Startups Build the Missing Logistics Layers for Orbit and Construction Sites

Launch is no longer the only bottleneck in space logistics. Alex Wilhelm framed the market as increasingly crowded for getting mass from Earth into an initial orbit, citing SpaceX, Rocket Lab, Blue Origin, Russia and others in the launch category. The harder question is what happens after a payload reaches space and still needs to move.

That is the market Tom Mueller says Impulse Space is building for: spacecraft that move, host, deploy, rendezvous with, or deliver payloads after launch. Impulse recently raised a $500 million Series D, bringing total capital raised above $1 billion. Mueller said the company has been oversubscribed in every round since seed, that insiders wanted to participate, and that the company could have waited but chose to raise while the market was hot. Asked whether Impulse is not capital constrained, he answered, “That would be true,” and allowed that even that might be an understatement.

The company’s first product, Mira, is the smaller and more flexible vehicle in the portfolio. Mueller described it as a highly propulsive spacecraft with storable rather than cryogenic propellants, meaning it can remain on orbit for years. It can point precisely, host payloads on board while moving them through experiments, deploy payloads such as CubeSats, and perform rendezvous and proximity operations. Impulse has flown three spacecraft so far, and Mueller said all three deployed CubeSats. Last year, using a Starfish Space kit on board, one Impulse spacecraft was commanded to approach another; the two came within 1,200 meters.

Mueller’s original concept for Mira was straightforward: if a launch vehicle is a container ship arriving at port, the space economy still needs trucks and vans to distribute cargo after delivery. Mira was meant to ride on missions such as SpaceX Transporter and, eventually, Starship, then move payloads to more specific orbits. Wilhelm called it a pickup truck: functional, affordable, and able to move around within its domain. The accompanying Impulse imagery reinforced that idea, showing an orbital vehicle maneuvering near another spacecraft and a gold foil-wrapped spacecraft carrying the line, “How’s my orbital maneuvering? Go to impulse.space.”

The commercial response, however, was softer than Mueller initially expected. Impulse tells customers that if they do not like where Transporter drops them off, Mira can move them to a different spot more cheaply than buying a dedicated launch, such as on Rocket Lab. But “most people were just happy where Transporter went,” he said. The product has still had commercial customers on each flown mission, but its stronger demand has come from government users.

Space Force, Mueller said, “really liked it.” Impulse has more than half a dozen Miras in production, mostly for government work. Some are a newer version capable of operating at geosynchronous orbit, about 22,000 miles out, where the radiation environment is much harsher. Wilhelm initially thought that meant Mira could move payloads from low Earth orbit to geosynchronous orbit. Mueller corrected him: Mira can move within low Earth orbit, for example from roughly 300 kilometers to 1,200 kilometers, “probably up and back,” but going from LEO to GEO requires more than four kilometers per second of Delta VChange in velocity; in orbital mechanics, this is the budget that determines how much maneuvering a spacecraft can do..

That distinction is central to Impulse’s second product. Helios is not a pickup truck. It is the long-haul system.

Helios is Mueller’s answer to the energy gap between low Earth orbit and higher-energy destinations. He called it “a rocket on top of a rocket”: a third stage added to a launch vehicle, initially Falcon 9. It uses a large aluminum tank holding 12 tons of liquid oxygen and liquid methane, and a pump-fed Deneb engine producing 15,000 pounds of thrust. Mueller said the engine is likely to be the highest-performing hydrocarbon engine ever built, with very high specific impulse — “gas mileage for rockets,” as he put it.

Helios is designed to be dropped off by Falcon 9 at around 200 kilometers, then perform two burns. In about eight hours, Mueller said, the payload can reach geosynchronous orbit at 22,000 miles. The first flights are planned for 2027.

The original use case was the commercial GEO satellite market: take a four-ton satellite, which Mueller described as the GEO “sweet spot,” directly from LEO to GEO on Falcon 9. Today, many commercial GEO satellites launch on Falcon 9 into a transfer orbit and then spend months raising themselves with electric propulsion. Electric propulsion is efficient but low-thrust; satellites are effectively collecting solar energy over months to create enough impulse. Helios instead burns all of its propellant in about 15 minutes across two burns.

