Hey, I'm John Furrier with the Cube. Here at the Cube's NYSE Studios, of course. We were Palo Alto Studio connecting Silicon Valley and Wall Street. This is a special news alert. Obviously, we have a lot going on in AI infrastructure and the breakout companies are starting to emerge. Seeing clearly the ones that are making it happen. Vast Data is one of them, ones we've been covering since the launch of the company. I can't remember, it seems like 10 years ago. It's only a few years ago. The founder and CEO of Vast Data is here inside the Cube again. Renan, great to see you. Thanks for coming in.
Congratulations. Thank you.
News we're about to break here on the Cube as you guys release the news. You have a funding, a Series F financing at a $30 billion valuation. And that's up from about 9 billion and change in a relatively short time. Congratulations. Give us the hard news. Announce it.
Yeah, it's the latest funding round. We've been growing and once in a while as we grow, we like to do these rounds. We find new investors that want to join the company. We give some liquidity to early investors and employees and we get a new mark. And so $30 billion is the latest and now we need to triple again.
It's been fun to watch the rise and accelerated velocity of the business on two fronts. One, obviously, the numbers. And we'll go through some of those, but more importantly, the technology. You made some good bets. We've covered that in the past and if you look at the last GTC from Nvidia, their conference where Jensen Huang pretty much is the bellwether event as well as supercomputing and a few others, but that's pretty much the big one. Storage, okay, was in the slide as again, another year of featuring, but much more strategic. And Vast, I remember when you launched, just to kind of do a history lesson here, you really felt strongly that you didn't want to be seen as a storage company. Obviously, good bet. Take us through that. Obviously, a good call, data platform operating system. I think we said that on the Cube. You guys took that, ran with it. Good job. We can debate that, but it is an operating system. Take us through that progression. Where did you see that escape velocity kick in and how does that map to today?
So, we started by building a new architecture. We call it disaggregated shared everything. That's what allows us to break those fundamental computer science trade-offs between price and performance and scale and resilience and ease of use. And then we built the storage system on top of that early days because we saw that you need very fast access to a lot of data to enable these AI systems. And then our customers came back and said, 'But we also need a structured data store.' And so we built the Vast Data Base. And they said, 'We need a new way of orchestrating compute across DPUs and GPUs. Training is different than inference and prefill is different than decode.' And so we built a data engine for data pipelines. And so as we add more and more parts and across geographies, we built the data space. It really is filling out that software infrastructure layer of this new AI stack. And that's why we call it the operating system. As we shift from predominantly training to inference and fine-tuning and reinforce learning, the weight moves from compute to data. You need things to be distributed at the edge, close to where the people are, very resilient. You need security in place so that nothing gets leaked and that everything is observable. All of that needs to be handled by the software layer and that's what we're about.
So, it's safe to say you guys made an architectural bet. A first principles bet.
I'm really glad you wanted to get that out there because let's talk about the market. Obviously, the market was growing significantly in the hyperscalers and AI infrastructure, the rise of the neo clouds, more specialty clouds. I just consider them all hyperscalers in general just as a blanket category. And then the enterprise certainly hot. The needs of the hyperscaler were unique. You hit those. This year the market's changed a bit. You're seeing inference, you mentioned that. What's the hot market for you right now? Obviously, you're continuing to service the main market. What's the main market opportunity for Vast right now?
So, as we said, it's a new stack that's getting built and Jensen put it really nicely. He said it's a five-layer cake. It's power, it's chips, it's software infrastructure, models and applications. And so we're that middle layer and we work with the underlying layers and we work with whoever is building models on top of us. That was the early days. Now the applications are starting to get built and the market is shifting from xAI and Mistral and Inflection over to the enterprise. The enterprise is starting to put AI workloads into production. They need it to be simple. They need it to be secure. We're enabling them to make that transition.
Okay, I need to ask you a question. Obviously, we have to do some of the business results in a second. The market has shifted from training to inference, you pointed that out. A lot of people didn't get that right. They went all in on training. And then training changed like, 'I need training and inference.' And there's fine-tuning. What's the success formula for a versatile, resilient architecture that I could get a stand up some infrastructure and get all of it?
Yeah, so training needs to be very large. There's a lot of GPUs that are very hungry for data to train on. And it needs to be very fast. And that's all you need for training. Inference requires high levels of resilience and low latency and the ability to add more and more tools on top as you build for these new models and applications. And so there are a lot more constraints for inference. Reinforce learning is combining those two because you want to infer on data as it comes in and then you want to fine-tune your model of the universe over time. And so really you need the combination of both.
And you need it to be across geographies because shallow inference will happen in the device, in the car, in the robot. More deep inference will happen at the edge and then training will happen at the core. And so you need that data space to make data accessible across all the different locations.
Got it. That's awesome. Market's going to grow certainly. The AI native applications need to have model choice. Sovereignty models is being talked about. Proprietary models in the enterprise certainly coming. Let's talk about the valuation and then how much did you guys raise? Talk about the amounts.
So, it's a billion-dollar round. It's a combination of primary and secondary. We've been not just growing very fast. We've been tripling every year roughly since we started. But we've also built a very efficient company. And so in parallel to growing so fast, we've been cash flow positive forever. We're now profitable. And so we don't need external cash infusions. And that's why it's a combination of primary and secondary.
A lot of people have not say complained to me, but like there's always been this dark art of revenue in storage business. Annual contracts versus longer contracts. People state certain amounts. How would you categorize your revenue? What give confidence to your numbers? Share the numbers if you can. Give people a taste of the relevance of the success.
So, as everyone, we do multi-year deals and we get long-term commitments up front. And so our sales numbers are larger than our revenue numbers as a consequence of that. I can share that so far we've sold more than $4 billion worth of systems, all software, not including any of the underlying layers of the stack.
4 billion in software like... or software numbers. That's right. No hardware.
That's right. Hardware is in addition to that. We don't sell the hardware. We partner with others to do that. Last year we exited with more than half a billion dollars in CARR, committed ARR. And so that's an annualized number. And as I said, we're roughly tripling year over year.
