Is Ndea profitable in 2026?

No. Ndea has no product, customers, or revenue as of 2026 and describes itself as a pure research lab with 3-5 years of runway from its seed round, per a May 13, 2026 analysis. Profitability is not the goal; proving its AGI thesis is.

Could scaling AI replace Ndea's entire approach?

That is the core risk. Ndea bets program synthesis beats pure scaling to AGI, but ARC-AGI-2 scores hit 68.8 percent by March 8, 2026 via scaling and tool-use, directly eroding the differentiation its thesis depends on.

Ndea was founded by Francois Chollet, creator of Keras and the ARC-AGI benchmark, and Mike Knoop, co-founder of Zapier. The lab publicly launched January 15, 2025, aiming to build AGI via deep learning-guided program synthesis.

Who is Ndea's biggest competitor?

OpenAI, Google DeepMind, Anthropic, and xAI, all reported to be pursuing program synthesis and test-time-compute techniques themselves as of a January 14, 2026 ARC Prize technical report, with far more capital and compute than Ndea's roughly 15-person team.

Does Ndea have any commercial product?

No. As of a May 13, 2026 report, Ndea has no commercial product, customers, or revenue, and has published no implementation detail about its architecture, remaining a research-only AGI lab since its January 2025 launch.

What is Ndea's technical approach to AGI?

Ndea fuses deep learning for intuitive pattern guidance with program synthesis for formal, compositional reasoning, an alternative to pure LLM scaling detailed in a January 16, 2025 report on the company's founding thesis.

YC Winter 2026 · United States

Ndea

Pre-product AGI lab betting program synthesis beats pure scaling.

Losing its protection

Assessed 27 July 2026 · Evidence as of 27 July 2026

The structural protection is eroding.

On the recordSince 27 July 2026 · Checkable from 2026 · Every call on the record Checkpoint: Ndea's first public ARC-AGI-2 leaderboard submission

The call

Ndea has no product, no revenue, and a thesis that the same frontier labs it's racing against could disprove just by scaling further.

Ndea is a small company trying to build artificial general intelligence by using a specific method called program synthesis, which they believe will work faster than the approach used by bigger competitors like OpenAI and Google. The problem is that these four giant labs have hundreds of times more money, more engineers, and nothing stopping them from using the exact same method that Ndea is working on, so there is no real reason customers or investors would stick with Ndea instead. The one thing that would change everything is if the big labs prove that their current approach (just making their AI models bigger and giving them tools to use) can already solve the hard thinking problems that Ndea is working on, which would mean Ndea's entire reason for existing disappears.

Why

AI pressure: OpenAI, Google DeepMind, Anthropic, xAI. These labs are scaling models plus tool-use on the same ARC-AGI-2 benchmark Ndea uses to prove its approach is needed. Could AI do the core job? Yes, likely.

Bigger players: OpenAI, Google DeepMind, Anthropic, xAI. Each has 100 to 1000 times Ndea's capital and faces no barrier to copying deep learning guided program synthesis. Could a giant copy it? Yes, easily.

What protects it. A two year research head start, Chollet's ARC-AGI benchmark credibility, and no rival lab-authored ARC-AGI-2 submission yet.

What would change this call

A frontier lab pushes ARC-AGI-2 past 85% via scaling plus tool use in 2026. Would show pure scaling matches Ndea's approach, undercutting its core thesis.
Ndea posts a proprietary top ARC-AGI-2 leaderboard submission during 2026. Would show Ndea's method still leads over scaling based approaches.
ARC-AGI-3 stays near 13% through 2026 even as frontier labs ship new flagship models. Would support Ndea's bet that scaling stalls on interactive reasoning tasks.

Key numbers

ARC-AGI-2 at 68.8% via frontier-lab scaling (2026-03-08)
$43M seed round led by NEA (2026)
About 15 employees, no product or revenue (2026)

Named risks

Frontier labs solving ARC-AGI-2 via pure scaling would publicly falsify Ndea's core thesis. Severity: high
No structural moat; program synthesis techniques replicable by rivals with 100-1000x more capital. Severity: high
Pre-product, pre-revenue status with only 3-5 years of runway before funding runs out. Severity: medium
ARC-AGI-3 interactive reasoning stuck near 13%, unclear whether Ndea's approach can move it. Severity: medium

Rivals, as read in July 2026

Rival	Direction	Evidence
OpenAI, Google DeepMind, Anthropic, xAI	gaining on	All pursuing program synthesis and test-time-compute; ARC-AGI-2 hit 68.8% via scaling by 2026-03-08
Sakana AI	contesting	Raised ~$135M Series B at ~$2.65B valuation in 2025, shipped Fugu Ultra multi-agent system June 2026
Axiom	contesting	AI mathematician built on Lean formal proof, adjacent alternative-architecture AGI bet, noted 2026-02-17

Questions people ask

Is Ndea profitable in 2026?: No. Ndea has no product, customers, or revenue as of 2026 and describes itself as a pure research lab with 3-5 years of runway from its seed round, per a May 13, 2026 analysis. Profitability is not the goal; proving its AGI thesis is.
Could scaling AI replace Ndea's entire approach?: That is the core risk. Ndea bets program synthesis beats pure scaling to AGI, but ARC-AGI-2 scores hit 68.8 percent by March 8, 2026 via scaling and tool-use, directly eroding the differentiation its thesis depends on.
Who founded Ndea?: Ndea was founded by Francois Chollet, creator of Keras and the ARC-AGI benchmark, and Mike Knoop, co-founder of Zapier. The lab publicly launched January 15, 2025, aiming to build AGI via deep learning-guided program synthesis.
Who is Ndea's biggest competitor?: OpenAI, Google DeepMind, Anthropic, and xAI, all reported to be pursuing program synthesis and test-time-compute techniques themselves as of a January 14, 2026 ARC Prize technical report, with far more capital and compute than Ndea's roughly 15-person team.
Does Ndea have any commercial product?: No. As of a May 13, 2026 report, Ndea has no commercial product, customers, or revenue, and has published no implementation detail about its architecture, remaining a research-only AGI lab since its January 2025 launch.
What is Ndea's technical approach to AGI?: Ndea fuses deep learning for intuitive pattern guidance with program synthesis for formal, compositional reasoning, an alternative to pure LLM scaling detailed in a January 16, 2025 report on the company's founding thesis.

Timeline

2026-07-14Ndea headquarters reported as San Francisco, United States
2026-06-23Sakana AI's Fugu Ultra multi-agent system compared against Anthropic models, highlighting rival alt-AGI approach
2026-05-13Analysis piece details Ndea's reported 43 million dollar seed and thesis risk
2026-03-08Survey shows ARC-AGI-1 at 93.0 percent, ARC-AGI-2 at 68.8 percent, ARC-AGI-3 at about 13 percent
2026-02-03ARC-AGI-2 Technical Report published on arxiv
2026-01-14ARC Prize 2025 Technical Report published; top Kaggle score 24 percent on ARC-AGI-2 private eval
2025-01-16Ndea's deep learning-guided program synthesis thesis detailed publicly
2025-01-15Ndea publicly launched, founded by Francois Chollet and Mike Knoop

Revision history

27 July 2026

First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.

In the same band

CellType OctaPulse The Token Company