# SigmanticAI

Aiolum story: https://aiolum.com/story/sigmanticai/ · added 2026-11-05 · updated 2026-11-06

**In short.** SigmanticAI (AI agents that write and verify chip-design code, USA): First paying customer onboarded, July 2025. Evidence grade E: no revenue figure was checked. Team: 2 co-founders; Y Combinator lists a team of 2 in November 2026. Launched 16 July 2025; still running.

AI agents that write and verify chip-design code · USA · AI-native · team of 2 to 10 · still running · group: Developer tools and coding agents

- Outcome: First paying customer onboarded, July 2025 (evidence grade of the headline number: E; no revenue figure)
- Launch: 16 July 2025; months from launch to the outcome: 0
- Team: 2 co-founders; Y Combinator lists a team of 2 in November 2026
- Outside funding: Y Combinator (Summer 2025); amount not published
- Main factor: Niche
- Meters, 1 to 4: evidence strength 1, copyable pattern 1, capital needed 2
- Tags: AI-native, Small team, Funded, Running, B2B, Semiconductors

Links:

- Visit sigmanticai.com: https://www.sigmanticai.com/

**The product.** A team of AI agents for chip designers. It writes RTL code (Verilog and SystemVerilog) and the verification code that tests it, runs the customer's own simulators and other EDA tools, debugs failures and works toward coverage closure. It is installed with "pip install sigmanticai" and its pricing page lists a free plan and a Pro plan at US$200 a month.

**Why it worked.** Chip design teams spend much of their time writing and debugging hardware code and testbenches by hand, with tools that the founders say have changed little since the 1990s. The two founders had worked on custom circuits and processors and on fine-tuning language models, and built a coding assistant for this one field.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts.

## Summary

SigmanticAI sells AI agents for RTL design. RTL (register-transfer level) is the level at which engineers describe a digital chip, in hardware description languages such as Verilog and SystemVerilog. The agents also write verification IP, reusable test code that checks a design in simulation, and run it on the customer's EDA (electronic design automation) tools, such as simulators from Cadence, Synopsys and Siemens or the open-source Verilator.

Rohil Khare (CEO, formerly at Amazon) and Tamzid Razzaque (CTO, formerly at Apple), both from UC Berkeley EECS, launched it on Y Combinator's site on 16 July 2025 as part of the Summer 2025 batch. The launch post said they had built a RISC-V CPU with it, "onboarded our first paying customer", and reached more than 90% accuracy on the VerilogEval benchmark. In November 2026 the company's pricing page lists a free plan, a Pro plan at US$200 a month and an Enterprise plan.

No revenue figure was found, so this story has evidence grade E. The headline is the founders' own statement of one paying customer in July 2025.

## Pricing

SigmanticAI prices in November 2026, from the company's pricing page (grade C)

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free |  | 0 | 150 credits a month |
| Pro |  | 200 | 2,000 credits per Pro engineer; US$0.15 per additional credit |
| Enterprise |  | custom | unlimited generations, on-prem deployment, SSO |

Prices are per month. The page publishes no yearly price. Credits are used by agent model calls; deeper reasoning calls use more credits. Pro usage beyond the included credits costs US$0.15 per credit up to a monthly cap that the customer's administrator sets.

## How much revenue does SigmanticAI make? The numbers, graded

A two-founder Y Combinator company that sells AI agents for chip-design code; one paying customer stated at launch, and no revenue figure found

