In late May 2026, Reuters reported that SpaceX had renegotiated the price the Pentagon is paying for getting Starlink connectivity on LUCAS one-way attack drones during the Iran campaign. Connectivity that had been costing about $5,000 per terminal at the land/mobility tier was reclassified by SpaceX as aviation-tier service, and increased the price to approximately $25,000 per terminal. The Pentagon agreed to the increase while strikes against Iran were ongoing.
As any modern technologist is well aware, LUCAS is the US' Low-Cost Uncrewed Combat Attack System, a one-way attack drone built by SpektreWorks and modeled on the Iranian Shahed-136. It saw combat in February 2026, during Operation Epic Fury against Iranian drone and missile infrastructure. The LUCAS platform's unit cost runs approximately $30,000 to $35,000. Its entire value proposition is being cheap enough to use in quantity against targets that would otherwise require Tomahawks at $1.3M per shot.
The math after the SpaceX price increase puts a lot of strain on that cost proposition. A $25,000 connection fee against a $35,000 airframe means Starlink represents roughly 70 percent of total LUCAS unit cost. Even taking the Pentagon's framing that the $25,000 is paid once per sortie rather than as a monthly recurring charge, the connectivity fee approaches the cost of the drone itself. The Drone Dominance Program is targeting one-way attack drones at unit costs as low as $2,300 in later production phases, granted with much more focus on quads vs. fixed wing platforms. At that price point the connectivity fee is an order of magnitude larger than the airframe, and the "cheap and attritable" model simply does not survive that math.
While the pricing dispute is a symptom, the reality is there is a deeper structural problem that is worth taking the time to look into.
Why the DoW Is In This Position
Starlink works, and that is why it is everywhere. Affordable, easy to deploy, low latency, and ubiquitous coverage in places where legacy systems do not reach, including both domestic and overseas. Full disclosure here, I have two terminals myself and the service is cheaper and more reliable than most commercial broadband alternatives. The Pentagon adopting Starlink for tactical connectivity was a rational choice based on capability and cost. Ukraine adopting it in 2022 was the same choice under far more pressure. U.S. Navy unmanned surface vessel testing, CENTCOM deployments, and dozens of drone programs reached the same conclusion.
The result is that one company now operates the connective tissue for a generation of uncrewed warfare. SpaceX runs roughly 10,000 satellites, more than 60 percent of all active orbital objects. OneWeb has fewer than 700. Amazon's Project Kuiper is in early deployment. No alternative matches Starlink's scale or coverage at the operational tempo the DoW needs.
As we well know, Elon Musk did not simply stumble into this position. He spent more than a decade and billions of dollars in private capital building a constellation specifically designed to be hard to replace, and to be a preferred provider for government customers including the DoW. It is also not fair to make a statement that the DoW is making bad procurement decisions by going with Starlink. It made rational decisions that compounded into a strategic dependency by design.
What Russia and Ukraine Already Know
On May 27, the Russian milblogger channel Rybar published an assessment of ground robotic combat systems on the Ukraine front. This piece attributes Ukraine's dominance in unmanned ground operations directly to Starlink (translated from Russian to English): "this is connected to the by-now-tiresome Starlink, which allows the Ukrainian Armed Forces to use ground drones very actively. Russian forces, unfortunately, cannot boast the same". The piece concludes that "a catch-up sprint is only possible with the development of one's own satellite communications."
This is a public admission from inside the Russian information space that the ecosystem of distributed, networked, low-cost uncrewed warfare runs on a foreign satellite constellation Russia cannot replicate. This strategic asymmetry is recognized and duly noted.
On the Ukrainian side, the same asymmetry is understood as a vulnerability. Starlink has been Ukraine's primary tactical communications backbone since the first weeks of the war. Ukrainian commanders and outside analysts have repeatedly described loss of Starlink as something that would severely degrade frontline operations.
