Arena Physica

Market position

Arena Physica positions itself against traditional commercial electromagnetic solvers and fragmented physics toolchains, arguing that verification through long simulation runs and physical testing is a primary hardware bottleneck, and against general-purpose frontier language models, which it reports failed its RF tasks while consuming more than 50,000 tokens per attempt.

Its stated differentiation is AI that is native to electromagnetic physics rather than general-purpose, combining a foundation model trained on proprietary simulation and measurement data with in-house RF expertise and a calibrated 140 GHz characterization lab, so predictions can be validated against silicon measurements.

Technology

The core technology is Heaviside, a physics-native foundation model trained on millions of designs and more than twenty years of proprietary simulation data. Heaviside-0 is reported to predict S-parameters directly from design geometry in about 13 milliseconds, or roughly 0.3 ms per board when batched on a GPU, which the company characterizes as approximately 800,000x faster than a commercial solver. Related research covers grayscale diffusion with flexible S-parameter conditioning for inverse design of RF passive components, and Atlas RF Studio is described as combining a data factory with the Heaviside-0 and Marconi-0 models, using fields as the training signal for generalization. Agent workflows in Atlas run staged graphs spanning planning, retrieval, simulation, solving, measurement and reporting, and the company's test capabilities are integrated with Atlas for measurement-simulation correlation, root cause analysis and yield management.

Go-to-market

The company sells through direct engagement with engineering organizations, offering forward deployed engineers who embed in customer stacks, and publicizes named design partners including AMD, Anduril Industries and Sivers Semiconductors.

Hardware and semiconductor engineering teams working on electromagnetically governed systems, including AI compute, defense technology, satellite and 5G communications, automotive radar, and high-speed networking and HPC interfaces.

Geography

Headquartered in New York City, where it also operates its RF and EM characterization lab; the company lists partner locations across the United States (including San Francisco, Los Angeles, Washington DC, Seattle, Austin, San Diego, Pittsburgh, St. Louis, Salt Lake City) and internationally in London, Barcelona, Sindelfingen, Aix-en-Provence, Waterford, Taipei and Tokyo.

History

Founded in 2019 in New York, the company spent roughly 2020-2024 building Atlas, an AI hardware engineer, while accumulating proprietary simulation and measurement data. In April 2025 it publicly launched Atlas and announced a $30M Series B, and introduced Atlas Edge for self-diagnosing machines. Over 2025-2026 it repositioned as Arena Physica around electromagnetic superintelligence, shipped Atlas RF Studio in beta and released the Heaviside foundation model. Research publications through 2026 cover Atlas RF Studio and the Heaviside-0 and Marconi-0 models, inverse design of multi-layer RF passives, and compute scaling under data constraints.

Risks & controversies

Performance claims such as ~13 ms electromagnetic prediction and ~800,000x speedup versus a commercial solver, and the reported failure of frontier general-purpose models on its RF benchmarks, are drawn from the company's own tests and materials. Third-party listings also disagree on basic company facts, giving founding years of 2019 and 2020 and headcount of both 51-200 and approximately 160.