Mueller said Helios can increase payload to the Moon on Falcon 9 by a factor of about four, and in a Mars case Impulse analyzed, increase payload by about five. The stage itself is priced at $25 million, while he put launch-vehicle cost at roughly $70 million to $100 million. From his perspective, adding $25 million to get multiple times the payload is why he calls Helios a compelling product.

Wilhelm compared the concept to a “Falcon 9 moment” for post-launch movement: the improvement that comes not from changing the first trip to orbit, but from making movement after orbit materially cheaper and faster. Mueller said Helios is already selling well “for something we haven’t flown yet.”

That last qualifier matters. Asked how confident he is that Helios works the first time, Mueller did not overstate it. “This is a premium product,” he said, and Impulse is taking its time because “it’s really gotta work.” But he added the standard aerospace caveat: “this is rockets,” and rockets are hard. If Helios does not light in space, he said, “that’s bad.”

SES is among the Helios customers Impulse has announced. Mueller described SES as one of the largest communications satellite companies in the world, and the mission fits the original design case: a large, single four-ton-class payload. Impulse has also sold what it calls “caravans,” or rideshare missions, carrying multiple payloads. Astranis is among the announced customers.

The rideshare product exists because small payloads rarely get direct access to GEO. Rideshares to GEO have historically appeared only occasionally, often when a very large satellite on a Falcon Heavy still leaves some unused capacity. That can allow a 300-kilogram payload to tag along, but such opportunities may come only every few years. Helios changes the cadence by creating, in Mueller’s words, a “highway to GEO.” The first rideshare mission is already sold out, and Impulse is filling the second.

Mueller estimated that around 20 things go to GEO in a year today. Impulse might capture half of that with a mix of rideshare and large communications satellites, potentially approaching a once-a-month cadence, though he said it will take time to get there. When Wilhelm called 20 GEO-bound objects a year “pathetic” for humanity, Mueller pointed out that GEO is different from low Earth orbit: three GEO satellites can cover the whole Earth, there are only so many useful orbital slots around the equator, and satellites cannot be packed too closely without interfering with one another’s signals.

Helios is designed as expendable for now. Its liquid oxygen and liquid methane are cryogenic, and Mueller said they will not store very long on the stage. There is no fundamental reason Helios could not be reused, but reuse would require propellant depots in both low Earth orbit and geosynchronous orbit.

The concept would be simple: fill Helios in LEO, take a payload to GEO, partially refuel at GEO, return empty, and cycle the vehicle again. The infrastructure is the missing piece. Mueller said propellant depots are probably five to 10 years away. SpaceX will need tanker operations to go to the Moon and do what it wants to do, he said, and he hopes that eventually leads to depots. He also said he has spoken with other companies working on depots and told them Impulse would buy propellant.

Asked whether Impulse wants merely to build the trucks, or whether it wants to build the full logistics network — depots, coordination, and orbital management — Mueller said he has not thought too much about gas stations in space and does not think that will be Impulse, though “it could be.” The company’s logistics ambitions more clearly include landers. Impulse began a Mars lander effort with Relativity and is trying to enter NASA’s CLPS lunar lander program. Mueller said the company is excited about supplying a Moon base and eventually bringing materials back as part of its in-space logistics strategy.

Helios can change lunar cargo economics, but only within limits. Mueller said it can increase payload on Falcon 9 from about 200 kilograms to about 1,000 kilograms. On Falcon Heavy, it can get “like three or more tons” to the Moon. That is enough for cargo such as rovers and supplies. Human landers or larger lunar infrastructure would require a bigger rocket and a much larger Helios variant.

