The views expressed in this article are those of the author(s) and do not necessarily reflect the official policy or position of the Department of the Air Force, the Department of Defense, or the U.S. Government.

In an August 28, 2023 speech, Deputy Secretary of Defense Kathleen Hicks announced the Replicator initiative, referring to “mass” as “the PRC’s biggest advantage.” Replicator has a relatively simple mission: to “field attritable autonomous systems at scale of multiple thousands, in multiple domains, within the next 18-to-24 months.” In her speech, the Deputy Secretary made clear that while there is still a need for exquisite capabilities like stealth aircraft, there was also a need to “leverage platforms that are small, smart, cheap, and many.”

Why DoD is Launching Replicator

Replicator builds on years of innovation and quickly accelerating experimentation. The initiative also heeds the lessons of Ukraine, by mixing the success of massive numbers of low-cost, attritable systems, with greater autonomy needed to survive in an even more highly contested environment.

Massive numbers of systems, like drones, can pose multiple operational dilemmas for an adversary. If kinetic munitions are used against them, then those munitions and the targeting assets are not fully engaged in other activities. If they are ignored, filtered-out, or otherwise left untouched, then they can accomplish their mission with limited interference. Significant numbers of autonomous systems also provide flexibility to U.S. and allied commanders as their decisions can shift from do I have the resources to how many systems do I want to use for a given problem.

Low-cost, attritable systems are an answer to questions posed in Ukraine. Despite losses of up to 10,000 drones per month, the Ukrainian military can still use them to generate combat effects. If the loss of small numbers of vehicles translated into a true loss of capability, as it would with crewed air- and sea-based platforms, then the usefulness of large numbers of autonomous systems would be questionable. Attritable has many qualities: it means systems that can be affordably lost; as the Deputy Secretary emphasized in a second speech on Sept 6, 2023: it means fielding “without long maintenance tails” of legacy platforms; it means that it’s easy to divest and “move on to the next thing” to respond to a changing adversary.

Autonomy is a necessary element to Replicator’s success. It solves two challenges:

1. The ability of U.S. and allied forces to control mass numbers of systems simultaneously. Autonomy allows a single operator to control multiple vehicles and operate at scale. 
2. Survival in the contested (anti-access, area-denial, or A2/AD) environment they must operate in. Without the need for a control link from operators, the Replicator systems can survive and continue to operate even under heavy jamming, all while U.S. forces deploying them can minimize their vulnerability to detection.

Beyond the direct applications to the DoD, there are also second-order effects that the Deputy Secretary explicitly courts. Her ‘commander’s intent’ to field operational capability fast, necessarily drives the department and commercial manufacturers to respond. Large defense contractors starting with clean-sheet designs are not set up to respond to this demand signal. Instead, this call can help bridge the gap from innovation pilots to operational capabilities. The ability to rapidly innovate, field, and scale up production can be critical capabilities before a possible confrontation in the Indo-Pacific, and will be even more important if a conflict ever actually kicked off. 

As the Deputy emphasized in her second speech, Replicator is also intended to drive culture change towards stopping “a larger aggressor from achieving its objectives.” This push makes clear the desire to address both the manufacturing and acquisition challenges that come with buying thousands of drones and the issue of developing concepts of operation for the platforms and preparing operational use by forward deployed forces. New, rapidly fielded capabilities challenge the PRC or any potential adversary to respond. Faster fielding of innovative systems “let us think and act differently” to deter aggression.

Scoping and Defining Replicator

When the Deputy Secretary announced Replicator and ‘unpacked’ the Initiative 10 days later, she called for a mass of low-cost, “all-domain attritable autonomy” – but what does that mean?

Fortunately, she defined many of these terms for us:

• Mass is multiple thousands. At a minimum, that means 2,000. Her intent, however, clearly drives us to think bigger if the market can handle it.
• Low-cost, for these purposes, is somewhat circularly defined: quantitatively it means whatever fits in the budget that allows the delivery of multiple thousands of systems.
• She also clarified all-domain in her second speech: from systems afloat to ground vehicles, aircraft to satellites, even active protection systems and countermeasures are within scope.
• Attritable, importantly, is not expendable. In the speech, the deputy secretary mentions a 3-to-5 year use window “before we move on to the next thing.” That means that unlike an expendable system, an attritable one must have the fuel, endurance, navigation and control to make it back to U.S. or allied control for re-use. It does, however, mean that the loss of some in combat or occasionally to lower reliability¹ would be acceptable.
• Lastly, autonomy should be considered on a spectrum. Autonomy can be a simple autopilot or cruise controls; it can be an automatic take-off and landing system; it can be an agent- or rule-based swarm; or it can be simple artificial intelligence (AI) running on the edge or a complex one running offboard.² For now, consider autonomy as this: the ability for a system to accomplish its mission having been tasked by an operator without significant further human involvement.