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Paying customers, July 2025 | a first paying customer | C | Founders' Y Combinator launch post, 16 July 2025: "onboarded our first paying customer". No later customer count was found. | [1] |
| Revenue | not published | C | No payment-processor profile, interview or filing with a revenue figure was found. See "What is not verified". | [1, 2, 3] |
| Prices | Free; Pro US$200 a month; Enterprise custom | C | Pricing page, read 5 November 2026. Pro includes 2,000 credits per engineer and costs US$0.15 per extra credit. | [4] |
| Products | Studio (available today) and Polaris Engine (alpha) | C | Home page, read 5 November 2026. Studio works with the engineer; Polaris is meant to work alone from a specification. | [2] |
| Benchmark result stated at launch | more than 90% accuracy on VerilogEval | C | Launch post. VerilogEval is a benchmark framework from NVIDIA. No independent run was found. | [1] |
| Delivery | a Python package, "pip install sigmanticai" (Python 3.9+) | C | Home page. Y Combinator's description still calls it a VSCode fork, the first form of the product. | [2, 3] |
| Deployment | cloud, or air-gapped and on-premises | C | Launch post; on-prem is in the Enterprise plan on the pricing page. | [1, 4] |
| Data handling | zero data retention; inputs not used for training | C | Terms page, effective 2 April 2026. | [5] |
| Payments | Stripe | C | Terms page: paid subscriptions and overage charges are billed through Stripe. | [5] |
| Team | 2 co-founders; team size 2 | C | Y Combinator company page, read 5 November 2026: founded 2025, San Francisco. | [3] |
| Funding | Y Combinator, Summer 2025 batch; amount not published | C | Y Combinator company page and the home page line "Backed by Y Combinator". | [3, 2] |
| Time from launch to the headline | 0 months | calc | Calculated: the launch post and the paying customer are both from July 2025. | [1] |

How to read this table: a grade C value is a statement by the company on its own site or its Y Combinator pages. A "calc" value was computed on this page. No value is grade A or B, and no row is a revenue figure.

## When did SigmanticAI launch, and how did it grow? Timeline

From a Y Combinator launch post with one paying customer to a two-product site with paid plans, in about 16 months. Each entry carries its grade.

- **2025** (grade C [3]): Rohil Khare and Tamzid Razzaque found SigmanticAI in San Francisco and join Y Combinator's Summer 2025 batch.
- **16 July 2025** (grade C [1]): Launch post "SigmanticAI - The Cursor for Hardware Design": natural language to Verilog and testbenches inside VS Code, a RISC-V CPU built with it, a first paying customer, 90%+ on VerilogEval.
- **2 April 2026** (grade C [5]): The terms page takes effect: zero data retention, credit-based billing through Stripe.
- **November 2026** (grade C [2, 4, 6]): The site sells Studio, a pip-installed agent, with Free, Pro (US$200 a month) and Enterprise plans; Polaris Engine is in alpha; the company publishes RTL Arena, a benchmark of AI models on Verilog tasks.

## Why did SigmanticAI work?

### One narrow field

General AI coding tools are trained mostly on software. SigmanticAI aims only at hardware code and its tests, and runs the result on the customer's real simulators, so the output is checked by the same tools that engineers trust.

### Fit with existing tools

The home page lists support for the main EDA vendors and open-source tools and says it generates the Makefiles and scripts for each, so a team does not change its flow to try it.

### Privacy for chip designs

Chip designs are valuable secrets. The company offers on-premises deployment and states zero data retention and no training on customer code.

## How it was built

Only what the company's own pages say. No page names the model providers.

- Y Combinator's description: fine-tuned Verilog language models, reinforcement learning and compiler feedback, refining code until it compiles and passes synthesis.
- First shipped inside a fork of VS Code; now a Python package that starts an agent from the command line.
- The home page describes a hierarchical system of more than 14 specialist agents, kept aligned by a component it calls SigmanticAI Brain.
- Credit-based billing through Stripe, as the terms page states.
- RTL Arena, the company's benchmark page, grades AI models by compiling and simulating their Verilog with Icarus Verilog.

Sources: [3, 1, 2, 5, 6]

## Who tried this before SigmanticAI, and why did they fail?

Two companies in our database of failed startups promised to make hardware design or AI hardware easier and failed. The values below are from that database and are not independently verified.

The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade.

