We are excited to announce our investment in Decart, an AI infrastructure company building the foundation for real-time, interactive AI. Backed by Sequoia Capital, Benchmark, NVIDIA, Radical Ventures, and Zeev Ventures, Decart has raised more than $450 million to date, having reached a $3.1 billion valuation in October 2025.
The AI stack wasn’t built for real time
AI is reshaping virtually every industry. But the infrastructure powering today’s AI boom — from the chips NVIDIA builds to the foundational models OpenAI trains — was largely designed for offline use cases: generating an image, producing a video clip, answering a question. These tasks tolerate latency. You submit a request, you wait, you get a result.
A growing share of the most valuable AI applications, however, don’t work that way. Generating video without lag, running inference for robotics, powering immersive gaming environments, and enabling live, interactive AI experiences all require something fundamentally different: real-time response. According to Decart, most video generation tools today take around 20 seconds to produce a single second of footage. For video to feel responsive in real time, a system needs to generate a new frame every 40 milliseconds. That’s effectively a 100x gap between where the current stack delivers and where real-time use cases demand.
Built from the ground up for real-time inference
Decart was founded to close that gap. Rather than optimizing around the edges of existing infrastructure, the company rebuilt the stack from scratch with real-time inference as the primary design constraint, working at both the processor level and the foundational model level simultaneously.
At the processor level, Decart’s software operates at the deepest layer of the hardware stack. Modern GPUs contain thousands of processors that spend a significant portion of their time sitting idle, waiting for data to travel across the chip and return. Decart eliminates that idle time, keeping every processor occupied at all times. The result is a processing engine that runs faster and at a fraction of the cost. Decart says it has reduced the hourly cost of generating content from hundreds or thousands of dollars to under $0.25, and that its customers are already seeing 25 to 100x cost reductions on hardware, including AWS Trainium.
This efficiency gain also addresses one of the most acute constraints in AI today: the global shortage of GPU compute. By making the same hardware go dramatically further, Decart is effectively expanding the supply of usable compute without waiting for new chips to be built.
At the model level, Decart’s flagship product, Oasis, is a real-time generative AI video model. Unlike a traditional video game that renders a pre-built world through game code, Oasis generates every pixel in real time in response to the player’s inputs. There is no world stored on a hard drive. The AI builds it frame by frame as you interact with it.
When Oasis launched in November 2024, it reached one million users in its first three days according to Decart. Decart has also since launched MirageLSD, a real-time video-to-video transformation model, expanding the category further. When the processor optimization and the model layer work together, the combined effect is what makes real-time inference at production scale possible. That is Decart’s core technical contribution and, in our view, a very defensible advantage.
An impressive level of commercial and technical depth for a two-year-old company
Decart was founded in late 2023 by Dean Leitersdorf and Moshe Shalev, both veterans of Israel’s Unit 8200. Despite being less than two years old, the company has established a level of technical, financial, and commercial cooperation with NVIDIA and Amazon that is noteworthy at any stage. That depth of partnership reflects how seriously the infrastructure layer of the AI industry takes what Decart has built.
The business metrics reinforce this. Despite raising over $450 million, Decart says it has spent less than $10 million of investor capital, funding its compute costs through revenue. The GPU optimization platform generates meaningful enterprise revenue through licensing agreements with cloud providers and AI laboratories, while the consumer product layer continues to grow.
Why we’re excited
Oasis puts Decart at the front of the race to build what researchers call world models, AI systems that can generate and interact with dynamic environments in real time. We believe this category represents a platform-level shift comparable to earlier generations of computing. World models are also widely seen as foundational to robotics: for a robot to navigate and act in the physical world, it needs to understand and predict how that world works, which is precisely what world models are designed to do. We believe we are in the early innings of that revolution.
In our view, the future of AI won’t be determined solely by who builds the largest data centers. It will also be won by the engineers who figure out how to get dramatically more out of the hardware that already exists. We believe Decart is one of those teams, and we’re glad to be partnering with them.
The views expressed in this post regarding market trends, technological development, and the future of artificial intelligence reflect the opinions of Alpha Partners Management, LLC as of the date of publication and are subject to change without notice. These statements are forward-looking in nature and are based on current beliefs and assumptions. Actual developments may differ materially from those expressed or implied. Venture capital investments involve substantial risk, including the potential loss of all invested capital. There is no guarantee that the trends or outcomes described will occur or that this or any other investment made by Alpha Partners will be successful. Past investment decisions are not indicative of future results.
