Vinci
Continuous Physics Reasoning — a Palo Alto-based Physics AI company building intelligence infrastructure for hardware engineering, starting from semiconductor physics
Vinci is a privately held company founded in 2023 and headquartered in Palo Alto, California. Per company disclosure, it operated in stealth for more than two years before its formal launch on December 2, 2025. The founding thesis is to convert physics simulation from a discrete event into a computational layer embedded continuously in the design process. We read this as an attempt to use AI to displace the bottlenecks of legacy FEA (finite element analysis) workflows: dependence on scarce specialists, the meshing step, and hours-to-days of compute latency.
Ph.D., Stanford University. His doctoral work addressed automating high-fidelity meshing for complex real-world geometries (computational geometry), widely regarded as one of simulation’s hardest problems. He subsequently led software at Carbon (per industry media profile). The company website cites roughly 15 years in physics simulation. Our observation: the founder’s academic specialty (meshing) coincides with the process step the current product claims to make unnecessary. This supports depth of domain understanding, but also places a self-validation burden on the claim of displacing an incumbent process.
Described in company disclosure as a pioneer in large-scale machine learning and autonomous systems. Third-party profiles list prior roles as SVP of Engineering at Iron Ox (agricultural robotics), COO of Mayfield Robotics, and president of Women in Machine Learning, with an academic publication record in robotics and reinforcement learning. The company frames the founding pair as uniting physics-based simulation with production-grade AI.
• The October 2026 Series B release quotes Kabaria only as “co-founder and CEO” and does not mention Osentoski’s current title or role. We could not confirm her current role within the management team.
• Osentoski’s background is drawn from third-party profiles and academic databases and has not been cross-verified against an official company biography. Kabaria’s “approximately 15 years” is self-reported promotional material.
• Senior hires (CTO, CRO, etc.) and board composition are undisclosed, as is whether the three Series B co-leads hold board seats.
Vinci’s core product capability is Continuous Physics Reasoning, which per company disclosure delivers deterministic, solver-accurate understanding of product behavior while the design is still in progress. Commercialized physics domains are thermal, thermo-mechanical (warpage and stress) and convective fluid behavior, with semiconductor design as the initial market. The company states that semiconductors are its “first proving ground, not its destination.”
(as of Dec 2025)
We organize the business into three pillars below. Pricing and contract structure are undisclosed; our characterization of the model as license-based enterprise software is an assumption.
The company says its Foundation Model for Physics operates zero-shot on new designs with no customer-specific training or fine-tuning. No customer data requirement and on-premises, behind-the-firewall operation are central selling points.
Combines automated design understanding and preparation, agentic orchestration and GPU-native physics kernels. It automates the setup work traditionally done by specialists, and is claimed to process models from hundreds of millions to more than 15 billion degrees of freedom in minutes.
Physics axis: thermal, thermo-mechanical and fluid, extending to memory and advanced computing, then vehicles, aircraft and satellites. Engineering axis: from understanding what will happen, to proposing what should change, to generating new designs from human intent.
Key milestones: Dec 2, 2025 — exit from stealth and public debut of the thermal platform (claimed up to 1000x speed advantage). Feb 24, 2026 — launch of the second core physics capability, thermo-mechanical (warpage) analysis; a representative benchmark ingested ~1GB layout files and completed the workflow in about 30 minutes, covering 100×100 cm boards with ~20-micron features and 100×100 mm packages. Oct 6, 2026 — Series B announcement, with convective fluid cited as commercialized.
• Revenue, ARR, contract count and tenor, customer concentration and named customers are undisclosed. The deployment base is cited as “three leading semiconductor manufacturers” (Dec 2025), and we found no updated figure. The claim that half of the world’s top 20 semiconductor companies benchmarked the product comes from the company website and secondary coverage.
• All performance figures (1000x speed, 15B degrees of freedom, ~30-minute runtime) are self-reported; we found no independent third-party verification.
• AMD participates through AMD Ventures and an AMD senior executive is quoted in support, but a commercial customer relationship is not stated. We do not treat AMD as a customer.
• Some media and databases list the legal entity as Vinci4D, Inc. Take care to avoid confusion with unrelated companies of similar name (e.g., Vinci SA, Vinci Capital Partners).
Vinci closed a $250M Series B roughly ten months after exiting stealth. Disclosed funding through Series A totaled $46M; adding this round gives approximately $296M cumulative (our simple sum). That Advent (global private equity) and Temasek (sovereign investor) joined Xora as co-leads suggests, in our assessment, a shift in the investor base from deep-tech VC toward large-scale growth capital.
CB Insights records a convertible note in November 2023. The company disclosed that Eclipse led the Seed round but did not state individual amounts.
