Breaking: Nixtla raises $16M Series A led by Energize Capital TimeGPT trained on 100B+ data points 68 million open-source downloads Customers include Microsoft, Zalando, Decathlon, Lyft Forecasting in three lines of code Named Fast Company Next Big Things in Tech 2024 Breaking: Nixtla raises $16M Series A led by Energize Capital TimeGPT trained on 100B+ data points 68 million open-source downloads Customers include Microsoft, Zalando, Decathlon, Lyft Forecasting in three lines of code Named Fast Company Next Big Things in Tech 2024
The Company Report - San Francisco Edition Vol. TS-1 - Time Series Intelligence
Nixtla company logo
Company Profile - Artificial Intelligence

Nixtla.

The company that decided forecasting shouldn't require a data science department - just three lines of code and a model that has read a hundred billion moments in time.

A logo on a white card, corn-yellow border, orange shadow. The founders named the company after nixtamal - the ancient way of preparing corn. The joke writes itself: they process raw time the way their ancestors processed maize.

~2022Founded
$16MSeries A - Feb 2026
68M+OSS Downloads
~28Employees
The Feature

A Foundation Model, But For Boredom

Here is a fact about forecasting that everyone in finance quietly knows and rarely says out loud: predicting the future is both the most valuable thing a business can do and the most tedious. Somebody has to decide how many jackets Decathlon ships to Lyon in October, how much electricity a grid needs at 6 p.m., whether a machine is about to fail. For decades this was the province of specialists - people who fit ARIMA models by hand and argued about seasonality. It was important work. It was also, structurally, a spreadsheet job.

Nixtla's entire thesis is that this should stop being a job and start being an import statement. The company, founded around 2022 by three engineers from Mexico who relocated to San Francisco, built something called TimeGPT - a foundation model trained not on the text of the internet but on more than 100 billion data points of pure time. Retail. Electricity. Finance. IoT sensors. The model learned what time series tend to do, the way a language model learns what sentences tend to do, and now it will forecast yours in, they insist, three lines of code.

Everyone else was training models to talk. Nixtla trained one to predict - and then gave the tools away before charging a cent. The strategy, in one sentence

The interesting part - the part that makes this a real company and not just a clever demo - is the order of operations. Nixtla did not start with a product to sell. It started with open source: a family of libraries collectively called the Nixtlaverse, including StatsForecast and NeuralForecast, that quietly accumulated something on the order of 68 million downloads and 16,000-plus GitHub stars. Data scientists reached for them because they were fast and they worked. By the time TimeGPT and the enterprise platform arrived, Nixtla wasn't introducing itself to the market. It was already the market's default.

What you can actually do with it

Concretely: you point TimeGPT at a table of historical numbers with timestamps, and it returns a forecast - with uncertainty intervals, if you want them - plus flags for anomalies it thinks are weird. You can fine-tune it on your own data, feed it exogenous variables (weather, promotions, holidays), and run it across thousands of series at once. The pitch to enterprises is numeric and refreshingly un-mystical: up to 42% more accurate than traditional methods, roughly 10x more efficient at inference. One retailer reported a 35% lift in store-level forecast accuracy. A large mobility company cut its forecasting false alerts by 85%.

Those are the kinds of numbers that survive a procurement review, which is the whole point. Nixtla sells the number, not the vibe. The enterprise product wraps the model in the things large companies actually require before they'll deploy anything: on-premises or cloud deployment, security, compliance, dedicated support. The latest releases - TimeGPT 2.1 and Nixtla Enterprise 2.0 - added multivariate modeling and "agentic" forecasting, which is the current term of art for letting the system reason about your data rather than just fit a curve to it.

Up to 42% more accurate. 10x more efficient. 35% better store-level forecasts. 85% fewer false alerts. Nixtla's reported customer results

Why the market believes it

There's a tell in the hiring data. TimeGPT now appears as a requested skill in job postings at companies like OpenAI, DoorDash, and Tesla. That's a strange and useful kind of validation - not a customer logo, not a press release, but employers deciding that fluency in your tool is worth screening for. It suggests the technology has crossed from novelty into infrastructure.

