Cosmic Robotics Is Building Mars With a Solar-Farm Alibi

The YC Summer 2026 company has a 10,000-pound robot, tens of thousands of panel installs, and the audacity to treat Earth as a construction rehearsal.

August 05, 2026

Subtitle: The YC Summer 2026 company has a 10,000-pound robot, tens of thousands of panel installs, and the audacity to treat Earth as a construction rehearsal.

Company: Cosmic Robotics YC batch: Summer 2026 YC primary partner: Ankit Gupta Website:https://www.cosmicrobotics.com/YC profile:https://www.ycombinator.com/companies/cosmic-robotics

At sundown, a utility-scale solar farm becomes a black glass ocean. The panels run to the horizon in military rows, the dust turns orange, and every module still has to be lifted by somebody. Ninety pounds of glass and aluminum, again and again, under a sun with the bedside manner of a blowtorch.

Cosmic Robotics has brought a ten-thousand-pound machine to this punishment. Cosmic-1 rolls on eight wheels, drags its batteries and computers through the dirt, finds a panel with cameras and high-accuracy positioning, lifts it with a great articulated arm, and lays it onto the rack. The company says the robot has installed tens of thousands of panels on some of America’s largest solar projects and more than doubled the labor productivity of conventional crews.

That would be an adequate hard-tech story. Naturally, the founders have attached Mars to it.

Cosmic’s stated mission is to build a city on the red planet. Solar farms are the proving ground; data centers are the next terrestrial market; lunar construction work under a NASA contract is the bridge into space. Revenue and field data from Earth are supposed to train the machines that will eventually build power, compute, and shelter where there are no construction workers, unions, roadside diners, or convenient replacement parts.

It is a magnificent piece of startup theater because the prop is real, filthy, and heavy enough to crush the stage.

The product: a power tool that learns

Cosmic-1 is not pitched as a sealed appliance for one repetitive gesture. The company describes it as an autonomous heavy-lift manipulation platform: a mobile base, robotic arm, perception system, swappable end effectors, and software that can be trained for new construction tasks.

In solar, it picks and places modules weighing about 90 pounds with claimed millimeter precision, operating in heat, rain, snow, and dust while fitting into existing crew workflows. In a recent company demonstration, engineers swapped the tooling and trained the same platform to detect, lift, and install large cooling pipes for data centers in ten days. That is not proof of production reliability, but it reveals the larger wager. Cosmic does not want to sell a solar-panel robot. It wants to sell the general-purpose missing limb of construction.

The immediate customer is a utility-scale solar engineering, procurement, and construction contractor staring at labor shortages, injury risk, schedule pressure, and miles of repetitive lifting. The next target is the hyperscale data-center builder wrestling with heavy mechanical components and ravenous demand. The eventual customer is anyone assembling large infrastructure in places hostile to human labor—including, if the physics and budgets cooperate, the Moon and Mars.

The problem is brutally legible. Construction remains full of tasks too variable for factory automation and too punishing for humans. A robot has to survive an unstructured jobsite, recognize imperfectly staged materials, avoid workers, tolerate weather, recover from exceptions, and show up tomorrow. A factory arm lives in a monastery. A construction robot lives in a bar fight.

Two founders, two routes to the dirt

James Emerick, co-founder and CEO, grew up around construction sites. YC says he previously deployed autonomous earthmoving machines as a robotics engineer at Built Robotics, then worked at Autodesk Research on adaptive construction platforms. He holds bachelor’s and master’s degrees in mechanical engineering from the University of Pennsylvania. His useful heresy is that construction is not conquered by “disruption” rhetoric; crews adopt tools that earn their trust.

Lewis Jones, co-founder and CTO, arrives from the other end of the atmosphere. YC lists propulsion work at Relativity Space and earlier experience at SpaceX, NASA’s Jet Propulsion Laboratory, and NASA’s Johnson Space Center. He studied mechanical engineering and economics at Caltech and was named to Forbes’ 30 Under 30. The company says Jones went looking for the robots that would build a Martian city, discovered an awkward vacancy, and helped start Cosmic.

YC lists these founder LinkedIn profiles:

Twitter/X links were not listed for either founder in the YC company bio. YC lists Ankit Gupta as the company’s primary partner.

