Skild AI open-sources its physics engine, up to 13 times faster than Nvidia's PhysX, in Nvidia's Newton
Steel lands inside NVIDIA's open-source Newton engine and makes fast, mixed-robot simulation free for every rival, including NVIDIA's own GR00T 2 team. What Skild keeps, the brain and its factory deployments, carries a $14 billion valuation on revenue no filing has ever shown.
Richard Tangthree-minute read
On November 9 at the Conference on Robot Learning (CoRL) in Austin, Skild AI$16.93B — Skild AI, private, latest valuation $16.93B put its Steel physics solver into NVIDIA's open-source Newton engine under an Apache license, the same day NVIDIA launched its own robot brain, GR00T 2 123. Skild measures Steel at up to thirteen times PhysX's throughput on mixed robot batches, a simulation figure, not training speed 1. The brain Skild keeps is valued at $14 billion, on revenue no filing has shown 4.
The speedup is biggest exactly where general brains train
GPU physics engines earn their speed by replicating one scene thousands of times 1. A general-purpose brain needs the opposite: arms, dexterous hands, humanoids and quadrupeds practising different tasks in one batch, and engines waste work padding unlike scenes into a common shape 1. Steel sizes work per environment and groups similar mechanisms, so its edge is narrow on identical tasks and widens as batches mix 1. Skild's own account of its data engine rates simulation as cheap to scale but thin in diversity; Steel is built for exactly that cell 17.
Steel's lead over PhysX widens as the training batch mixes more kinds of work
The advantage is heterogeneity, exactly the workload a general-purpose brain needs
Data
| Steel vs PhysX | |
|---|---|
| Identical tasks (geometric mean) | 3.8x PhysX |
| 5 task types | 9.2x PhysX |
| 10 task types | 10.4x PhysX |
| 15 task types | 12.5x PhysX |
| 20 task types | 13x PhysX |
NVIDIA holds every seat at this table
Steel was built in collaboration with NVIDIA−4.65% — NVIDIA, down 4.65 percent today 1. NVentures, NVIDIA's venture arm, owns Skild stock alongside SoftBank, Samsung and LG 6. Newton and Isaac Lab host Skild's training and now ship its solver 13.
By chief executive Deepak Pathak's account, a Blackwell assembly line at NVIDIA's Houston plant has run Skild's system since June, from first data to a working line inside a week on unchanged equipment 4. GR00T 2, the rival brain, claims up to 2.5 times the training throughput of NVIDIA's previous framework 3. Every outcome pays NVIDIA: the platform, the rival model, the stake.
The revenue receipt is one trade article
More than $100 million in annual recurring revenue and 60-plus paying customers, figures stated as of September 2026, come from a single trade article published on November 4 5. Skild's own January funding release claimed about $30 million of 2025 revenue 6.
In Seoul on November 10, Pathak confirmed deployments, the NVIDIA line, an LG battery plant and a warehouse run by Market Kurly, a Korean online-grocery firm, without putting a revenue number on any of them 4. LG is shareholder, customer and suitor at once: it signed a physical-AI MOU with NVIDIA in August, and its chairman met Microsoft's−1.81% — Microsoft, down 1.81 percent today chief executive on physical AI in September 8.
Freeing the physics layer is the right bet for the field: varied, cheap simulation speeds robot learning for everyone, and Pathak expects robots spreading through factories within the next year or two 4. Three receipts would settle the wager: a filing or audited customer count behind the $100 million, the first rival brain trained on Steel, and GR00T 2 deployments set against Skild's factory lines 4. Any one of them tells investors more than another benchmark win.
Get Robotics & Defense by email, free
Robots, self-driving vehicles, drones and AI in defense. In our daily email, every weekday.
More about NVIDIA
NVDAFiscal Q2 2027Revenue rose 106% to $96.2B, 92.5% of it from Data Center, “driven by the ramp of our Blackwell Ultra infrastructure”. Gains on equity stakes of $7.8B lifted net income to $59.7B.
Show as a table
| Line | Value |
|---|---|
| Revenue | $96.2B |
| Gross margin | 75.0% |
| Operating margin | 66.2% |
| Net income | $59.7B |
| Hyperscale | $48.7B |
| AI clouds & enterprise | $40.3B |
| Data Center | $89.0B |
| Edge Computing | $7.2B |
| Gross profit | $72.1B |
| Cost of revenue | $24.1B |
| Other income | $7.8B |
| Operating income | $63.7B |
| Operating expenses | $8.4B |
| Tax | $11.8B |
| R&D | $7.1B |
| SG&A | $1.4B |