Some reporting indicates that the price tag for Replicator will be “in the hundreds of millions of dollars rather than billions,” and in her second speech, Deputy Secretary Hicks specifically says it will “use existing funding” rather than “asking for new money in FY24.” While potentially separate, additional media reports and the House Appropriations Committee (HAC) proposed Defense Appropriations Bill itself call for a DIU-led “hedge portfolio” and Replicator may leverage portions of this hedge portfolio and $1.8B in AI-related funding in the FY24 budget. A reasonable budget for Replicator may then be on the order of $400M to $800M. The FY24 committee proposal³ from HAC⁴ matches the approximate reported $1B hedge portfolio and also places a “DIU Fielding” effort roughly halfway between the $400M to $800M discussed below, with a set-aside of just over $200M for support to Combatant Commands.

Affordability, Quantities, and System Mix

Given the quantities desired, and a rough expected budget, we can estimate a range of unit costs for potential platforms:

UxS shows viable paths to meet the cost range:

While the math is straightforward, the consequences for these various quantities and price ranges are more complex. Particularly in operational capability, as well as industry capacity, these offer very different solutions to the Deputy Secretary’s intent.

With average unit costs ranging from $8K to $400K, this can scope the usable capabilities for unmanned⁶ systems (UxS). For example, while UAS (Unmanned Aerial Systems) can be found to be within this price range, only the smallest classes of satellites (which the DepSecDef mentioned in her Sep 6 remarks) are likely to figure significantly into the fielding of thousands of systems. A review of US-based offerings of different UxS shows viable paths to meet the cost range⁷:

Based on the average cost range and the unit costs above, it is possible to create a diverse mix of platforms under the cost cap of $800M – or focus on masses of purely low-cost vehicles. In the air, the DoD could choose to buy many low-end UAS as well as small numbers of higher-end-but-still-affordable XQ-58 Valkyries for $6.5M⁹ each. For example, the Department could buy 10,000 low-end small commercial-grade UAS ($60K each) for $600M. For mid-range capabilities, they might purchase 2,000 higher-end military-focused UAS ($250K each) for $500M. For a high-low mix, 50 XQ-58s ($6.5M each) and 5,000 of the commercial grade UAS ($60K each) would run $625M. All of these options fall within the assumed budget.

The same mix can apply to other domains. However, since Replicator focuses on mass, we will simplify and look only at purchases of lower-cost UxS, not as part of a high-low mix.

Unmanned Systems in Indo-Pacific

UAS in the Indo-Pacific can likely perform Intelligence, Surveillance, and Reconnaissance (ISR) missions, be communications relays, and perform critical electronic warfare and decoy missions during a conflict. As only small vehicles are affordable, aerial logistics is not a viable option at scale.

While this review won’t dive deep, some smaller UGVs cheaper than the average may play a role in Replicator, but will ultimately be limited to other theaters given the distances and water involved in the Indo-Pacific. 

As demonstrated by Task Force 59 under the US Navy’s 5th Fleet, USVs and UUVs will be major players in Replicator. The challenge for USVs will be delivering on the scale of thousands given the price points involved. Likewise, given the costs and progress noted by the SAC report on the FY24 budget, larger UUVs will not fit the Replicator budget. Smaller UUVs, however, can provide capability at affordable cost points. ISR, communications, bathymetry and ship tracking are all viable pre-conflict, and EW and decoy missions are possible if a conflict began.

Finally, there are certain classes of space systems that will exceed the cost limitations while others that will fall within the range.¹⁰ Decision makers will need to consider the cost of ground infrastructure to support the satellites. Licensing data from commercial offerors is an option to minimize those costs. As with most low earth orbit satellites today, they would be well placed for ISR and communications missions.

