Skip to main content

10 Gigawatts of Renewables = 3.2 Billion Events a Day in the Databricks Lakehouse

Summary

  • Nexalis processes 3.2 billion operational events per day from a 10-gigawatt renewable energy fleet—covering wind, solar, and storage assets—and makes this industrial data available in Databricks in under four seconds through a zero-copy Delta Sharing architecture.
  • The Nexalis platform normalizes hundreds of OEM protocols and naming conventions through an AI-powered contextualization engine, producing AI-ready tabular data aligned to a unified asset hierarchy of site, asset, and class governed by Unity Catalog.
  • Production deployments with EDF Power Solutions, Clenera, and Calpine with Accure have delivered measurable results including 15% less equipment downtime and identification of $1 million in solar losses.

10 Gigawatts of Renewables = 3.2 Billion Events a Day in the Databricks Lakehouse

Watch: 10 Gigawatts of Renewables = 3.2 Billion Events a Day in the Databricks Lakehouse
Industrial AI's real bottleneck isn't the model — it's the data foundation. Operational data has the classic 3 V's at extreme scale: Variety (hundreds of OEM protocols and naming conventions), Velocity (1-second sensor streams), and Volume — 3.2 billion events a day from a 10 GW renewable fleet.
Nexalis is the cloud-native industrial data infrastructure layer for the AI era. A zero-copy architecture makes OT data available in your Databricks environment — cleansed, shaped, and AI-ready in under 4 seconds — unlocking the full analytics ladder: descriptive, diagnostic, predictive, and prescriptive.
Three production deployments: EDF Power Solutions on Maverick6 (solar + battery); Clenera (1.6 GW today → 5+ GW); Calpine with Accure (predictive battery degradation deep inside the BESS).
Same architecture, same pattern — heading next into conventional generation and nuclear. Anyone can dump events into a bucket. Cleansed, shaped, and AI-ready in less than 4 seconds is data engineering.
Talk By: Quentin Morel, CEO & Co-Founder, Nexalis ;

Chapters

FAQs

What is Nexalis and how does it work with Databricks?

Nexalis is a cloud-native industrial data infrastructure layer that collects OT data from wind, solar, and storage assets and delivers it to Databricks in under four seconds via a zero-copy Delta Sharing integration. Its edge agents, contextualization engine, and governance layer normalize the extreme variety, velocity, and volume of industrial sensor data into AI-ready Delta tables.

Why is industrial data so difficult to use for AI applications?

Industrial data presents three extreme challenges: variety (hundreds of OEM communication protocols and inconsistent naming conventions), velocity (real-time 1-second sensor streams), and volume (3.2 billion events per day from a 10 GW fleet). These factors make the data nearly invisible to AI without a specialized contextualization layer.

What results have Nexalis customers seen in production?

Three production deployments are referenced in this video: EDF Power Solutions on Maverick6 (solar and battery), Clenera managing 1.6 GW growing to 5+ GW, and Calpine with Accure for predictive battery degradation inside BESS systems. Reported results include 15% less equipment downtime and $1 million in solar losses identified.

How does Nexalis enable the full analytics ladder for renewable energy operations?

Nexalis delivers cleansed, time-aligned, tabular data into Databricks that supports the full analytics progression—descriptive, diagnostic, predictive, and prescriptive. Multi-agent orchestration then enables automated fleet monitoring and autonomous decision-making at scale across the renewable asset portfolio.