The customers reinforce it. Microsoft, whose venture arm M12 backed the company, deployed TimeGPT-1 with sub-five-second response times. Prudential, Unilever, Decathlon, Lyft, Zalando, Nestle, and Grab appear on the roster. In February 2026 the company raised a $16 million Series A led by Energize Capital, with True Ventures and GreatPoint Ventures along for the round. When your customers' investors are also your investors, the market is telling you something. Nixtla's answer to what it does with the money is characteristically flat: accelerate production-ready systems that solve real forecasting and decision-making problems. No manifesto. Just fewer steps between a company's data and its next decision.

If there is a risk here, it's the obvious one that shadows every foundation-model company: the big labs and cloud providers can decide time series is worth their attention, and Amazon, Google, and Meta all have forecasting tools of varying seriousness. Nixtla's defense is the same as its origin story - it got there first, it's the default in the notebooks where this work actually happens, and it has spent years making the boring part easy. In a field full of companies promising to change everything, there's something almost contrarian about one whose ambition is to make prediction dull, cheap, and one line long.

The Byline

Three Engineers, One Model

MM

Max Mergenthaler Canseco

Co-founder & CEO

Serial entrepreneur and engineer with a background in building data science teams. The public face of the company and its funding story.

AG

Azul Garza Ramirez

Co-founder & CTO

Applied-math and economics background, an experienced machine-learning engineer, and a lead researcher behind TimeGPT.

CC

Cristian Challu

Co-founder & Chief Scientific Officer

Holds a PhD in machine learning and drives the scientific work under the Nixtlaverse and the foundation model.

The three met at a previous startup and decided to build an open-source time-series model together.

The Catalog

What They Ship

2024 - FOUNDATION MODEL

TimeGPT

A generative pre-trained model for time series, trained on 100B+ data points, that forecasts and detects anomalies across retail, energy, finance and IoT in a few lines of code.

2022 - OPEN SOURCE

Nixtlaverse

StatsForecast, NeuralForecast, MLForecast and HierarchicalForecast - the open-source libraries with tens of millions of downloads that made Nixtla a default.

2025 - ENTERPRISE

Nixtla Enterprise

On-prem or cloud deployment with security, compliance and support - now with multivariate modeling and agentic forecasting in Enterprise 2.0.

The Record

How It Happened

2022

Nixtla founded

The founders launch the open-source Nixtlaverse and incorporate the company in San Francisco.

2023

Open source scales

StatsForecast and NeuralForecast reach millions of downloads and thousands of GitHub stars.

2024

TimeGPT launches

A foundation model that forecasts in three lines of code arrives; Fast Company names it a Next Big Thing in Tech.

2025

Enterprise matures

TimeGPT 2.1 and Nixtla Enterprise 2.0 ship with multivariate modeling and agentic forecasting.

2026

Series A

Nixtla raises $16M led by Energize Capital to advance time-series intelligence.

The Roster

Who Forecasts With Nixtla

MicrosoftPrudentialUnilever DecathlonLyftZalando NestleGrab
Marginalia

Four Things Worth Knowing

The name Nixtla nods to nixtamalization, the ancient Mesoamerican process for preparing corn - a Mexican fingerprint on a Silicon Valley company.
TimeGPT was trained on more than 100 billion data points spanning retail, electricity, finance and IoT.
TimeGPT now shows up as a requested skill in job postings at OpenAI, DoorDash and Tesla.
Nixtla gave away its core technology as open source - tens of millions of downloads - before building a paid product.
The Broadcast

Watch & Listen

The Questions

FAQ

What is Nixtla?

Nixtla is a San Francisco company building time-series intelligence tools, best known for TimeGPT - a foundation model for forecasting and anomaly detection - and the open-source Nixtlaverse libraries.

What is TimeGPT?

TimeGPT is a generative pre-trained foundation model for time series, trained on over 100 billion data points, that produces forecasts and detects anomalies with just a few lines of code.

Who founded Nixtla?

Nixtla was co-founded by Max Mergenthaler Canseco (CEO), Azul Garza Ramirez (CTO) and Cristian Challu (Chief Scientific Officer), who met at a previous startup.

Who uses Nixtla?

Enterprise customers including Microsoft, Prudential, Unilever, Decathlon, Lyft, Zalando, Nestle and Grab, along with a large open-source developer community.

How much funding has Nixtla raised?

Nixtla raised a $16 million Series A in February 2026 led by Energize Capital, bringing total funding to roughly $44 million including earlier backing from Microsoft's M12, GreatPoint Ventures and True Ventures.

Pass It On

Share This Profile