political story for using them near commercial traffic. Skip those and the fleet is a parade.
Where the Technology Still Falls Short
Cathy Fang of Taiwan’s Research Institute for Democracy, Society and Emerging Technology has pointed at the gaps that matter: resilient communications, AI-enabled autonomy and command-and-control, and multi-vessel coordination. That list is not a brochure. It is a warning label. Building thousands of drone boats is the visible half. Keeping them communicating, navigating, and coordinating after satellites, shore networks, and command centers become targets is the half that decides whether the money was spent on a capability or on inventory.
Resilient communications is the phrase everyone nods at and few budgets fully fund. A link that works in an exercise is not a link that works when the other side is hunting emitters. Boats may need to go quiet, share data in bursts, ride commercial waveforms when that is wise, and fall back on inertial and visual navigation when GPS is a liar. None of that is exotic in a lab. All of it is expensive to prove at sea, in weather, at scale.
Multi-vessel coordination is the sleeper problem. One clever boat is a demo. Fifty boats that do not collide, do not all ping the same satellite, and do not all chase the same false contact are a system. The software has to degrade gracefully. If the commander loses the picture, subgroups should still do something sensible rather than mill around or, worse, all sprint at the last reported track.
What has to survive contact: Navigation without reliable GPS Links that assume jamming Local decisions when the shore link dies A human who can still veto the swarm
I have sat through enough autonomy briefings to distrust the phrase “set and forget.” The honest version is supervised autonomy with a plan for unsupervised minutes. Those minutes are where accidents and escalations live. A program that budgets for the software and the rules, not just the hulls, is the one worth taking seriously.
Shipyards Already Know the Hard Physical Work
There is a reason this concept is plausible on this island and would be a fantasy on many others. Taiwan builds ships. It builds the electronics that go into ships, phones, and factories. The missing pieces are specific, not general. Secure comms, autonomous navigation, sensors tuned for clutter, command software, and the glue that lets a large number of vessels act together. Those are acquirable. They are also the pieces a rival will try to break first.
Capacity is the quiet constraint. A yard that can finish forty unmanned boats a year is a start, not a surge. Surge means extra shifts, second sources for engines and composites, and a test regime that does not become the bottleneck. Defense production fails in the queue more often than it fails on the drawing board. If several yards split the work, you gain resilience and you inherit integration pain. Both are real.
Labor is the other quiet constraint. Skilled trades are not a slide you can duplicate. A program that hires and trains, rather than assuming the workforce appears when the contract is signed, will look slow in year one and less fragile in year four. In my experience, the programs that skip that step announce a number and then quietly redefine it.
American Software, Local Steel
The partnerships matter because they attack the gap Fang described rather than the gap photographers notice. Taiwan can supply yards, electronics manufacturing, and, if the order is real, production scale. American defense-technology firms can supply much of the autonomy layer and the command software. That is not charity and it is not a simple export. It is a bet that a combined stack reaches the water faster than a fully indigenous stack that is still in requirements.
There are frictions worth naming without melodrama. Export controls can slow software and radios. Classification can slow integration. A firm that wants its code to stay proprietary will clash with a buyer that wants the right to modify it after the contract ends. Those fights are normal. They become dangerous only if they are discovered after the boats are already in the water and the link standard turns out to be a private dialect.
A healthier version of the deal writes down interface standards early, keeps a local team that can maintain the stack, and assumes the outside vendor might be unreachable on the worst day. Dependence is acceptable as a starting point. Dependence with no exit is a single point of failure wearing a partnership badge.
| Layer | Likely home | Main risk |
| Hull and yard work | Domestic shipbuilders | Cadence below the advertised fleet size |
| Electronics and sensors | Local industry plus partners | Supply pinch on specialized parts |
| Autonomy and command software | Mostly outside firms at first | Link loss, vendor lock-in, export delays |
| Coordination of many boats | Shared, still immature | Swarm behaves badly when the network degrades |
| Peacetime users | Navy and coast guard | Requirements diverge and split the line |
Read that table as a planning argument, not a scorecard. The rows that look “solved” are the ones a rival will not bother to attack first. The rows that look immature are the campaign.