Full transcript

[00:07] Good afternoon everyone. I'm Quentin Moral, the CEO of Next Alice uh and co-founder. Um so today um I'm going to share what we have done with 10 gigawatt of renewables assets. Um you will see Next
[00:19] Alice is positioned as an industrial data infrastructure. So we really position ourselves as uh the data foundation for industrial AI, uh which makes a lot of sense with our collaboration with Databricks. Um so
[00:33] very happy to to share with you what we have done um uh so far in the renewables uh space. Um so let me start with uh with a number. So when we talk about a fleet of 10
[00:46] gigawatt of renewables assets, uh it represents a huge uh volume of data. Um so in this case, 3.2 billion of operational events per day
[00:58] uh with a 10 10 gigawatt fleet of assets. So here, we want to understand the landscape. We are talking about um a fleet of assets deployed um in the field, so wind, solar, and
[01:11] storage assets um with uh inverters, wind turbines, uh a lot of different devices, and all of those devices they produce uh a huge number um uh of data.
[01:23] And to be very a bit more specific about what this means, um here we are really talking about the classic Vs. Um the first V is about the variety of the data. Um
[01:36] hundreds of OEM, um it's very complex to get the data with the different communication protocols coming from OEMs, um and everything around naming conventions uh is also part of this uh
[01:49] variety of uh industrial data. Second V is about the velocity of the data. All of this data is generated in real time. It's very hard to to manage.
[02:01] And the last V is the volume. So here we have this 3.2 billion events a day with this 10 giga watt of of assets. Um And today
[02:13] um really this industrial data is almost invisible with AI because of these challenges.
[02:26] So that's where Celectis provide a solution to these three challenges. So let me break down our solution in four categories. Um the very first one is the ability to collect industrial data.
[02:40] So here I'm talking about edge agents deployed as close as possible as the data source. So here we are able to get data from assets in OPC UA, Modbus,
[02:52] DNP3, all of these complex way of integrating industrial data. Second component is the contextualization and the centralization of the data. So here we have an AI
[03:06] engine able to contextualize data. So it is asset aware. So everything is about having a reference model where every single complex vendor tags is
[03:20] normalized against a unique data model. The third one is about governance. So everything around around permissions, lineage, and audit is very important. So
[03:33] that's a part we are doing very well with Databricks. So we are unlock this this industrial data. And of course the last one is to make this data really
[03:45] AI ready, it needs to be shareable and provided to any types of application. And in this case I'm going to share how we do that with with Databricks.
[04:02] So, this is the slide high level slide about how NexaLisa and Databricks works well together. On the left you have everything around OT data. So, SCADA, turbines, inverters, and meters.
[04:14] As I said earlier, NexaLisa will take care of all the data foundation. So, everything around ingesting data, contextualizing data, we take care of that. And the way we interact with Databricks
[04:28] is through a zero copy mechanism with a a delta sharing technology provided by by Databricks. Everything is done in real time under 4 seconds.
[04:41] Um when we do that um the data is provided into the customer's Databricks environments in a very seamless manner. We share tables from the NexaLisa
[04:55] environment to the customer environment and so the customer has a direct access to all the cleaned and structured data we provide for them. So, it unlocks AI
[05:08] and any type of business intelligence application. Um so, our clients are able to build seamlessly AI and applications uh thanks to the clean, structured
[05:21] industrial data we provide for them. So, um the goal is to facilitate um operations and to make the data as as clean and structured as possible.
[05:37] So, let me describe a bit further what this means um in the with this data cleaning and and and structure. So, as I explained earlier,
[05:50] industrial assets generate a huge volume of data. Um the core feature of Nexelis is to be able to manage this time series data in
[06:02] a very efficient manner and to make sure this data is uh seamlessly available in the Databricks environment. To do that, we developed what we call a smart aggregation. So,
[06:15] this complex second-level data, we convert it into a time-aligned uh tabular format. So, we create the tables uh and use the Delta Sharing uh
[06:28] technology. So, we share the data to Databricks. Um it is meant to be cost-effective and performant um because having very noisy data, time series data natively in
[06:43] Databricks will not make things super um cost-effective. So, we are making sure that the data we provide into the Databricks environment is make sense, is valuable, and can be directly um uh used
[06:57] by uh any AI features provided by Databricks. Of course, contextualization is key. Uh regardless of the sites, regardless of the vendors, regardless of the equipment
[07:10] manufacturers, we make sure the data is always um unified and normalized uh with our contextualization uh and reference data model. So, that's
[07:22] that's key. Uh the third part is about asset hierarchy. It's also something very necessary. If you want to do industrial AI, you need to have the right asset hierarchy, so which means uh
[07:36] data being being structured by site, asset, and class. Uh so, we are talking here about um a fleet level uh data normalization. And of course, the last one I talked
[07:49] about uh governance. So, everything we are doing with a Unity Catalog, all of that we leverage everything around uh security and governance of the data. Uh so, we have the right lineage and versioning and
