Announcing Encord’s $60 million Series C funding

Today, we are thrilled to announce that Encord has raised $60 million in Series C funding led by Wellington Management to scale our AI-native data infrastructure as physical AI hits an inflection point. This milestone brings Encord’s total funding to $110 million and is a testament to the dedication of our team and a reflection of AI’s rapid progress within the real world.

The round also included participation from existing investors Y Combinator, CRV, N47, and Crane Venture Partners, and new investors Bright Pixel and Isomer Capital. It brings our company’s total funding to $110 million and enables us to accelerate the development of a universal data layer for production-ready AI.

In the coming years, AI will touch every aspect of our day-to-day lives, and the reliability of AI models will be tested as never before. Encord’s technology will ensure that AI models are always trained and run on the best data available—continuously, and at scale.

AI is entering the physical world

AI is breaking out of the chatbox and into our physical environments, where the reliability of AI robots depends entirely on the quality of data that powers them. 

The past few years of AI applications have been text-based – large language models (LLMs) trained on internet datasets of articles, comments, and websites. The performance of LLMs has tended to improve with the amount of training data. From GPT-3 to GPT-4, for example, the size of the training set increased from 45 terabytes to roughly 1 petabyte. 

But for the emerging wave of physical AI applications – think autonomous vehicles, delivery drones, and humanoid robots – data scale is only half the battle. Vast amounts of multimodal data must be captured, organized, and enriched across the model lifecycle – often in near real time.

As AI companies move from prototype to production, they often encounter new and unexpected challenges in data governance, infrastructure, and transformation, from small numbers of malicious data points poisoning a model, to sharp jumps in computationally intensive multimodal data processing needs , to custom development to integrate humans into their data workflows. 

All of which means that AI companies are often blind-sided by data quality issues just as they prepare to enter the market.

Legacy data infrastructure can't keep up

The root problem is that traditional enterprise data approaches focused on cloud-based infrastructure were never designed to help companies manage, curate, or annotate and align their AI data. 

As a result, AI companies are unable to use legacy platforms to make their data usable. (One of our Product leaders has likened data management for training and running AI models to looking for a million needles in a billion haystacks.) Even when legacy software works in the prototyping phase, it is likely to fall short when transitioning to production. 

The upshot? In order for companies to dependably deploy AI at scale - and to capitalize on the massive growth potential of physical AI, in particular - they need a revolution in data infrastructure to match the one we’re already seeing in AI more broadly.

Encord’s AI-native universal data layer is our response. Our platform helps AI companies with every data automation and processing task they encounter, from facilitating data generation in the pre-training phase to aligning models in accordance with human feedback, post-training. Specifically, our data layer handles:

Curation. Specifying and curating data to pinpoint the most relevant and useful data for the AI model.

Management. Indexing and integrating data with storage facilities, enabling a complete view of the data and the ability to filter, trace and manage it.

Annotation and alignment. Annotating and labeling training data. Evaluating and aligning models to the requirements of user applications. Getting the right data into the model while getting the wrong data out of the model.

Building the missing data infrastructure for the future

In the next few years, analysts project that hundreds of millions of AI robots will come online. The physical world will pose new challenges for today’s AI applications in the form of myriad unexpected conditions: electric outages, inclement weather, and other unforeseen circumstances. 

And even non-technical companies will need to incorporate AI into their offerings and operations to remain competitive. 

At Encord, our work with AI leaders like Woven by Toyota, Zipline, and Flock Safety gives us a glimpse into the future, and we’ve already seen that massive increases in data storage and computation needs will require a new generation of purpose-built, AI-native data infrastructure. 

Many of the world’s leading physical & next-gen AI companies trust us to meet that need, as highlighted by the stunning growth we’ve seen in the last twelve months. Encord’s platform grew from 1 petabyte to over 5 petabytes over that period, while our revenue from physical AI customers increased 10x. 

In the new world, every company will require a universal data layer to help manage, curate, annotate, and align the data on which they build and deploy AI models. This imperative will only grow more urgent as the ways in which robots and other AI systems learn continue to evolve.

Our commitment is to provide that missing data layer, and in doing so, to help achieve AI solutions previously thought unsolvable by technology. 

If you’re an AI leader building Physical AI, book a demo to see how Encord can help you train and build your AI on the right data. 

If you’re interested in joining our team, check out our Careers page. We’re hiring across all teams in San Francisco and London. 

It’s early on in the AI revolution, and we can’t wait to help build what’s next.