Drones--the hot "physical AI," rare earths remain crucial

Drones are upgrading from "battlefield tools" to a broader and larger industrial chain.

According to Wind Trading Desk, Zornitsa Todorova, thematic FICC research analyst at Barclays, stated in the latest report that the global drone market has doubled from about $20 billion (2020) to over $40 billion (2025) in the past five years, and is expected to reach $50 billion in 2026, pointing to a market size of $250 billion in 2035.

Barclays highlighted the most easily overlooked "true constraints" of this business: “Constraints are increasingly extending beyond defense budgets—now evident in AI capital expenditure, energy and critical minerals.” The unit price is indeed dropping—single-use drones are commonly priced between $20,000 and $50,000—but for scaled combat and sustained supply, the real spending—and real bottleneck—are shifting more to computing power, power supply and key minerals, with rare earths still crucial.

Barclays sees drones as a typical form of "AI entering the physical world": defense budget expansion is only one thread, and the other three—AI capital expenditure, energy and power grid, critical minerals—are tying defense capabilities to broader investment cycles, inflation, and industrial policy.

Besides, military demand is an accelerator but not necessarily the end point. Barclays expects the civilian share to rise from about 55% now to about 65% by 2035; agriculture, warehouse inventory, and last-mile delivery will become the main battlefields for new growth: drone spraying can cut costs by about 70%, reduce operation time by over 90%, and cut water usage by about 90%; in warehouses, inventory process time can be reduced by 50%; in delivery, there are already scaled operations with an average delivery time of 19 minutes.

Drones Enter the Stage of "Explosion in Patents and Revenue"

Barclays defines "takeoff" with two sets of data: market revenue expansion and innovation density surge. The number of global drone patents granted grew from less than 200 in 2014 to nearly 8,000 in 2024 (a 45-fold increase), mainly driven by defense R&D investment and rapid commercial expansion.

Estimating the market size itself isn't easy—many companies don't disclose revenue by product line—Barclays' approach is to aggregate company disclosures, industry research and demonstration materials for multi-source estimates, and notes that forecast convergence is high for 2024-2025, standard deviation around $5 billion.

In a broader framework, Barclays categorizes drones under "Physical AI": Embodied intelligence covers humanoid robots, autonomous vehicles, advanced industrial automation, and drones, jointly viewed as racing toward a trillion-dollar opportunity by 2035. Drones are positioned as the second largest segment, ranging from $150 billion to $350 billion, with the base case at $250 billion; autonomous vehicles are the largest segment (around $250 billion-$550 billion).

Growth Is First and Foremost a Defense Story: Low Cost, Single Use, and Swarms Become the New Standard

The report’s conclusion is straightforward: defense demand is the dominant growth driver for the drone market, accounting for about 40%-50% of related revenue and contributing "up to half" of recent growth. It uses battlefield data to show that "scaling up" is becoming the norm: Ukrainian drone production jumped from about 800,000 units in 2023 to nearly 5 million units in 2025, including about 2 million FPV small systems; in the first weeks after the Middle East conflict erupted in 2026, nearly 2,000 drone strikes were recorded.

Within defense scenarios, Barclays roughly divides drones into two ends:

High-value dedicated platforms: Unit price can reach millions of dollars, featuring long endurance, high performance, multi-tasking.Low-cost, highly scalable platforms: Mainly short-range, limited endurance, common single-use missions, unit price usually under $50,000.

The key change is: the ability of low-cost platforms to scale up “more, faster, and denser” increasingly depends on AI shifting capability from hardware to software—navigation, obstacle avoidance, teamwork—making swarm actions transition from “labor intensive” to “model driven”.

Cheap Units Don’t Mean Cheap Systems: Defense Spending Moves ‘Upfront to Computing Power and Software’

The report stresses a structural shift: Physical AI is moving defense expenditure from traditional "platforms and personnel" to R&D, computing power, data and software, with a tendency to lower costs in operation and organization.