Design Tradeoffs

The Deputy Secretary has laid out the budget and business characteristics that she expects to see from systems delivered under Replicator.

The challenge of the PRC drives two additional fundamental qualities that any system under Replicator must meet in addition to the mantra of small, smart, cheap, and many. To challenge the A2/AD environment, they must first be able to reach the operational environment and second, once there, be able to do a mission-relevant action or activity.

These two operational qualities drive design choices. A deep dive on UAS options helps to illuminate these price/capability tradeoffs.¹¹ We chose UAS in part because there is a listing already pre-vetted and available for purchase on the Defense Innovation Unit’s (DIU) Blue UAS website.¹² The available platforms can be categorized as either small rotary wing, large rotary wing, or fixed wing/tilt rotor, with accompanying capabilities and limitations for payload mass, range, altitude, speed, and cost. 

Unlike in Ukraine, where the frontline may be measured in meters, the Taiwan Strait is approximately 100 miles (161 km, 87 nmi) wide. It is one of the shorter distances for potential conflict across the wider South China Sea.

As a rough metric for analysis, any vehicle should be able to “get into the fight” (for the purpose of analysis here, at least 50 miles), persist long enough to have a mission effect, and then return. Deploying from a nearby ship or aircraft is certainly possible, but launching at sea or in the air is more technically difficult, and will likely increase the cost. Launching from further away, allied territory for example, will require a platform with more fuel and greater range, necessarily requiring a larger airframe and greater cost. 

Because of the distances involved, Figure 1 above makes clear that a fixed-wing/tiltrotor platform would be preferred for UAS. Some much larger rotary-wing drones may be able to cover parts of that distance if launched nearby. 

Of course the ability to perform mission-relevant actions also requires tradeoffs. 

• For ISR missions, higher altitude UAS can “see” more land and sea at a time (see Figure 2a), but also need better propulsion, better cameras, and so will cost more. Likewise, for a given camera, better resolution can be achieved at lower altitudes, but then more vehicles are required for equal coverage. For more complex missions where multiple drones cooperate to improve their total sensor performance, there is also a tradeoff between better sensors on larger platforms and the ability to coordinate between vehicles. Again, more complexity and size means more expensive platforms and fewer of them.
• For electronic warfare missions, larger vehicles can provide more power, but this increases their size and complexity, therefore increasing their cost. Similarly, to ensure radars can see EW emissions, they must be able to achieve sufficient altitude (see Fig. 2B). The ability to reach higher altitudes will also drive cost. Quantity itself has value in electronic warfare, and so a direct cost/quantity tradeoff also applies. Communications relay missions will have similar drivers.
• For a decoy mission, larger vehicles present a larger target for missiles, and faster, higher flying UAS are more likely to be detected by radars directing those missiles. That increased size, of course, comes with increased cost.
• For vehicles used as munitions, the mass of the payload will directly impact its effectiveness. However, increases in mass will also drive larger vehicles and therefore cost or the vehicle will lose range, a critical consideration in the Indo-Pacific.

For all mission types, there must be sufficient onboard processing to enable autonomy. Whether for navigation, communicating with other vehicles, or performing other mission functions, greater autonomy will drive complexity and therefore cost. Each vehicle will also need a mission payload, regardless of what it is intended to do. That cost must also be accounted for within the acquisition, and scaling payload production and development may be a parallel challenge to fielding thousands of vehicles. Simple, standardized payloads and interfaces can reduce that challenge.

None of these tradeoffs are impossible to make. Engineers and program managers make them regularly. What is important is to understand how the mission and budget constrain the choices.

Importantly, as the implementation of autonomy is key to making these large quantities operationally viable, Replicator must assess where to exist on the spectrum from autopilots to AI. Limited pilot-in-the-loop control will limit the ability to perform adjustments on the fly. This means that either significant post-launch autonomy must be developed and tested, or the vehicle missions should favor non-lethal¹⁴ missions. If an agent-based, rule-based, or other type of autonomy is developed, the risks of mission failure in an unfamiliar environment or of inadvertent escalation must be addressed. Nevertheless, an autonomous system able to perform certain non-lethal tasks, to include target/object detection, electronic warfare/decoy missions, and other similar tasks can persist in the face of threats that would eliminate a remotely piloted platform.