An Industry, or Just a Procurement Line
If the navy follows through on something like 1,320 vessels, the result could be more than another weapons buy. It could be the anchor customer for a domestic line that builds autonomous vessels in volume, with hardware made locally and much of the technological nervous system supplied from outside. That sentence has a conditional in it on purpose. Proposed is not appropriated. Appropriated is not delivered. Delivered is not sustained.
Anchor customers change supplier behavior. A yard will invest in a jig if it believes year three exists. A software firm will station engineers if the integration work is multi-year rather than a pilot. A training pipeline appears when hiring managers stop treating the program as a rumor. Kill the follow-on order and those investments reverse. I have watched that reversal in other sectors. It is quiet and expensive.
The coast guard track helps, oddly enough, because it gives the line a customer who is not waiting for a crisis. Patrol, inspection, and infrastructure watch are dull missions. Dull missions pay for spare parts. A defense industry that only builds for the worst day tends to atrophy between worst days.
Scale Versus the Boats You Can Actually Crew From Shore
Thirteen hundred is a political number as much as an operational one. It signals mass. Mass is the point, because attrition is the point. A concept that collapses after the first twenty losses is not a concept. At the same time, every boat needs a berth, a maintainer, a charging or fuel plan, and a place in a network that has finite bandwidth. Past a certain density you are managing a traffic problem, not a fleet problem.
A staged buy would be the adult version of the announcement. Prove communications in a jammed exercise. Prove that twenty boats can share a picture without stepping on each other. Prove that a yard can hold a monthly cadence. Then multiply. The temptation, especially under public pressure, is to multiply first and discover the network ceiling later. That order of operations produces warehouses.
There is also the question of mix. A fleet of identical small attackers is easy to describe and awkward to employ. Surveillance craft, infrastructure watchers, decoys, and a smaller armed set will not share every component, but they can share enough to keep suppliers honest. Variety costs money. Uniformity costs options. The better programs argue about that trade in public instead of hiding it inside a single hull count.
What a Blockade Would Actually Have to Defeat
A blockade is not a line of ships and a press release. It is a sustained effort to control traffic, insure nothing, and still look like law rather than war to at least some outside audiences. Unmanned boats complicate the picture in a few concrete ways. They multiply the number of contacts a blockading force must classify. They give the island a way to keep watching lanes even if larger ships are held back. They threaten the assumption that small, cheap probes can be ignored.
They do not, by themselves, keep commercial ships sailing. Insurance, convoy politics, and the willingness of third countries to treat a closure as illegal matter more than any single weapons class. I say that because unmanned fleets are sometimes sold as a substitute for diplomacy and stockpiles. They are an ingredient. A blockade fails or holds on the whole recipe: fuel, food, munitions, repair parts, and the political endurance of people who are not in uniform.
Still, raising the tactical cost is not nothing. A force that expected to loom and intimidate, and instead has to spend interceptors on boats that cost less than the interceptor, has a worse week than its plan assumed. Worse weeks accumulate. That is the entire theory, and it is a theory with a paper trail in other seas, not a theory that requires faith.
Command in a Place That May Not Stay Quiet
Command-and-control is where elegant fleet charts go to get humbled. If the shore station is the brain, the brain is a target. Distributed command, boats that can prosecute a last order, and human teams that can reconstitute a picture from fragments are the unfashionable answers. Artificial intelligence helps rank contacts and propose routes. It does not remove the need for a person who can say no when the picture is wrong.
A practical architecture, if I were sketching it on a whiteboard, would assume three states. Fully connected, when the exercise looks like the brochure. Degraded, when some links are up and the boats cluster into local groups. Isolated, when a boat or a handful of boats have only their last tasking and their own sensors. Training should spend more time in the second and third states than in the first. Most demos do the opposite. That is how you buy a surprise.
Useful test: pull the shore link at minute twenty and see whether the group still has a job, a boundary, and a way home.The test is almost insultingly simple. It is also the test many programs postpone because it embarrasses the slide. Programs that run it early find out whether they bought boats or bought a dependency.
Money, Timelines, and the Risk of a Paper Fleet
Nine hundred thirty million dollars over a multi-year buy can disappear into integration without anyone stealing it. Radios cost more than hulls once you demand them to be hard to jam. Software licenses and engineering support cost more than people expect, then cost again when the waveform changes. Test ranges, safety boats, and insurance for trials are line items that never make the press note. A serious budget shows those lines. A promotional budget hides them inside “systems.”
