Skild AI raises $1.4 billion in SoftBank-led Series C, lifting its valuation past $14 billion on a bet that one model can run any robot
Skild AI said on 14 January 2026 it had closed a $1.4 billion Series C led by SoftBank Group, valuing the company at more than $14 billion. NVIDIA's venture arm, Macquarie Capital and Bezos Expeditions joined, alongside new strategic backers LG, Schneider Electric, CommonSpirit and Salesforce Ventures. The start-up, founded in 2023, said its Skild Brain software went from no revenue to about $30 million within a few months of 2025.
Skild AI, the company building a single robot-control model meant to work across many kinds of machines, said on 14 January 2026 that it had raised $1.4 billion in a Series C round led by SoftBank Group. The company put its valuation at more than $14 billion, a price that sets a public benchmark for companies selling robot intelligence rather than robots.
The size of the step is the headline for investors. In July 2024 Skild announced a $300 million Series A at a $1.5 billion valuation. Measured against that round, the new price is more than nine times higher after roughly eighteen months, and the Series C alone is worth close to a tenth of the stated valuation. Skild did not publish the share class terms, the split between primary and secondary capital, or whether the $14 billion figure is pre or post money.
The investor list mixes financial and industrial money. Besides SoftBank, Skild named NVentures, the venture unit of NVIDIA, together with Macquarie Capital and Jeff Bezos through Bezos Expeditions. Existing backers Lightspeed, Felicis, Coatue and Sequoia Capital put in more. New strategic investors were LG, Schneider Electric, the US hospital operator CommonSpirit and Salesforce Ventures, and the round also drew Disruptive, 1789 Capital, IQT, TF Capital, Andra Capital, Palo Alto Growth Capital, Alpha Square, Mirae Asset and Destiny.
Skild describes its goal as omni-bodied intelligence. The company says the Skild Brain is a unified foundation model that can control quadrupeds, humanoids, tabletop arms and mobile manipulators without having been built around any one of them, and that it generalises across both tasks and hardware. That positioning separates Skild from humanoid makers that train models for their own bodies, such as Figure AI, Apptronik or 1X Technologies.
The technical claim rests on data. Skild says it feeds its model from four streams, which are large-scale simulation that produces synthetic experience, internet video of people doing tasks, teleoperation sessions that link camera images to joint movements, and data returned from robots working at customer sites. In plain terms, simulation and human video give breadth cheaply, while teleoperation and field robots supply the scarce, physically accurate examples that tell a model how a specific motor should move.
According to the company, the Series C will be used to scale each of those data sources sharply and to fund basic research on model architectures, learning algorithms and data collection methods. Skild did not give a headcount target, a compute budget or a timeline for new model releases in the announcement.
Commercial proof was thin but not absent. Skild said its revenue climbed from zero to about $30 million over a few months of 2025, from deployments in security, construction, delivery, data centres, warehouses and factory assembly. It did not name customers or say how much of that figure is recurring, hardware related or tied to pilots.
- Skild AI
- Series C
- 2026-01-14
- 1400
- 14
- No
- Skild AI
- Series A
- 2024-07-09
- 300
- 1.5
- No
- Figure AI
- Series C
- 2025-09-16
- 1000
- 39
- Yes
- Apptronik
- Series A-X extension
- 2026-02-11
- 520
- No data
- Yes
- 1X Technologies
- Series B
- 2024-01-13
- 100
- No data
- Yes
- Physical Intelligence
- Series B
- 2025-11-20
- No data
- No data
- No
Amounts and valuations as disclosed by each company or lead investor; null where not disclosed. Skild's Series C valuation was given as more than $14 billion and is shown at that lower bound. Valuation step-up = 14 / 1.5 = about 9.3 times. Revenue multiple in text = 14,000 / 30 (USD m) = about 467 times.
As of Oct 1, 2026
The comparison with other heavily funded robot companies shows how investors are pricing software. Figure AI announced a $1 billion Series C at a $39 billion post-money valuation in September 2025, but Figure builds its own humanoid and factory. Apptronik closed a $520 million Series A extension on 11 February 2026, taking its Series A to more than $935 million at three times the earlier Series A price. Physical Intelligence, Skild's closest software rival, closed a CapitalG-led Series B in November 2025 without the lead investor disclosing terms. Skild's round therefore sets the clearest public price yet for a hardware-agnostic robot model.
For the wider industry the deal matters because it pulls hardware makers, chip suppliers and end users into one cap table. NVIDIA supplies the compute that simulation-heavy training depends on, LG makes appliances and has its own robot ambitions, Schneider Electric automates factories and buildings, and CommonSpirit operates hospitals and care sites. Each can become a deployment site and a data source, which is the asset Skild says it most needs.
The risks are plain. A valuation near 470 times 2025 revenue leaves no room for a slow ramp, and Skild has not shown independent benchmarks proving that one model outperforms embodiment-specific systems on safety or reliability. Robot makers that build their own models may treat a horizontal supplier as a competitor rather than a vendor, and the company still has to prove its software works on hardware it does not control.
ROBOTNESS analysis
Skild's Series C turns the robot brain into a stand-alone asset class, but the price assumes that robot makers will license intelligence rather than build it.
The evidence for the thesis is the cap table. SoftBank, NVIDIA and a dozen financial investors paid more than nine times the 2024 price for a company whose product is a model rather than a robot, and industrial groups such as LG and Schneider Electric chose to own a piece of the software layer rather than wait for it to mature.