Mueller called that concept “Mega Helios.” It is not a current program and Impulse is not working with SpaceX on it, but Impulse has studied the exercise. A third stage for Starship might use something like 50 tons of propellant and six Deneb engines, and could take roughly 30 tons to the Moon. Mueller emphasized that these were rough numbers and speculative, but used the example to illustrate what might become possible with a larger in-space stage.

The politics of government payloads are another constraint. Mira has clear Space Force demand, but Helios has to work through NSSL, the National Security Space Launch lanes, to fly government missions. Mueller described that process as highly contested and politically charged, the same system SpaceX sued to enter years ago and that newer launch companies now want access to. Helios can do for government what it does for commercial customers, he said, and he is confident Impulse will find a way to fly government payloads because the U.S. government is the world’s largest customer.

Mueller’s lunar interest is not primarily nostalgia or science fiction. Tom Mueller said the Moon is more important in the near term than Mars because it has what humanity would need to build megastructures in space. The first category of megastructure he expects is AI data servers.

His thesis is that compute eventually needs to move to space. Power demand for compute, he said, has grown at least 15% per year, a steep exponent that becomes crushing over decades. In Mueller’s framing, space offers the power environment that could support that growth. He said he has been talking about this idea for a long time, adding narrowly that “Elon’s already doing it.”

The second part of the thesis is materials. Mueller said it takes about 20 times less energy to bring material from the surface of the Moon to low Earth orbit than to bring material from Earth’s surface 200 miles up to LEO. He argued that the economics of building large structures in orbit eventually shift toward lunar resources, additive manufacturing, and concepts such as mass drivers on the Moon.

Impulse does not require Starship or other new heavy launch systems to work. Mueller said the company works well with Falcon 9. But larger launch vehicles could lower its costs. Starship, he said, could carry three Helios stages at a time. Even if Starship were sold at Falcon 9 pricing, that would reduce Impulse’s launch cost per Helios by one-third.

That scalability depends on Impulse’s manufacturing strategy. Mueller said Impulse is “extremely vertically integrated,” a SpaceX-like operating philosophy that consumes capital but gives the company control over cost, schedule, and quality. Asked where the billion dollars goes, he said mostly employees and vertical integration.

His critique of legacy space vendors was blunt. In his view, cost-plus contracting and prime-contractor incentives have created a “space industrial complex” that is expensive and slow. He recalled a saying from SpaceX: if it takes a month to get a quote for a part, you do not want the part. The quote cycle itself previews the likely price and schedule.

Impulse has passed 500 employees after more than doubling in the past year, and has open requisitions for about 200 more. Mueller said recruiting has become harder because there are more space startups, and experienced people from SpaceX and elsewhere increasingly start their own companies and recruit their colleagues. He called the result the “SpaceX mafia,” visible in El Segundo and the surrounding area, with space, robotics, and other companies spinning out of that talent base. A SpaceX IPO, he suggested, could further supercharge that movement.

Orbital congestion, meanwhile, is not something Mueller dismisses as a century-away issue. He said it will require more active management sooner than that. Large LEO constellation operators, in his view, are generally good stewards: maneuvering satellites and bringing them down when they are done. But he said some players are leaving stages up and polluting space, which he called “really bad.”

Tessa Lau said Dusty Robotics has built “the most accurate mobile robot on the planet,” but its purpose is not to look like a worker. Its robot prints floor plans directly onto construction sites.

Construction layout is still often manual. Lau described the current process as someone taking printed blueprints into the field, then using measuring tape and string on hands and knees to mark where everything goes inside a building. Dusty automates that process with what she called “an inkjet on wheels”: a mobile robot that takes the digital blueprint and prints it on the floor.

The robot is not only printing walls. A building’s documents contain separate sheets for plumbers, electricians, framers, and other trades. Dusty consolidates those drawings and prints them in one pass, so every trade can see not only its own work, but what everyone else is supposed to build around it. Lau said that shared visibility eliminates conflicts during construction.

The FieldPrinter 2 was shown driving on a concrete floor and printing lines and text. The on-screen claims highlighted three capabilities: “Print text clearly and efficiently,” “Print curves with precision,” and “Guarantee 1/16" accuracy.” That last claim is central to why the product can replace layout work rather than merely annotate it.