That loss has already happened under operator control. In September 2022, Musk denied Starlink coverage to support a Ukrainian naval drone attack on the Russian Black Sea Fleet at Sevastopol, citing concerns about nuclear escalation. Also in September 2022, Reuters reported Musk ordered Starlink shut down near Beryslav during Ukraine's Kherson counteroffensive. In March 2025, Musk publicly suggested he might cut Ukraine off if Kyiv walked away from a U.S. mineral deal. In July 2025, a global Starlink outage cut U.S. Navy unmanned surface vessel testing and left platforms adrift.
Russia is admitting the dependency and Ukraine is living the vulnerability. Unfortunately, DoW is now in the same position with LUCAS.
Starshield Is Not the Full Answer
DoW Is well aware, and has been working on solutions to this challenge for a while. Starshield is a SpaceX product line designed for government and military use, sold to the Pentagon under a 2023 contract framework, with enhanced encryption, anti-jam features, and access to a separate satellite layer. Starshield terminals were in use across more than a dozen drone systems when the Iran campaign launched, including LUCAS. The Pentagon was considering an additional purchase of more than 3,500 Starshield terminal subscriptions, including 100 at the aviation tier, when the LUCAS dispute became public.
Starshield in not an answer in solving the dependency problem. Reuters reported that Starshield terminals are designed to access both the dedicated Starshield constellation and the broader Starlink network of roughly 10,000 satellites, and that the two systems remain intertwined at the infrastructure layer. The LUCAS pricing dispute involved Starshield terminals operating against Iran, not commercial Starlink. The leverage SpaceX exercised was over a product the DoW had contracted for specifically as a military-use solution, on military missions, against an adversary the U.S. was actively striking. If Starshield is the supposedly Pentagon-controlled alternative, the LUCAS episode demonstrates that "Pentagon-controlled" still depends on SpaceX pricing decisions, SpaceX operational decisions, and the personal positions of SpaceX leadership.
The same week Reuters published the LUCAS pricing story, the U.S. Space Force awarded SpaceX a $2.29 billion firm-fixed-price contract to build the Space Data Network Backbone using Starshield as the core platform, with a fully operational prototype required by December 31, 2027. The Space Force has stated it intends to award a second SDN contractor by summer 2026, but in practice SpaceX's proprietary Starshield encryption and existing relationship make rival entry difficult on equal footing. The Pentagon zeroed out funding for the Space Development Agency's Transport Layer Tranche 3 in FY26 to shift emphasis toward the Starshield-based SDN; Congress reinstated $500 million to keep multi-vendor options alive.
The Pentagon is simultaneously fighting SpaceX over per-sortie pricing and signing them to a $2.29B contract that deepens integration. Both decisions are individually defensible. Together, they describe a procurement system that has not caught up to its single-vendor exposure.
Starlink Is Not the Only Dependency
The Starlink situation is the most visible example of a pattern playing out across the autonomous systems enterprise. The same architectural choice the DoW made on satellite communications, picking the best commercial option because it was the best commercial option, is producing similar single-vendor concentrations at other layers of the autonomy stack.
The C2 and autonomy software layer is the clearest parallel that we should be watching out for. On May 19, 2026, the Office of the Under Secretary of War for Research and Engineering selected Shield AI to integrate its Hivemind autonomy software onto LUCAS as the AI pilot for swarming, with an operational demonstration planned for fall 2026. Hivemind will allow LUCAS to "sense, decide, and act independently, without human intervention" with one operator commanding a swarm of autonomous systems. Hivemind is also embedded in Shield AI's V-BAT, in Kratos MQ-58 Valkyrie integrations, and in multiple CCA-adjacent prototyping efforts.
Anduril's Lattice operating system sits in a similar position one layer up. Lattice is the mesh networking and C2 backbone for Anduril's own family of systems and an expanding portfolio of third-party platforms. In March 2026, the Army awarded Anduril a $20 billion, 10-year enterprise contract that consolidated more than 120 procurement instruments into a single vehicle, effectively embedding Lattice as a permanent infrastructure layer across the force. In November 2024, the Space Systems Command awarded Anduril a $99.7M IDIQ contract to deliver Lattice as the mesh networking backbone for the modernized Space Surveillance Network, with full deployment mandated by end of 2026. Lattice is the C2 layer where a growing share of joint operations will live.
