Normal Computing
Normal Computing builds AI-accelerated verification software (Normal EDA) for complex chip designs and develops physics-based ASICs (Normal ASICs) implementing a proprietary thermodynamic computing architecture, selling to Fortune 500 semiconductor design and manufacturing institutions via enterprise field sales and deep co-development partnerships.
- Company typePrivate
- Founded2022
- HeadquartersNew York, United States
- Headcount51–100
- GTM typeB2B
- OfferingSoftware
What Normal Computing does
Normal Computing is a US-domiciled, privately held deep-tech company founded in 2022 and headquartered in New York, with additional offices in San Francisco, London, and Copenhagen. The company sells AI software and is developing physics-based silicon targeted at the semiconductor industry's verification and energy-efficiency bottlenecks, with stated end customers being Fortune 500 semiconductor design and manufacturing institutions. As of Aug 2025 it had grown to more than 45 engineers, researchers, and operators drawn from Google Brain, Google X, Meta, Palantir, IBM Research, NVIDIA, and Intel.
The company operates two converging technology stacks. Normal EDA is an AI-accelerated verification stack for complex IP and SoC silicon design that auto-formalizes natural-language chip specifications into a logically consistent ontology and from that generates hierarchical test plans, SystemVerilog stimulus, UVM-compatible testbenches, and human-in-the-loop traceability links, integrating with Synopsys VCS and Cadence XCelium simulators. Normal ASICs is a physics-based ASIC line built on a proprietary "Carnot" thermodynamic computing architecture that uses Langevin dynamics, non-equilibrium driving, and asynchronization to emulate stochastic differential equations in silicon, claiming up to 1000x energy efficiency on targeted AI and scientific workloads; the CN101 chip was taped out in Aug 2025, with CN201 planned for 2026 and CN301 for late 2027/early 2028. Both stacks are reinforced by a portfolio of open-source libraries (thermox, posteriors, Outlines, DSPy contributions, Branches, DRAMPyML/DRAMBench).
The business model is enterprise software licensing for Normal EDA (quote-based, Typeform-gated demos, SOC 2-compliant onboarding) paired with deep co-development of Normal ASICs alongside the largest semiconductor design and manufacturing institutions; hardware revenue is not yet recognized and the company is still in chip characterization and benchmarking. Go-to-market is high-touch enterprise field sales — the company hired its first Chief Business Officer (Craig Churchill) and a Senior Enterprise Account Executive (Aaron Sines) in 2025 — supplemented by an open-source developer footprint (GitHub, Discord), a content-led blog, and tier-1 PR coverage. To date the company has raised more than $85M cumulatively across a $8.5M seed (Jun 2023, led by Celesta Capital and First Spark Ventures with Micron Ventures), undisclosed follow-on seed extensions (cumulative seed ~$35M by mid-2025), ARIA Scaling Compute Programme selection (£50M, Oct 2024), and a $50M strategic round in Mar 2026 led by Samsung Catalyst Fund.
Normal Computing firmographics
Firmographics- Name
- Normal Computing
- Legal name
- Normal Computing Corp.
- Website
- https://normalcomputing.com
- Company type
- Private
- Founded year
- 2022
- Operating status
- Operating
- Headcount range
- 51–100 employees
- Short description
- Normal Computing builds AI-accelerated verification software (Normal EDA) for complex chip designs and develops physics-based ASICs (Normal ASICs) implementing a proprietary thermodynamic computing architecture, selling to Fortune 500 semiconductor design and manufacturing institutions via enterprise field sales and deep co-development partnerships.
- Ownership category
- akta.pro rank
Normal Computing industry classification
Industry- Product category
- Electronic Design Automation (EDA) Software
- NAICS
- Custom Computer Programming Services (541511), Instruments and Related Products Manufacturing for Measuring, Displaying, and Controlling Industrial Process Variables (334513), Bare Printed Circuit Board Manufacturing (334412)
- SIC
- Services-Prepackaged Software (7372)
- akta.pro primary industry
- EDA Software Platforms (Digital Implementation, Verification & Signoff) (HDAHAFAA)
- akta.pro secondary industries
- Hardware/Chip Design IP Cores (Processor, Interface, Analog, Security) (HDAHAFAF), EDA Cloud/SaaS, Design Data Management & Collaboration (HDAHAFAJ), AI Datacenter Power, Cooling & Racks (liquid cooling, power delivery) (HDAAAAAH)
Keywords
Where Normal Computing is headquartered
LocationHeadquarters
- HQ city
- New York
- HQ country
- United States
- HQ region
- North America
Offices4 records
Markets served
Normal Computing business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Operations, Marketing or Sales, Infrastructure
Revenue model
- Enterprise software licensing — Normal EDA: Normal EDA is positioned as an enterprise AI verification stack sold to semiconductor design and manufacturing organizations; pricing is quote-based, gated behind a Typeform demo request, and delivered with SOC 2-grade security. Commercial model implies subscription / enterprise licensing with annual contracts.