Xora, a Temasek-backed deep-tech VC, led the Series A, with participation from Khosla Ventures and Eclipse. Subtracting the $36M Series A from the $46M total implies roughly $10M of earlier capital including Seed (our calculation).
Participants include AMD Ventures, Madrona and existing investors Eclipse and Khosla Ventures, among others. Disclosed uses of proceeds are broader physics coverage (from memory and advanced computing to vehicles, aircraft and satellites) and expanded engineering functionality (design-change recommendations and intent-driven design). By our calculation, this round is roughly seven times the Series A.
• Amount raised: $250,000,000 (Series B)
• Valuation: $1.5B (stated as “valuing the company at”; pre- vs. post-money not specified)
• Co-leads: Advent, Temasek, Xora Innovation
• Participants: AMD Ventures, Madrona, Eclipse, Khosla Ventures and others undisclosed
• Cumulative capital raised (our sum): approximately $296M ($46M + $250M)
• The press release does not specify whether $1.5B is pre-money or post-money. The Series A valuation is also undisclosed, so the step-up multiple between rounds cannot be calculated.
• The mix of primary issuance versus secondary sales, per-investor amounts, ownership percentages and preferred terms are undisclosed. The “others” participating are unidentified.
• Seed figures conflict across sources. Seedtable lists the entire $46M as Seed, VCBacked shows $36M as total raised, and CB Insights shows $46.5M across four rounds. We used the company press release (total $46M, Series A $36M) as the baseline.
• Without revenue figures, we cannot assess the $1.5B valuation on a revenue-multiple basis.
Vinci’s differentiation claims concentrate on three axes: (i) a zero-shot architecture that applies to new designs without customer data, (ii) compute scale and speed at manufacturing resolution, and (iii) deterministic, solver-verified outputs. In our assessment, (i) and (iii) map directly to semiconductor customers’ IP security and reliability-qualification requirements and are design choices that lower adoption friction. At this stage, however, all figures rest on self-reporting.
The company contrasts Vinci with surrogate models that must be trained or tuned per design, stating that it works on new geometries without separate training and runs behind the firewall. This can lower adoption barriers in an environment where chipmakers are reluctant to move design data off-site or allow it to be used for training. The flip side is that a moat built on accumulated customer data is likely to be weak.
Disclosures cite manufacturing-grade models of hundreds of millions to more than 15 billion degrees of freedom processed in minutes, and thermo-mechanical analysis of ~1GB layouts in about 30 minutes. The company claims to retain full resolution where existing tools rely on geometry simplification (blocks or averages). If independently benchmarked, this could become a core moat; today it is unverified.
The company positions Vinci not as a replacement for numerical solvers but as a “physics intelligence layer” that reasons over physical laws, with solver-grade methods used for external verification. The design directly targets hallucination concerns about generative models, which we view as favorable for acceptance in hardware design, where reliability requirements are strict.
The disclosed team strength is the pairing of computational geometry and simulation expertise (Kabaria) with production ML and robotics operating experience (Osentoski). AMD Ventures’ participation and a public supporting quote from AMD’s senior vice president of technology and engineering read as signals of technical acceptance within the industry, though neither is evidence of a commercial contract.
Competitive landscape (our view): Incumbents in thermal and thermo-mechanical analysis are traditional CAE/EDA solver vendors (Ansys, which we understand was acquired by Synopsys in 2025; Cadence; Siemens, among others), and Vinci also competes with a newer cohort of Physics AI startups. Incumbents hold advantages in installed base, certification and workflow lock-in, while Vinci attacks on speed, automation and mesh-free operation. This landscape reflects our general market understanding, which we did not independently verify in this research, and we did not obtain vendor-level share or pricing data.
- • $1.5B valuation formed without disclosed revenue or customer metrics, leaving it heavily reliant on the technology narrative
- • No independent verification of performance and accuracy claims; zero-shot generalization to new physics and industries is unproven
- • Concentration in a single segment (semiconductor thermal/thermo-mechanical); possible dependence on a few large customers (concentration undisclosed)
- • Incumbent solver vendors embedding AI features, plus certification and ecosystem lock-in
- • Execution risk as physics coverage and validation requirements rise sharply in aerospace, vehicles and satellites
- • Rising complexity of advanced packaging (2.5D/3D) is structurally expanding demand for thermal and warpage analysis
- • The pace of product extension from thermal to thermo-mechanical to fluid offers partial evidence of roadmap execution
- • Extending the same architecture across physics and industries could widen the TAM beyond semiconductor CAE
- • Advent and Temasek joining the round provides capacity for large-scale sales build-out, global expansion and follow-on financing
- • Evolution to design-change recommendations and intent-driven design would move value capture from an analysis tool to a design decision platform