The speed improvement is the core business case. Lau said Dusty is tracking at roughly 10 to 17 times the speed of a manual layout crew. A job that would normally take several weeks can often be done in about a day. On large data centers, manual layout can take months. Crews are usually two to four people, she said, partly because string-based layout requires someone on either end. Dusty can reduce that work to days.

The product works from the industry’s standard design tools. Lau said Autodesk Revit is the de facto standard, with AutoCAD also widely used; Dusty supports both. Its process begins with plugins in Revit and AutoCAD that extract information, preprocess it, and upload it to Dusty’s portal. That portal is where the trade designs come together for additional checks and field verification before the robot prints.

Lau distinguished between design coordination and field coordination. General contractors already run BIM coordination processes to overlay models and resolve basic conflicts, such as ductwork interfering with conduit. But a coordinated model is not necessarily constructible in the field. Field coordination means checking whether the design matches existing conditions and can actually be built. Dusty’s portal, she said, shifts that work left by bringing more of it online before the field print.

The second-generation FieldPrinter incorporates learnings from field deployments and customer feedback about what works across job-site conditions. Wilhelm raised a possible launch date for the second generation, and Lau said it should be checked rather than accepting it. Across all generations, Dusty has completed more than 330 million square feet of layout, and Lau said the number is accelerating.

Dusty’s sweet spot is large, complex construction: data centers, hospitals, and similar projects. Tessa Lau said the company has worked on everything down to single-family homes, but large floor plates with dense coordination needs are where the value is clearest. Data center construction is driving demand because Dusty can help projects finish with less time and less money.

Lau said some of the biggest names in data centers are starting to mandate Dusty. That means the requirement is pushed down to whichever general contractor is building the facility, and those contractors then come to Dusty because they have to use it. Asked whether the motive is speed or accuracy, Lau collapsed the distinction: saving time saves money, eliminating rework saves time, and therefore also saves money.

Dusty’s business model is a robot subscription. Customers rent the device in a structure Lau compared to leasing a car. The subscription includes the equipment and a software license that allows use of the full system across as many projects as the customer wants.

There is also an implementation and training layer. Dusty trains the software users and the field operators separately because they are often different groups. Customers can become Dusty certified through that process. Lau also described Dusty Academy, a self-serve training system intended to remove adoption friction. Some customers watch the videos and are largely trained by the time Dusty arrives; in some cases, she said, Dusty can ship the system and customers can use it out of the box.

That adoption strategy matters because Dusty is selling into an industry often caricatured as technology-resistant. Lau said there are many tech-savvy people in construction, especially younger workers accustomed to smartphones and video games. The robot is operated with a joystick-like interface, making the learning curve closer to a familiar controller than a specialized industrial machine.

The more important trust-builder is accuracy in the field. Lau described an early hospital job in San Jose during COVID. A skeptical superintendent, masked and distrustful of technology, began checking the robot’s printed lines with a tape measure within 30 seconds. Within five minutes, she said, he pulled down his mask to show a wide grin.

The company’s fleet has grown to about 150 deployed robots across the U.S. and Canada, up more than 100% from the prior year by Lau’s account. Dusty plans to have around 200 by the end of the year and more by the middle of the following year. Utilization per robot has been roughly constant, but Dusty is building features meant to increase where and how often the robots can be used.

Dusty’s original operating assumptions fit large, open projects: data centers, hospitals, and other spaces where the robot can be set loose across a broad floor plate. Smaller jobs are harder. Tessa Lau gave the example of a customer renovating one operating room at a time. An operating room may contain dense mechanical, electrical, and medical infrastructure, making layout valuable, but the space is too small and constrained for Dusty’s earlier assumptions.

The company’s answer is a feature called flexible control. Traditionally, Dusty depends on surveyed control points: precise reference points around the perimeter of a building, placed by surveyors using benchmarks or monuments. Lau compared them to registration markers for a standard printer. In large construction projects, those points allow the robot to locate itself accurately within the building coordinate system.