- Hardware (future) — Normal ASICs / thermodynamic computing chips: CN101 (tape-out Aug 2025) and roadmap chips CN201 (2026) and CN301 (late 2027 / early 2028) are intended to be commercialized as Physics-Based ASICs for the semiconductor industry. No chip sales revenue is yet disclosed; the company is currently in characterization and benchmarking.
- Strategic co-development with semiconductor partners: Normal Computing delivers software and hardware 'with the largest semiconductor design and manufacturing institutions' through deep partnership arrangements; this implies collaborative development, professional services, and likely joint IP/licensing revenue tied to specific chip programs.
Go-to-market motion2 records
Distribution channels3 records
Marketing channels7 records
Normal Computing product offering
Product offeringCore offering
Normal Computing sells Normal EDA, an AI-accelerated verification stack that auto-formalizes natural-language chip specifications into SystemVerilog test plans, stimulus, and UVM-compatible testbench code for Fortune 500 semiconductor design and manufacturing institutions. It is also developing Normal ASICs, a roadmap of physics-based thermodynamic computing chips (CN101, CN201, CN301) that exploit stochasticity and dissipation for orders-of-magnitude more energy-efficient AI and scientific workloads. The platform is reinforced by a suite of maintained open-source libraries (thermox, posteriors, Outlines, Branches, DRAMBench/DRAMPyML, Fuji-Web).
Product overview
Normal Computing delivers a two-pronged AI-for-silicon platform comprising (1) Normal EDA, an AI-accelerated verification stack that auto-formalizes natural-language chip specifications into SystemVerilog test plans, stimulus, and UVM-compatible testbench code, integrating with industry-standard Synopsys VCS and Cadence XCelium simulators, and (2) Normal ASICs, a physics-based ASIC product line built on thermodynamic computing principles (Carnot architecture) that the company is commercializing through a chip roadmap starting with the CN101 tape-out in August 2025 and progressing through CN201 (2026) and CN301 (late 2027/early 2028). The platform is reinforced by a suite of open-source AI/ML libraries developed and maintained by the company, including thermox (thermodynamic computing simulator), posteriors (uncertainty quantification), Outlines (controlled LLM generation), DSPy (language model program optimization), Branches (LLM reasoning visualization), DRAMBench/DRAMPyML (hardware auto-formalization benchmark co-developed with Fraunhofer IESE), and Fuji-Web (autonomous web agent). Together, Normal EDA and Normal ASICs realize the company's "co-design era" vision in which AI software and silicon share the same physics.
Differentiator
Problem solved
Functional benefit
Brands
- Normal EDA: AI-accelerated co-design and unified AI verification stack for silicon engineering teams, including structured ontology, intelligent verification planning, and testbench code generation.
- Normal ASICs
Products and services
- Normal EDA Holistic AI-accelerated verification stack for complex IP and SoC verification that auto-formalizes natural-language design specifications into hierarchical test plans, SystemVerilog/UVM stimulus, and testbench code; integrates with Synopsys VCS and Cadence XCelium simulators. Sold to Fortune 500 semiconductor companies under enterprise contracts with SOC 2-grade security.
- Normal ASICs Physics-based ASIC product line built on Normal's Carnot thermodynamic computing architecture that exploits stochasticity, fluctuations, and dissipation to enable stochastic, stateful, and asynchronous computing. Targets orders-of-magnitude more energy-efficient compute for AI and scientific workloads; commercialized through deep co-development with leading semiconductor manufacturers.
- CN101 Thermodynamic Computing Chip First thermodynamic computing chip implementing Normal's Carnot architecture via proprietary Lattice Random Walk (LRW); targets up to 1000x energy efficiency on linear algebra, matrix operations, and stochastic sampling for AI workloads. Foundational milestone validating the Normal ASICs roadmap.
- thermox Open-source JAX-based thermodynamic computing simulator providing fast and exact Ornstein-Uhlenbeck processes with zero discretization error and up to ~800x speedup vs. diffrax; serves as a digital baseline for thermodynamic linear algebra experiments with one-line APIs (e.g., thermox.linalg.solve).
- posteriors Open-source Python library for uncertainty quantification and Bayesian computation using PyTorch and its functional API; includes Laplace approximation methods that can leverage thermox for sampling. Used by researchers and ML engineers for probabilistic AI workloads.