The company was founded in 2024, is based in San Francisco, and YC lists a team of eight. TechCrunch reported in April 2025 that Cosmic raised a $4 million pre-seed round led by Giant Ventures, with HCVC, MaC Ventures, and angel investors participating.

Traction, with the usual warning label

The strongest public signal is physical deployment. YC and Cosmic say the machine has completed tens of thousands of installation cycles, installed panels on major U.S. solar projects, and more than doubled crew productivity. A company LinkedIn post says it has installed megawatts of solar on live projects. These are company claims, not independently audited operating data, but they are more substantial than a robot waving delicately at a trade-show booth.

There is some third-party corroboration around the work itself. The U.S. Department of Energy selected Cosmic Robotics as a semifinalist in Round 8 of the American-Made Solar Prize, awarding $50,000 and describing its system as robots that pick and place modules on single-axis trackers alongside human crews. Recent company announcements say Cosmic later won the Set! phase. KUKA has publicly featured the system working with solar installer Sunstall. YC says Cosmic is also building lunar construction robotics under contract with NASA, though public materials reviewed for this profile did not disclose the contract value or program details.

What remains private matters just as much: revenue, paid fleet size, unit economics, uptime, maintenance burden, customer concentration, injury record, and the duration of the productivity gains. “Tens of thousands” sounds large until a national infrastructure market begins counting in millions.

The rivals are already in the field

Solar automation is not virgin territory. Built Robotics has pursued autonomous heavy equipment and solar piling. Terabase Energy markets a digitally managed construction system with automation for utility solar. Luminous Robotics, Planted Solar, AES’s Maximo, and other specialists attack module handling, layout, installation, or site logistics from different directions. Traditional construction-equipment giants have distribution, service networks, financing, and a century of institutional scar tissue.

Cosmic’s proposed advantage is a machine that can migrate across tasks and markets. If the same expensive hardware installs solar modules this quarter, data-center pipe next quarter, and lunar infrastructure someday, utilization improves and the learning system compounds. If every new task demands new tooling, safety validation, integration, and months of field babysitting, “general purpose” becomes a very costly euphemism for consulting.

That is the central technical risk. Demonstrations reward a clean success. Jobsites charge rent for every edge case. Dust blinds sensors. Mud steals traction. Glass flexes. Components arrive bent. Workers improvise. Network coverage vanishes. A machine that doubles productivity on the happy path can still wreck the schedule if a specialist technician must fly in whenever reality gets creative.

There are business risks too. Hardware consumes capital. Construction customers buy cautiously and punish downtime. Incumbents can copy features or bundle service. A NASA contract makes excellent narrative fuel but can distract a young company from the earthly market paying this year’s invoices. Mars is a useful north star; it is a terrible quarterly customer.

Why this matters

The AI boom is colliding with the stubborn mass of the world. Models can generate a data-center design in seconds; somebody still has to install the cooling pipe. Cheap solar modules do not become a power plant until workers lift, align, fasten, wire, inspect, and repeat. Ambition is abundant. Skilled bodies and construction hours are not.

Cosmic’s best insight is that space robotics may need to be born as industrial equipment. A machine trained only in a pristine aerospace lab will learn the manners of a diplomat. A machine dragged through heat, freezing rain, mud, broken pallets, and real project deadlines may acquire the temperament required for another planet.

The skeptical version is equally clean: the road to Mars is littered with Earth companies that used the heavens to make an ordinary product glow. Cosmic must prove that its robot is not a grand science-fiction wrapper around a narrow solar installer—and that general-purpose autonomy survives contact with invoices.

Still, there is something bracing about the sequence. Do not begin with a humanoid folding towels in a laboratory. Begin with a giant wheeled arm under the murderous sun, lifting the same heavy slab ten thousand times without asking for a motivational seminar. Make it useful in the dirt. Make it pay. Then point it at the Moon.

The future, Cosmic insists, is not imagined. It is built. For once, a startup has arrived with grease on the sentence.

Sources

Editorial note: Deployment totals, productivity improvements, environmental performance, data-center transfer speed, and NASA work are attributed to Cosmic/YC unless otherwise stated. Public sources reviewed did not provide independent audited performance data.

Hero image

/Users/michaeljames/Documents/Codex/2026-05-23/twice-weekly-research-one-startup-from/assets/cosmic-robotics-cyberpunk-hero.png

Original image generated with the built-in image-generation tool. No company logos, protected marks, or recognizable branded machine designs were requested.