[08:02] access control of the data. So, now let me talk about what in the end we deliver for uh renewables and
[08:15] industrial um operators. And we classify those into um the full analytics ladder uh into four levels.
[08:27] What we have here on the left is um everything that's descriptive. Um so, it's basically reporting. So, what happened last week? What happened last month? Um we are talking about fleet
[08:40] KPIs, business intelligence, and reporting. That's the first step and the first thing we deliver is about uh this first uh level, which is the descriptive one.
[08:52] Then, if we go one step further, we are talking about diagnostic. So, why something happened? Understanding the root cause of um of an event across sites and the fleet of
[09:05] assets. That's the second one. If we want to go further, um this is where it becomes interesting is everything that's predictive. Uh what happens? What will happen next? Uh how
[09:19] we can anticipate failure? Uh yield forecasting thanks to machine learning uh with the time series data. That's the the third step. And then, I believe the future is also about everything that's
[09:33] prescriptive. So, everything we can do with agents. So, what to do now to make operations uh better um and having agents recommending and acting uh thanks
[09:45] to uh operational data. What we've seen so far, so from stranded data to live intelligence,
[09:57] those are key figures uh we are comfortable to share today uh with the value we deliver to to our customers. So, we are talking about a 6 weeks um uh of contract, so to live data. Uh so,
[10:10] we want to make things fast. We want to do uh rapid time to value. Um we usually usually start with a one or two sites to demonstrate the value, and then we are talking about a full fleet deployment in
[10:23] a couple of months. Um on average, we see a 15% less uh unplanned uh downtime with uh for example, in renewables, uh we have uh recently done uh
[10:36] uh identified $1 million losses of uh solar generation for a single site uh in a in a for a solar uh solar utility scale solar developer. Uh and we are also able to demonstrate
[10:50] uh under 1 year of a demonstrated ROI. Um So, that's also a proof point showing that we now we can fully leverage the power
[11:02] of of AI um and and data bricks and to have a real impact on operations and to provide value and uh operational efficiencies.
[11:20] A few words on something very important in the industry is everything around uh sovereignty, governance, and security. Um so, putting OT data in the cloud is something sensitive. Uh and uh it's
[11:32] worth it because of everything I said earlier. Um, to gain more uh, efficiencies being able to leverage cloud and AI is something uh, very important today. But,
[11:44] sovereignty, security, and governance are super important. So, today we are able to work within the operators uh, own tenant. So, as I said earlier, we we take care of the uh, data foundation and
[11:57] then we are able to share the data to the customers Databricks environment. Um, we fully leverage the Unity Catalog features of Databricks for govern and
[12:09] audited uh, data uh, access. Um, and uh, everything around compliance and uh, control end-to-end
[12:21] by design and access control and residency of the data. We have clusters in uh, North America and Europe uh, and we are also able to select the right uh, the same cloud provider to avoid uh, egress cost. So, for example, if a
[12:35] client is running Databricks in Azure in Europe, we will set up the same environment uh, on our end to make sure we we avoid uh, uh, egress cost and the data sharing is the very uh, efficient
[12:48] and and cost-effective. So, here um, everything we have done today so far is in renewables. We are now uh, so it's
[13:00] uh, a a multi-gigawatt and scaling uh, with these uh, uh, fleets of uh, of assets, solar, wind, and battery. Now, we are extending our scope to conventional generation, power utilities, and oil and gas.
[13:13] This is the exact same issue. Uh, we are talking about the same industrial data problem, the same architecture, uh, and the same consumption story with a lakehouse provider with Databricks.
[13:31] So, a few words about a feature we have just released, which is Next Gen AI Co-work. It is an AI workspace for the industry. The goal is to build agentic workflows
[13:43] to improve operations. And we have classified this feature into three core principles. The very first one is what we call human to agent. Human to agent
[13:56] is about having an interface where human can ask any questions about industrial assets. So, it's in plain language. It provides live KPIs, asset topology, and
[14:09] structure analysis. But then, that's not rocket science. We are seeing a lot of chatbot and a lot of things going on with that, and everyone is used to
[14:21] ChatGPT-like interface right now. Now, the the step further everything with agent to human. Having autonomous agents
[14:33] built with missions and objectives are something we are also seeing as a providing key values to our customers. So, they are able to identify
[14:45] root cause analysis. They are able to make their own KPIs and report all of that into the preferred channel. It can be emails, Teams, WhatsApp. All of that is provided from the agents to the
[14:59] humans. And the last step is about the agent to agent so coordination. What we are seeing and what we are setting up right now is to have agents coordinating with each other,
[15:12] which means an agent specialized in monitoring a fleet, another one being able to receive reports from an agent and then making decisions about
[15:25] an an analysis and making sure that the operators gets only what's needed based on automated agents and automated agents. So,
[15:37] feel free to ask me any any questions. I'm available. Thank you for your attention.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.