Sunday raises $165 million at a $1.15 billion valuation to move its glove-trained Memo robot from demos into homes

Sunday has trained its household robot without a single trajectory of teleoperation data, and investors have now put a price of $1.15 billion on that approach. The Silicon Valley company said on 12 March 2026 that it had raised an oversubscribed $165 million Series B to take Memo, a two-armed robot that clears dining tables and loads dishwashers, out of staged demonstrations and into ordinary homes, starting with a beta programme this autumn.

The round makes Sunday one of a handful of robotics companies valued above $1 billion on a product that has not yet shipped. What the investors are paying for is a method of collecting training data as much as a machine. The questions that matter are how that method works, what the price implies when set against rivals that raised money in the same quarter, and which numbers will show by 2027 whether the approach holds up outside a lab.

Coatue led a round sized for deployment, not research

Coatue led the round and its partner Thomas Laffont joins Sunday's board, the company said in a post titled "No More Demos". Bain Capital Ventures, Fidelity Management & Research Company, Tiger Global, Benchmark, Conviction and Xtal Ventures also took part, alongside unnamed others. Sunday did not say whether $1.15 billion is a pre-money or post-money figure, and it gave no breakdown of how the money will be spent. It did say it had already grown its engineering team threefold, research fourfold and data operations fivefold to prepare for deployment. On the stated figures, the new money equals 14.3% of the valuation.

The product commitments are specific. Sunday said public deployment begins with a beta rolling out to users in the autumn and described Memo as the first autonomous home robot it intends to put in households this year. Its beta page sets the terms. Participation is free and invite-only, the first households receive a numbered limited-edition Memo with delivery and set-up by Sunday staff, and the first 50 members get a 3D-printed miniature. Owners will schedule tasks by voice or through a Sunday app, and the company said the beta will mix supervised, collaborative and fully hands-off operation. General availability depends on the beta; Sunday said it expects to ship Memo at scale in 2027-2028. It also disclosed a number most rivals keep private. Building one Memo by hand today costs about $20,000, and Sunday expects volume manufacturing to cut that by at least 50%, which implies a build cost of $10,000 or less. Retail pricing will come closer to launch.

Those terms describe a company that has yet to prove its robot can work unsupervised in a stranger's kitchen. The more consequential question is how it intends to teach the robot to do so.

The glove, not the robot, is the core asset

Sunday's method rests on the Skill Capture Glove, a wearable it says is a one-to-one copy of Memo's hand in geometry and sensor layout. The glove records motion through force sensors, cameras and joint potentiometers. Sunday recruits contributors it calls Memory Developers, who wear the gloves while doing chores in their own homes, and its website says the company has shipped thousands of gloves across the United States. A software layer the company calls Skill Transform then strips human-specific details from the recordings, aligning both the movement data and the camera images so that they look as if Memo had produced them. Sunday said in November 2025 that this conversion succeeds 90% of the time.

The economics explain the appeal. Teleoperation, the standard way to collect robot training data, needs one robot and one trained operator for every stream of demonstrations, usually in a lab. A glove costs far less than a robot and can be worn in thousands of real homes at once, so the data reflects cluttered counters, pets and odd floor plans rather than a test kitchen. Sunday says its cycle from capturing new data to a deployed capability takes less than a week.

The results Sunday published with its first model, ACT-1, on 19 November 2025 show what that data bought. In a task it calls Table-to-Dishwasher, the robot carried out 33 distinct and 68 total dexterous interactions with 21 objects, including wine glasses and ceramic plates, while travelling more than 130 feet. ACT-1 was also run in Airbnb rentals it had never seen, using a 3D map of each home to find the table and dishwasher, which Sunday described as the first single end-to-end model to combine long-horizon manipulation with map-conditioned navigation. The company also showed the robot pairing and balling socks and pulling an espresso shot. One disclosed figure tempers the demonstrations. Sunday says its contributors move at about half natural human speed and its models run at 60-80% of that pace, which by our arithmetic puts Memo at roughly 30-40% of the speed of a person doing the same chore.

Memo is designed to yield when people push it

According to Sunday's specifications, Memo stands 1.7 metres tall, weighs 170 pounds (about 77 kg), reaches 0.8 metres horizontally and 2.1 metres vertically, and has 27 degrees of freedom across two seven-joint arms, two four-joint hands, the torso and the lower body. It navigates at up to 1 metre per second and runs for four hours on a one-hour charge, with self-charging promised for the beta. Sunday contrasts Memo with humanoids that need power to stay balanced. It says Memo is passively stable, stays upright if power is cut in any pose and yields under compliant control when bumped. The company also says it does not need data from customers' homes to train the robot because its contributors supply it.

The founders' records explain why investors gave the method credit. Chief executive Tony Zhao, who left a Stanford PhD in 2024 after working under Sergey Levine and Chelsea Finn, led the ALOHA project and worked at Google DeepMind and Tesla Autopilot. Chief technology officer Cheng Chi, a Columbia PhD advised by Shuran Song, created Diffusion Policy. Sunday credits the two with ALOHA, Diffusion Policy and the UMI gripper, techniques that it says much of the field now builds on.

Bigger cheques went to rivals that disclosed less

Sunday raised less than the quarter's largest robot-AI rounds but disclosed more about its price. Skild AI raised $1.4 billion at a valuation above $14 billion on 14 January, Apptronik closed a $520 million extension on 11 February that took its Series A past $935 million, and Rivian spinout Mind Robotics raised $500 million on 11 March. Among home robots, 1X has priced its NEO humanoid at $20,000 for early access or $499 a month with priority deliveries in 2026, and Fauna Robotics launched its 107 cm Sprout developer humanoid on 27 January.

Sunday's round equals 14.3% of its stated valuation, against 10.0% for Skild and 2.6% for Figure's September 2025 Series C. In other words, investors bought a larger share of an earlier-stage company. In the ROBOTNESS database, the four US robotics rounds recorded for January to March 2026 total $2.585 billion, of which Sunday's accounts for 6.4%.