The strongest counter-argument is that vertical integration keeps winning in robotics so far. Figure, Tesla and the leading Chinese humanoid makers train models on their own fleets, and they have little reason to hand control and data to an outside supplier valued at a fraction of Figure's price.
Bull case. If Skild's revenue keeps compounding from the roughly $30 million it reported for 2025 and the Brain runs on many vendors' arms and quadrupeds, Skild becomes the default operating layer for smaller makers that cannot afford their own model teams. Fresh capital also buys the simulation compute and teleoperation data that late entrants will struggle to match.
Bear case. Foundation models for robots could commoditise as open models from chip suppliers and research labs improve, squeezing licence prices. A valuation set at more than 460 times trailing revenue would then need a long period of growth to justify, and later investors could demand structure or a lower price.
- First half of 2026: whether Skild names paying customers or hardware partners beyond the sectors listed in January.
- May 2026: SoftBank Group's full-year results for any disclosure on the size and accounting of its Skild stake.
- Through 2026: rival rounds at Physical Intelligence and humanoid makers, which will show whether the $14 billion price becomes a benchmark or an outlier.
- Series C, $1.4 billion
- 2023
- 14 January 2026
- More than $14 billion
- About $30 million, up from zero within a few months (company figure)
- SoftBank Group
- NVentures (NVIDIA), Macquarie Capital, Bezos Expeditions; Lightspeed, Felicis, Coatue, Sequoia Capital increased stakes
- Series A, $300 million at $1.5 billion (July 2024)
- LG, Schneider Electric, CommonSpirit, Salesforce Ventures
Why it matters
The round puts a public number on a question the industry has argued about since 2023, namely whether robot intelligence can be sold separately from robot hardware. A price above $14 billion for a model developer that reported about $30 million of 2025 revenue says that the largest technology investors believe the answer is yes, and that the value will accrue to whoever controls the training data and the model rather than to whoever bends the metal.
It also matters for capital allocation across the sector. A single $1.4 billion cheque raises the bar for rival model developers that now need comparable sums to buy compute and data. Smaller arm, quadruped and mobile-robot makers gain an option to license a capable model rather than build one, which could speed deployment at the lower end of the market.
Rival analysis
Physical Intelligence is the closest analogue, a model-first company whose CapitalG-led Series B in November 2025 came without disclosed terms. Both sell the idea of a general robot policy, but Skild stresses control across very different bodies, from quadrupeds to humanoids, while Physical Intelligence has emphasised manipulation. Google DeepMind and NVIDIA are the platform-scale rivals, since both publish robot models that hardware makers can adopt with little or no licence cost.
Vertically integrated humanoid makers are the other front. Figure AI, priced at $39 billion post money in September 2025, trains its own Helix model on its own robots; Apptronik and 1X follow similar paths. They are potential customers only if their in-house models fall behind, which makes Skild's growth partly dependent on rivals' engineering failures.
Valuation context
Using company figures, Skild's valuation rose from $1.5 billion in July 2024 to more than $14 billion in January 2026, a step-up of about 9.3 times. Against roughly $30 million of 2025 revenue the implied multiple is about 467 times, far above what listed industrial automation groups trade on and closer to the pricing of frontier AI model companies.
The round size of $1.4 billion is about 10 percent of the stated valuation, a conventional dilution for a late-stage private round. Because Skild did not say whether the valuation is pre or post money, or disclose preferences, the economic price for new investors may be lower than the headline suggests.
Supply-chain implications
Skild's main inputs are compute and data rather than parts. NVIDIA's participation through NVentures aligns the company with the GPU supplier whose simulation and training stack most robot developers use, and simulation is one of the four data streams Skild names. Teleoperation rigs and field robots are the costly inputs, and the strategic investors give access to sites where those robots can work.
On the hardware side Skild depends on third-party arms, quadrupeds and humanoids, whose actuators, reducers and batteries come from global supply chains with heavy Asian content. That leaves its deployments exposed to the same component pricing and export-control risks as the robot makers it serves, even though it does not carry inventory itself.
Signals to watch
The first signal is customer disclosure. Skild named sectors but no customers in January, and named contracts with logistics, construction or data-centre operators would show whether the $30 million is repeatable revenue.
The second is partner hardware. Announcements that LG, Schneider Electric or other investors are shipping machines with the Skild Brain would test the omni-bodied claim outside Skild's own labs. The third is follow-on pricing at rivals such as Physical Intelligence, which will show whether investors treat $14 billion as a reference point.
Analyst view
Thesis: Skild has bought itself the scale to be one of two or three independent robot model suppliers, but the valuation prices in a licensing market that does not yet exist at size. Confidence medium. The investor roster, the revenue ramp and the stated data strategy are verified and coherent; the lack of named customers, independent benchmarks and deal terms limits conviction.
We would upgrade confidence on evidence that third-party hardware makers pay for the model on production fleets, and downgrade it if the leading humanoid makers keep models in house while open models from NVIDIA and Google close the performance gap.
Questions you should be asking
How much of the roughly $30 million of 2025 revenue is recurring software income, and how much comes from pilots or hardware pass-through?
Is the more than $14 billion valuation pre or post money, and did the round include secondary sales or structured terms?
Which robot platforms run the Skild Brain in paid deployments today, and what are the measured success and intervention rates?