This places defense systems at the intersection of four ledgers: Defense budget + AI capital expenditure + Energy + Minerals, with the latter three increasingly becoming new constraints.

The defense budget itself continues to rise: SIPRI data shows global military spending reached $2.7 trillion in 2024; the UN's scenario projections put this at $3.5 trillion by 2030, and over $4.4 trillion by 2035; in a more aggressive scenario, with military spending at 5% of global GDP, 2035 spending could reach $6.6 trillion.

Structurally, NATO members’ "equipment and related R&D" share of spending has risen from 24% in 2014 to 30% in 2025 (with NATO’s total spending at about $1.5 trillion in 2025), which Barclays sees as an early signal of “tilting toward infrastructure and equipment.”

The report also notes that electricity demand has surged; IEA estimates data centers now consume about 1.5%-2.0% of global electricity; by 2035 this is projected to rise to 4.4%, or about 1,600 TWh. What’s trickier are local constraints rather than nationwide averages: AI data centers have power intensity similar to aluminum smelting and other heavy industry but demand is highly concentrated. In the US, for example, nearly half of data center capacity is concentrated in five regions; Barclays believes the issue isn’t “whether the country generates enough power”, but “whether power can be reliably delivered to emerging demand nodes”, such as Northern Virginia (PJM), Texas and parts of the Midwest.

Critical Minerals Are Not Just a Cost Item, but the Boundary of Strategic Autonomy

The drone hardware stack’s dependence on critical minerals runs through the entire chain:

Propulsion systems: High-performance motor magnets require rare earths (such as Pr, Nd, Sm, Dy) and nickel, copper.Structural framework: Aluminum, titanium, magnesium, tantalum are used for lightweighting and strength.Communications and navigation: Beryllium, gallium, germanium, indium are used for fiber optics, sensors, radar and imaging.

The report identifies 52 types of critical minerals, with China being the dominant supplier of 40. Rare earth concentration is even more extreme: it includes 17 elements, 14 classified as critical minerals, with China’s supply share of these 14 exceeding 90% for each.

Supply chain “de-concentration” is underway, but Barclays’ conclusion is rather cold: over the next five years, non-China regions plan to add more than 50,000 tons/year of rare earth mining capacity, over 40,000 tons/year separation capacity, and at least 70,000 tons/year magnet/alloy capacity, but projects actually reaching FID are few; rare earth mining takes 10-20 years from exploration to commercialization, and mid- and downstream tech and experience gaps are especially hard to make up.

After Defense Incubation, Civilian Will Complete the Story

The report places drones’ “end-game demand” on productivity improvement, not simply defense cycles. It outlines three types of deployment along the “military first—cost reduction—civilian diffusion” pathway:

Agriculture: By the end of 2024, DJI estimates its global active agricultural drone fleet at about 400,000 units (from about 80,000 in 2020); extrapolating at the same growth rate, the agricultural drone fleet could approach 3.5 million units by 2035. Spraying scenarios show: compared to manual backpack spraying, costs drop by about 70%; compared to tractor spraying, costs drop by about 50%; time more than 90% shorter, water use drops by about 90%.Warehouse inventory: After UPS Supply Chain Solutions in Kentucky "Velocity" warehouse introduced Verity autonomous drones, inventory process time dropped 50% within months.Last-mile delivery: Alphabet’s Wing and Walmart, since cooperating in Dallas-Fort Worth in 2023, have achieved thousands of weekly orders with average delivery time under 19 minutes, connecting 18 Walmart Supercenters via a "hub-and-spoke" model.

Risk Disclosure and DisclaimerThe market carries risk. Investment needs caution. This article does not constitute personal investment advice and does not take into account the individual user's specific investment objectives, financial situation, or needs. Users should consider whether any views, opinions, or conclusions in this article are appropriate for their specific circumstances. Investment is at your own risk.