But bringing a surveyor into a single operating room to establish control points is impractical. Flexible control allows Dusty to operate without those surveyed points by scanning existing conditions — walls, columns, and other features — and aligning to them. That opens use cases where the robot can still print accurately in smaller or more constrained spaces.

Lau said 99% of construction layout happens on the floor before it is transferred elsewhere. Even wall outlets are first marked on the floor with annotations such as an outlet location 18 inches above finished floor. Overhead work is handled similarly: Dusty marks an X on the ground, and a plumb laser transfers the point upward. That is why a floor-printing robot can affect walls and ceilings without itself climbing them.

Supply chain has not been a current blocker, according to Lau, because Dusty learned during COVID’s disruptions to stock long-lead parts. When the company needs to build, it keeps the required parts on hand.

The long-term opportunity, in Lau’s view, is not only printing the layout. Layout sits at the center of construction. All design and pre-construction work funnels through it, and all execution depends on it. If layout is wrong, downstream work goes wrong. Because Dusty started at that linchpin, she said, the company has permission to move both upstream into better design and coordination tools, and downstream into instructions for future job-site robots.

Those instructions may be literal. Dusty prints QR codes as part of the layout. Lau described them as machine-readable instructions for future robots: where they are and what to do. Wilhelm called Dusty the scribe for future construction robots. Lau preferred another term.

Unlike many uses of “orchestration,” Wilhelm said this one made sense. Dusty sits where design intent becomes field instruction. If more robots enter construction, Dusty wants its printed layout to become the shared reference layer they build from.

Dusty’s upstream opportunity is field coordination: reconciling coordinated design with the physical conditions and human constraints of a real site. Tessa Lau said even when design coordination is complete and the digital model looks like the future hospital, “the moment that design hits the field, everything goes sideways.” Columns are not exactly where expected. Concrete can be an inch short. Existing conditions diverge from the model.

Constructibility is a separate problem. Designers may never have spent time in the field or tried to build the thing they drew. Lau gave the example of a staircase that looked perfect on screen. The tolerances worked. The available space appeared sufficient. But the superintendent rejected it because workers could not physically get their hands between the wall and staircase to install screws. The model had enough geometric clearance, but not enough buildable clearance.

Dusty is not trying to reinvent every sensing tool required for that process. Lau said scanning tools already exist. Instead, Dusty now has a service partner network that bundles scanning, coordination, and Dusty layout into a single service. For customers, that means they do not need to buy or learn multiple tools. For partners, Lau said, the combined service can be priced significantly higher than layout alone. Asked whether partner economics are attractive for Dusty, Lau indicated that the bundling is a net positive in distribution and revenue terms.

The larger capital market has not necessarily rewarded that kind of practical robotics as loudly as humanoids or AI. Wilhelm contrasted Dusty’s publicly announced roughly $70 million in funding with the much larger sums being directed toward humanoid robot companies. Lau said venture capital is faddish and currently chasing AI returns. Dusty, by contrast, is “solidly building a company” with a useful product deployed nationwide and beginning to go international. She expects the “physical AI” part of the world to wake up over time.

Lau’s ambition is to take Dusty public. She said she does not know when, but believes the company is “well on our way.” The readiness case she described is not a single metric but an operating position: Dusty has figured out go-to-market, knows how to sell, has built a recognizable construction-tech brand, and has strong market momentum.

The point of comparison with Impulse is not that space tugs and layout robots share a customer or a technology stack. It is that both companies are selling enabling infrastructure after the headline platform has done its visible work. Launch gets mass to orbit; Impulse argues that customers still need a logistics layer to put it where it becomes useful. Architects and engineers create the model; Dusty argues that builders still need a field coordination layer that turns that model into reliable instructions on the job site. In both cases, the business is being built around the unglamorous handoff where expensive work can otherwise stall, drift, or have to be redone.