Timelines deserve the same skepticism. A partner announcement is not a boat. A boat at a show is not a certified system. A certified system is not a monthly delivery. If public language stays at the level of “considering roughly 1,320,” treat the fleet as an intention. Intentions shape industry behavior, which is useful. They do not yet complicate a blockade.
The political risk runs both directions. Delay too long and the industrial window closes, partners wander off, and the concept becomes a study. Rush the buy to own a headline and you field boats whose communications fail the first serious exercise. Either failure is familiar. The narrower path is a public cadence: this many hulls this year, this network test passed or failed, this follow-on released only if the test held.
How Rivals Are Likely to Answer
A larger navy does not have to pretend small boats are irrelevant. It can jam, spoof, net, and shoot. It can flood the same water with its own unmanned craft so that classification becomes a brawl. It can threaten the yards and the piers rather than the boats at sea. It can treat commercial traffic as cover and dare the defender to shoot into a busy lane. All of those answers are available. None of them are free.
The interesting response, from a planning view, is the cheap one: electronic attack and decoys aimed at the coordination layer. If the swarm cannot agree on a picture, mass becomes congestion. That is why the software gap is not a footnote to the hull count. It is the hull count’s premise. Spend on boats and starve the network, and you have handed the other side an efficient target set.
There is a diplomatic answer too, and it sits outside the acquisition chart. Partners who insure ships, who run cables, and who do not want a closure normalized have leverage that no boat replicates. A sea drone program that is briefed as the whole strategy will disappoint. A program briefed as one way to raise cost, beside stockpiles, energy planning, and outside support, is harder to mock.
What Investors and Suppliers Should Actually Watch
For suppliers, the tell is not the speech. It is the contracting rhythm. A multi-year framework with options, interface standards published enough for a second source, and a coast guard buy running in parallel are signs of an industry. A single prototype award with a press tour is a sign of a prototype. Engines, composites, maritime radios, and navigation units that can live without clean GPS are the unglamorous bottlenecks. Software firms that can work on local infrastructure, not only in a distant cloud, will be the ones still useful if links get ugly.
For anyone reading markets rather than maneuvers, the spillover is real but easy to overstate. A defense line can pull welders and electronics capacity that commercial yards also want. It can also leave dual-use skills behind if the program is structured that way. It will not, by itself, re-rate an entire equity market. Treat claims that it will as marketing.
- Watch whether more than one yard gets serial work, not just a demonstrator
- Watch whether communications trials are described with failures, not only successes
- Watch whether the coast guard order survives contact with the navy’s wish list
- Watch export-control delays on the software side before you believe the schedule
- Watch training billets. Boats without maintainers are inventory
Those five checks are deliberately dull. Dull checks are how you tell a buildout from a briefing.
A Narrow Water, a Wide Industrial Bet
Step back and the proposal is a bet on asymmetry in a place where asymmetry is not optional. Matching hull counts with a continental shipbuilding base is a losing spreadsheet. Complicating the spreadsheet is available. The tools are small vessels, local yards, an electronics industry that already runs hot, and outside autonomy software that has spent the last several years learning what breaks at sea. The price, on the order of $930 million for a navy concept near 1,320 small attack drones, is large for a new class and small beside the platform it is meant to hassle.
The bet fails if communications and coordination stay brochure-deep. It fails if production stays at demonstration scale while the public number stays in the thousands. It fails if armed and unarmed missions are mashed into one hull to satisfy every stakeholder. It has a chance if the island treats the nervous system as the main buy, uses the coast guard to keep the line alive, and accepts that some of the software will come from partners who must not become a single point of failure.
I do not think unmanned boats replace a navy, a stockpile, or a political strategy. I do think a crowded strait punishes forces that can only afford to watch with expensive ships. A large number of smaller craft, built where the steel and the circuit boards already are, is a serious attempt to change that punishment. Whether it becomes a fleet or a footnote depends on tests most people will never see: a link pulled on purpose, a yard hitting a monthly number, a group of boats that still has a job when the shore goes quiet.
That is the part worth watching. Not the render of a sleek hull. The Tuesday when twenty of them have to share one ugly picture, in weather, with the network half gone, and still come home with something a commander can use. If that Tuesday works, the $930 million starts to look like a down payment. If it does not, it is a very modern way to buy boats.