Finally, while the concept of ‘attritable’ platforms for a potential conflict with the PRC is not novel, this effort to do so quickly, at scale, and for operations rather than research, development, test, and evaluation (RDT&E) is new. Given the need to operationalize the Replicator-delivered technology, there will also be a need to develop and exercise test and training capabilities with the units and personnel who will operate these systems. Such analysis is beyond the scope of this post. Nevertheless, these elements are important to the success of Replicator. And while we have laid out some of the viable capabilities and considerations given the initiative’s scope, we will not recommend a particular path, vendor, nor the acquisition approach to get to fielding.

Conclusions and Recommendations

Industry has viable solutions to meet the intent of the Replicator Initiative at the cost and quantities the Department of Defense needs. Meeting the operational needs of USINDOPACOM will be more challenging given the distances involved compared to many current commercial offerings. 

Prices for UAS are the best of the ‘UxS’-type vehicles, aligned with the Initiative, with smaller USVs also viable solutions. Surprisingly, small satellites made by the commercial sector are also well within the cost constraints and can deliver operational capability. However, there are substantial infrastructure and strategic considerations needed before making satellite purchases.

Finally, the ability of American industry’s current capacity to manufacture thousands of drones across all domains is unknown, as is the effect on the market by having the DoD as a high volume customer. Further analysis is appropriate, though should occur concurrently with Replicator, not as a prerequisite. 

As the department reviews commercial offerings, it should also consider the following recommendations:

1. While a mix of short and long-range capabilities is appropriate, in the Indo-Pacific, longer range capabilities should predominate.
2. With a great variety of platforms and vendors, the advantages to investments in fielding simple, light, and preferably interchangeable payloads outweigh the typical challenges in standardization.
3. All domain capabilities are important, but the mobility and altitude of UAS and the endurance and payload capacity of USVs are key factors in preparing for a conflict in the Indo-Pacific.
4. DIU and other innovation organizations have led the connection with commercial companies to enable Replicator. Their ongoing efforts should seek to understand the UxS industry’s manufacturing constraints to help guide DoD investment strategies.

Replicator sends a powerful message to DoD acquirers, commercial partners, and US adversaries. Smart decisions on which capabilities to replicate and how to prepare operators to receive them will determine its deterrent value.