### Wattage (2013 to 2015; funding US$2M; cause: Unit Economics)

- Promise: Anyone could design custom electronics with no technical skills, through an online interface and 3D printing.
- Why it died: 3D printing could not match the quality of normal manufacturing, and teaching customers cost too much.
- Record: 

### Rain AI (2017 to 2025; funding US$118M; cause: Product/Tech Failure)

- Promise: Brain-inspired AI chips with 100 times better energy efficiency for inference.
- Why it died: Developers could not easily program the chips, and the company spent its money before it had proof in real use.
- Record: 

## What changed

Wattage tried to remove the engineer from hardware design and sold to people with no skills. SigmanticAI sells to professional chip engineers and keeps them in control. Rain AI built new silicon; SigmanticAI sells software and needs no manufacturing.

What did not change: chip design is a small market of expert buyers with long sales cycles and strict security rules, which is why the company offers on-premises deployment.

## Can one person copy SigmanticAI now?

**Verdict: Only with deep domain skill.**

The pattern is an AI coding agent for one technical field, checked by that field's own tools. Both founders studied EECS and worked on circuits and processors before they started. Without that knowledge, a solo founder could not judge the output.

1. Pick a field with its own languages and checking tools, where general coding assistants do poorly.
2. Run the generated code through the customer's own tools and loop until it passes.
3. Offer on-premises deployment and no data retention if the customer's code is a trade secret.
4. Publish an open benchmark so buyers can compare models and tools.

**What is still open.** The same approach for other specialised code: PLC programs for factories, FPGA bitstreams, scientific simulation input files.

## What could go wrong for SigmanticAI?

- **What is not verified.** No revenue figure was found. We looked for a payment-processor profile, a founder interview with figures, and a filing. The only traction statement is the founders' launch post of 16 July 2025, which says they "onboarded our first paying customer"; it names no customer and gives no amount, and no later customer count was found. The Python package page on PyPI answered with a bot challenge and could not be read. A reader could look for a funding announcement, a customer case study or an interview with the founders.
- **Claims without a method.** The home page says the product is "50x" faster than manual development of verification and IP, and the launch post claims more than 90% accuracy on VerilogEval. Neither page gives the method or the data behind these numbers.
- **Competition and model providers.** No other story on this site sells AI for chip design; the nearest are AI coding tools for software, such as TypingMind's chat interface for models. The company does not name its model providers. Y Combinator's page speaks of fine-tuned Verilog models, the pricing page says credits are used by "agent model calls", and the RTL Arena page compares models by "total API cost", so the product may depend on outside model APIs; the pages do not say which.
- **Product still changing.** Y Combinator's description calls the product a VSCode fork; the site now installs it with pip. The second product, Polaris Engine, is marked alpha, and phase 4 of the roadmap is "In progress" and phase 5, specification to silicon, has no status mark.

## Sources

1. Y Combinator Launches, 16 July 2025: SigmanticAI - The Cursor for Hardware Design — https://www.ycombinator.com/launches/NzI-sigmanticai-the-cursor-for-hardware-design (grade C). Read in full. Founders' own launch post. Source for the first paying customer, the benchmark claim and the founders' background.
2. SigmanticAI home page, read 5 November 2026 — https://www.sigmanticai.com/ (grade C). Read in full. Source for the products, the install command, the EDA tools, the founders and the claims.
3. Y Combinator company page: SigmanticAI — https://www.ycombinator.com/companies/sigmanticai (grade C). Read the company data in the page: founders, team size, batch, founding year and description.
4. SigmanticAI pricing page, read 5 November 2026 — https://www.sigmanticai.com/pricing.html (grade C). Read in full. Source for the plans, credits and overage price.
5. SigmanticAI terms of service, effective 2 April 2026 — https://www.sigmanticai.com/terms.html (grade C). Read in full. Source for data handling and billing through Stripe.
6. SigmanticAI RTL Arena benchmark page — https://www.sigmanticai.com/rtl-arena.html (grade C). Read the method section. The leaderboard loads by script and was not shown.
7. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Contact: aiolum@livyden.com. Run SigmanticAI? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: SigmanticAI).