It's important to note that Shield AI and Anduril are only doing what good defense tech companies are supposed to do, which is to build software that the DoW wants to adopt at scale. The DoW has made rational choices to adopt them because the alternatives are slower, weaker, or do not exist.
However, the cumulative effect is a Pentagon that increasingly depends on a small number of private companies for the layers that matter most: the connectivity that ties uncrewed systems together (SpaceX), the autonomy that pilots them (Shield), and the C2 mesh that coordinates them (Anduril Industries). Each individual dependency is the product of competitive selection. The aggregate dependency is not something any single procurement decision created, and it is not something any single procurement decision will solve.
The LUCAS pricing dispute is what one of these dependencies looks like when the vendor decides to exercise it. There is no structural reason the same dynamic could not appear at the autonomy or C2 layers if commercial incentives ever push in that direction.
What Has to Move
The DoW cannot build a new Starlink, nor can it build its own Hivemind or Lattice versions (nor should it be trying to). Suggesting the DoW stand up an alternative LEO constellation in the near term, or that it should develop government-owned autonomy software to compete with Shield AI, ignores how those companies got where they are and what it would cost to replicate them. The alternatives that exist (OneWeb, Project Kuiper, multiple smaller autonomy startups, traditional primes building competing C2 stacks) are years away from matching the scale and capability the DoW depends on today.
What the DoW can do is start treating commercial dependency itself as a category of strategic risk, and structure procurement around it.
Contracting structures for short-duration use cases have to match the operational profile. Aviation-tier monthly fees for connections that exist for minutes are a structural mismatch, whereas per-sortie pricing for expendable platforms (negotiated before the next conflict) is a fix the DoW could ostensibly implement on its own.
Multi-vendor requirements need teeth. Congress restored Tranche 3 funding for the SDA Transport Layer because the multi-vendor architecture had value the Pentagon was about to write off. That principle has to apply across the stack, including in autonomy and C2. Should we be considering that the DoW should be paying to keep alternatives alive if they are technically inferior? The alternative to redundancy is the exact position the DoW found itself in during Operation Epic Fury.
Independence at the infrastructure layer has to mean something. If Starshield satellites share orbits, ground stations, or operational control with commercial Starlink, the dependency has been relabeled rather than reduced. If Hivemind integrations require recurring Shield AI support that can be priced or withdrawn, the dependency exists at the software layer instead of the connectivity layer. The DoW needs to negotiate data rights, source code escrow, and operational continuity provisions for every layer where it is exposed.
Some of these dependencies are real, durable, and not going away in the next decade. The DoW needs procurement, doctrine, and budget structures that account for that rather than pretending the next contract or the next congressional appropriation can buy out of it.
An Honest Conclusion
It is important to note that I am not making an argument against Starlink, against Hivemind, or against Lattice. All three are transformative capabilities and it would be wholly irrational for the DoW not to use them. The argument is that any military depending on a small number of irreplaceable commercial vendors for the layers that matter most needs to plan for the day a vendor's leadership decides their interests are not the same as the country's interests.
The Pentagon spent decades building redundant secure communications precisely because single-point-of-failure dependencies are how wars get lost on the wrong day. The transition to commercial LEO, commercial autonomy, and commercial C2 has reintroduced that vulnerability under new labels. Russia is publicly acknowledging the asymmetry on the satellite side. Ukraine has lived the consequences three times in three years. The DoW just paid a 400 percent connectivity surcharge in the middle of a strike campaign because it had no leverage to refuse.
The LUCAS pricing dispute is the warning shot, and the $2.29B SDN award is the response. Whether the response is a real fix or a more expensive version of the same dependency depends on how the next eighteen months unfold, and on whether the DoW's procurement system can learn faster than the vendors can extract.