Executive Summary
The 2026 Iran war (28 February to 8 April, 39 days) exhausted Western premium-munition stockpiles: the Saudi and Gulf bloc burned through roughly 86 percent of its Patriot PAC-3 missiles (CSIS Last Rounds), and United States stocks were drawn down (CNN, 21 April 2026). Lockheed Martin's PAC-3 ramp to 2,000 a year by 2030 is too slow to help Lithuania before 2028. Two import options also closed: Wild Hornets STING was export-blocked (The War Zone, March 2026), and Helsing's HX-2 was paused in January 2026 over a 25 percent take-off failure rate and electronic-warfare problems with its AI targeting (Bloomberg, 19 January 2026). Lithuania is absent from Macron's Île Longue partner list of 2 March 2026 — absent, not explicitly excluded — and is not a party to the Polish-French Treaty of Nancy (signed 9 May 2025; Tusk-Macron Gdansk on 20 April 2026 was its first follow-up). This is a gap Lithuania must close on Lithuanian terms. Granta Autonomy, run by chief executive Gediminas Guoba (Litvinas is a co-founder, not the CEO), produces several hundred FPV drones a month with a stated target of thousands a month (dronelife.com, 30 September 2024). NanoAvionics (Kongsberg-owned satellite company) and Brolis Defence (infrared sensors, electronic-warfare optics) are not FPV producers. The recommended next step is an industrial study by the Ministry of National Defence, the Ministry of the Economy, and the State Defence Council that maps a realistic path: a conservative year-six target of 30,000 to 50,000 FPV a month (sixty- to one-hundred-fold scale-up from the Granta baseline), a labour need of about 960 workers at that volume scaling toward 20,000 for half a million a year (almost certainly needing Polish and Ukrainian co-production), and European funding pull via EDIRPA (310 million euros across 27 states), PESCO loitering-munition strands, and SAFE (6.3 billion for Lithuania — a loan serviced from defence budget, not a grant). Scale, vendor mix, and dispersal are determinations for the Lithuanian government and Seimas.
The Problem
Russia produces Shahed-class one-way attack drones at the Yelabuga plant at roughly 3,000 to 5,500 a month, and the March 2026 monthly tally of strike-Shaheds reached 4,186 (Euronews, 2 April 2026). The Geran-3 jet variant cruises at 300 to 370 km/h at up to 9,000 metres altitude. Russia also makes around 75,000 KAB glide bombs a year (RUSI, 2025), released from Belarusian airspace at 130 to 200 km standoff — putting every Lithuanian site east of Kaunas inside engagement range. On 15 April 2026 the Russian Ministry of Defence published a list of 21 European drone-coproduction targets that explicitly named Vilnius (pravda.com.ua, 16 April 2026, with corroboration from UNITED24, Meduza, and TASS). A cyber attack typically comes before a missile strike: on 29 to 30 December 2025, the Russian state hacker group known as Sandworm wiped the digital control systems of two Polish power stations using stolen authentication certificates (ESET via welivesecurity.com; Dragos).
Granta Autonomy's verified output at the start of 2026 is several hundred FPV drones a month, with a stated target of thousands a month (Guoba, dronelife.com, 30 September 2024). Eltonas (a Vilnius component supplier) and Astera Industries are smaller secondary producers. There is no domestic production above the FPV class. Lithuania's stockpile of imported strike munitions — JASSM, GMLRS, Spike — cannot be resupplied at saturation tempo because the Western production lines are backlogged through 2028. Ukrainian operator data (RUSI, 2025) puts the kill probability for FPV drones in a heavy electronic-warfare environment at 0.20 to 0.30, which means a 700-euro drone produces a real cost-per-kill of about 2,333 to 3,500 euros — still 300 to 2,500 times cheaper than the targets it destroys, but an order of magnitude above the headline 700-euro figure that often circulates. The 23 September 2025 Šakalienė amendments authorise the military to engage airspace violations only; they are not a legal predicate for cross-border deep strike or production licensing (kam.lt; militarnyi.com).
Without action: In a saturation scenario modelled on Ukrainian operational data, Lithuanian strike-munition stocks would be exhausted in 7 to 14 days, and Western resupply would not arrive inside the engagement window. The Article 5 tripwire would fail at the strike-fires layer rather than at air defence, and the regional industrial geography would set around Polish and Latvian producers without a Lithuanian share.
Lithuanian Context
Lithuania's 2026 defence budget is 4.79 billion euros, or 5.38 percent of GDP, with a weapons envelope of about 1.7 billion. A 350-million-euro annual operating cost at full-rate production would absorb roughly 20.6 percent of that envelope — a real crowd-out the Ministry of Finance and Seimas Budget Committee will have to reconcile. Manning is the binding floor: at Ukrainian wartime productivity of about 25 FPV drones per worker per year, a 50,000-a-month line needs about 960 workers (achievable inside Lithuania's roughly 14,000-strong defence labour pool), but a half-million-a-year line needs 20,000 workers and almost certainly requires Polish and Ukrainian co-production. European funding pull runs through EDIRPA (310 million euros total, realistic Lithuanian share under 50 million), PESCO loitering-munition strands LUMINA, SKYRAPTOR, and TALON, the European Defence Fund (70 to 150 million euros over the multiannual financial framework cycle), and SAFE (6.3 billion euros for Lithuania, a loan envelope serviced from the defence budget per lrt.lt — not a grant). The NATO Drone Coalition is a Ukraine-procurement-pooling vehicle co-led by Latvia and the United Kingdom (20 letter-of-intent signatories, 2.75 billion euros in 2025, 30,000 drones per mod.gov.lv), and the NATO Drone Wall (initial operating capability end-2028) is a layered counter-drone and integrated air-and-missile-defence mesh — neither is a strike-FPV offtake pool. Anduril's Anvil-M is a counter-drone variant for United States base defence, not a strike production model.