1.  Random failures are acceptable, but systemic faults that cause repeated, significant losses would not be acceptable, even for an attritable system.
2. Given the complexity and cost for training AI for a combat environment and in the interest of fielding many cheap systems quickly, most autonomy under Replicator will probably be simpler autonomy.
3.  Pg 251 – 254 from https://www.congress.gov/118/crpt/hrpt121/CRPT-118hrpt121.pdf.
4. Notably, the Senate Appropriations Committee (SAC) proposal does not include the same “DIU Fielding” funding that HAC includes at line 281 in the Defense-wide RDT&E table. It does, however, support the President’s budget funding of attritable aircraft at $126M and $177M for two Marine Corps programs under Marine Corps Advanced Technology Development, (page 217 and Navy RDT&E table line 19). While both the HAC and SAC reports increase funding for that line generally, HAC specifically adds more than $32M for “low-cost attritable aircraft technology.”
5. *This simple first-order approach neglects non-recurring engineering (NRE), test & training and operations costs, and other cost factors; however, with the objective to field within 18-24 months, most solutions must be nearly off-the-shelf, reducing NRE. Operations and sustainment costs are typically accounted for separately, but if included would serve to decrease the allowable unit cost to achieve the quantity desired.
6. The term “unmanned” is used here rather than “remotely piloted” given the expectation from the Replicator announcement that significant levels of autonomy should be included. While “uninhabited” and “uncrewed” are gaining broader usage, unmanned is more widely used and understood.
7. Non-US-based offerings would undermine the desired industrial base development, but some allied and friendly capabilities, like those being fielded by Ukraine, are likely cheaper. For Ukraine in particular, however, it is unlikely that they would spend industrial capacity only to export weapons abroad.
8. 
   Table 2 Notes:
   * No overt US response to 2019 and 2023 downings of RQ-4 and MQ-9 by Iran and Russia, respectively
   ** Average unit price according to analysis of data from Jane’s Markets Forecast; Reference: Wilson, Bradley, Ellen M. Pint, Elizabeth Hastings Roer, Emily Ellinger, Fabian Villalobos, Mark Stalczynski, Jonathan L. Brosmer, Annie Brothers, and Elliott Grant, Characterizing the Uncrewed Systems Industrial Base. Santa Monica, CA: RAND Corporation, 2023; pp. 18-19; https://www.rand.org/pubs/research_reports/RRA1474-1.html
   ***See RAND report from footnote 9.
   ^{^} While not a random sample, the MANTAS T12 had a $460K per unit cost in a single sale to the UK, and reporting indicates that Ukrainian systems have unit costs around $250K. If purchased using a data-as-a-service approach, costs can be even lower. 90-day missions using various configurations of Saildrone run an average of $290K to $670K, with minimum costs far lower, and cost efficiencies as missions extend beyond 90 days. Data based on analysis of GSA Advantage posted costs.
   ^{^^} See RAND report from footnote 9. Note that using the all class average obscures variation, including much lower costs for smaller vehicles. Examples: Iver UUV, $53K – $82K; Orca XLUUV, $109.6M; The USN is already investing in other small UUV partnerships: Razorback torpedo tube launch and recovery UUV, and Lionfish and Viperfish SUUVs should be more cost effective than their larger cousins.
   ^{^^^} While some companies are likely able to drop costs lower, publicly available data suggests that production-like quantities of satellites typically have unit costs between $5M ($99.4M for 20 buses) to $8M ($2.4B for 300 satellites). These costs may not include payloads to accomplish a mission, which would drive the relevant costs higher. For Cubesat-class designs, costs may be lower, especially at higher production rates; systems like Starlink are likewise expected to beat the ‘typical’ costs. While rates for imagery-as-a-service are not widely published, they appear to range from $5 – $100 per square km, with US government contract costs significantly lower on a dollar per square km rate.
   ^#Data is limited, but some news reports and some science bloggers indicate Starlink satellite unit costs are within the listed range.
   ^% Planet, a major player in imaging from Low Earth Orbit, may have satellites that cost “a few thousand dollars” according to some older media reporting. Given their use of the cubesat form factor originally designed for student projects and the cost of components, cubesat-class vehicles can reasonably be built at scale for tens of thousands of dollars.
9. Low quantities are $6.5M/copy with a recent Navy buy a bit higher. Unit costs go down as quantities go up.
10. Strictly speaking, the DoD could probably buy cubesat or Starlink space vehicles that are within the $8K – $400K range. However, this does not include ground infrastructure to support the vehicles in orbit. There are strategic considerations to commercial vs US government ownership and control. These should be weighed alongside the cost.
11. For simplicity, we will limit our analysis to UAS, but similar choices will apply to USVs and UUVs; satellites to a much lesser extent.
12.  Since these UAS are already cleared for DoD purchase, they should have favorable delivery schedules compared to other commercial vehicles not on the list. Other offerors, of course, should be considered.
13. 
    Table 3 Notes:
    ^{^}  Platforms listed were from DIU’s Blue sUAS list, https://www.diu.mil/blue-suas-2, plus the Easy Aerial Tern, the RQ-11, RQ-28, and Black Hornet. Prices were gathered from GSAAdvantage where possible, and news reporting on prices or contract values and quantities if not listed on the GSA website. For performance specifications, including payload capability, altitude, and speed, DIU descriptions and vendor specification sheets were used where available. Vehicles were sorted into vehicle type categories and the minimum and maximum ranges for the category identified. Caveats: Data listed is the best case – for example, the altitude (service) ceiling is likely not achievable with the maximum payload mass, and since maximum endurance is also generally calculated at zero payload mass, the range is also affected (see next note). As some gaps exist in the data, and the list of potential vendors is not exhaustive or a true random selection, it is best used to understand the relative capabilities of different vehicle classes.
    ^* This is calculated as the product of the vehicle’s max speed and its endurance. As battery life/endurance is lower at maximum speed, the true 1-way range is expected to be less.
14.  “Non-lethal” is used rather than “non-kinetic,” though in most cases they have significant overlap. Non-lethal is chosen only because certain uses of non-eye-safe lasing to disable a system may cause inadvertent harm.