- DRAMBench and DRAMPyML Open-source Apache 2.0 benchmark and framework for evaluating AI systems that auto-formalize natural-language DRAM chip specifications (JEDEC DDR2-DDR4, LPDDR2-LPDDR4, GDDR5/GDDR6, HBM2) into timed Petri net models that can drive SystemVerilog assertions, stimulus generation, and formal analysis.
- Branches Open-source library for visualizing and prototyping graph-based, objective-driven AI reasoning and planning with LLMs; includes webapp demos applying Tree of Thoughts to HumanEval and Game of 24 benchmarks. Used by AI/ML developers and researchers.
- Stochastic Processing Unit (SPU) First experimental electrical-circuit implementation of thermodynamic matrix inversion via 8 LC oscillators with all-to-all couplings plus FPGA; demonstrated approximate inversion of 4x4 and 8x8 matrices on real printed-circuit-board silicon with predicted O(d^2) time-complexity vs. O(d^3) for digital matrix inversion.
Companies that use Normal Computing
Customer profileSegments4 records
Ideal customer profiles3 records
Normal Computing technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
Integration2 records
AI capability13 records
Feature7 records
Normal Computing partnerships and signals
Strategic signalPartnerships
Four partnerships are on record, tiered flagship — joint research partnership., flagship — anchor uk/government partner., strategic — research consortium partnership. and operational — recruiting platform integration..
- Fraunhofer IESEflagship — joint research partnership.Co-developed the DRAMBench benchmark and the DRAMPyML framework with Normal Computing. Joint publications: 'DRAMPyML: A Formal Description of DRAM Protocols with Timed Petri Nets' (DV-Con Europe 2026), 'A Formal Description of Communication Protocols Using Petri-Nets' (MBMV 2026), 'Autoformalizing Memory Device Specifications using Agents' (ICLR VerifAI Workshop 2026).
- Advanced Research and Invention Agency (ARIA, UK)flagship — anchor uk/government partner.Normal Computing was selected for ARIA's £50M Scaling Compute Programme aimed at revolutionizing AI hardware costs. ARIA is the UK government's 'invention agency' funding high-risk research.
- Maryland Quantum-Thermodynamics Hub (University of Maryland)strategic — research consortium partnership.Normal Computing publicly backs the Maryland Quantum-Thermodynamics Hub, which secured three additional years of funding (per UMD press). Normal is a research collaborator in thermodynamic computing.
- Ashbyoperational — recruiting platform integration.Normal Computing hosts its careers page on jobs.ashbyhq.com/NormalComputing; job applications are processed through Ashby (noted in Privacy Policy).
Scale indicators12 records
Recent moves6 records
Expansion highlights6 records
Normal Computing competitors and assessment
Company assessmentMarket position
Competitive moat5 records
Key risks6 records
Key highlights7 records
Customer concentration
Normal Computing social profiles
Digital presenceNormal Computing compliance and trust
Trust signalCompliance3 records
Normal Computing financial estimates
Financial estimateRevenue estimate
Valuation estimate
Normal Computing leadership team
Management profileNumber of profiles
Profiles17 records
Normal Computing funding detail
Funding detailFunding overview
Funding rounds5 records
Investors13 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Normal Computing M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Normal Computing
What does Normal Computing do?
Normal Computing sells Normal EDA, an AI-accelerated verification stack that auto-formalizes natural-language chip specifications into SystemVerilog test plans, stimulus, and UVM-compatible testbench code for Fortune 500 semiconductor design and manufacturing institutions. It is also developing Normal ASICs, a roadmap of physics-based thermodynamic computing chips (CN101, CN201, CN301) that exploit stochasticity and dissipation for orders-of-magnitude more energy-efficient AI and scientific workloads. The platform is reinforced by a suite of maintained open-source libraries (thermox, posteriors, Outlines, Branches, DRAMBench/DRAMPyML, Fuji-Web).
Is Normal Computing a public or private company?
Normal Computing is a private company. It is classified as venture growth investor backed and is currently operating.
When was Normal Computing founded?
Normal Computing was founded in 2022. It employs 51 to 100 people.
Where is Normal Computing based?
Normal Computing is headquartered in New York, United States, in the North America region.
How does Normal Computing make money?
Three revenue lines are on record. Enterprise software licensing — Normal EDA is the primary driver. The others are hardware (future) — Normal ASICs / thermodynamic computing chips and strategic co-development with semiconductor partners.
Does Normal Computing have an API?
No public API is recorded for Normal Computing.
What industry is Normal Computing in?
Normal Computing's product category is Electronic Design Automation (EDA) Software. Its primary akta.pro industry code is HDAHAFAA, EDA Software Platforms (Digital Implementation, Verification & Signoff), with a secondary code of HDAHAFAF, Hardware/Chip Design IP Cores (Processor, Interface, Analog, Security). Its NAICS code is 541511 and its SIC code is 7372.