Your test vehicle records more video than the logger can hold. Some teams cut drives short; others slow collection down.
Beamr Lossless lets the logger hold nearly twice the footage, so the vehicle spends more time collecting and less time swapping and uploading.
What we measured, and where it runs:
-> 47% smaller than raw, on recordings from eight 8-megapixel cameras.
-> 12-bit Bayer in, 12-bit Bayer out.
-> On the GPU encoder that typically sits idle, separate from the compute cores.
Test Beamr Lossless on your own data and pipeline: https://t.co/8PmjGSBS9a
Half the world called in "sick" today, and it has absolutely nothing to do with a certain video game 🤒🎮
But if you work with AV/ADAS data and can spare 3 minutes between "symptoms", you’d be doing the whole industry a huge favor by helping us build its new benchmark.
3 minutes, and then you can get back to your "recovery"...
👉 https://t.co/6UUl597dJO
Beamr CEO Sharon Carmel published the Q3-2026 Letter to Shareholders today.
Highlights:
1. We expect 2026 revenues to grow about 50% over 2025.
2. In autonomous vehicles, we’re working with 12 companies, including Tier-1 suppliers and global AV programs.
3. Beamr is where teams already work: integrated with RTMaps by Intempora, a dSPACE company; a Technology Partner in the VAST Data Cosmos Community; and on AWS Marketplace.
4. Our second study on NVIDIA Cosmos Curator published this month showed 45.9% smaller video data vs. standard encoding, while preserving ML fidelity.
5. At petabyte scale, even a small number of production customers could have a substantial impact on our future revenues.
Read the full letter -> https://t.co/RDYzWwayAa
Our second NVIDIA Cosmos Curator study runs on a much broader set of autonomous vehicle video data: 96 clips, 3 vision-language models, under production-like generation settings.
The results are clear-cut: 45.9% smaller files with Beamr’s Content-Adaptive Bitrate compression (CABR) compared to standard NVENC constant-quality encoding, at mid-range levels - while caption meaning is preserved.
The added shift from Beamr’s optimization is practically unnoticeable: Qwen2.5-VL showed only about 0.003 additional semantic shift, within the uncertainty of the measurement.
Our conclusion: it’s possible to map the right compression point. In deployment, that means measuring file size and downstream task performance for your specific model and content, then choosing the lowest rate that still holds the performance you need.
@NVIDIADRIVE
-> Read the blog: https://t.co/jVsusR2SuC
-> Test on your own video data: https://t.co/FM3mSIFLQS
We've been talking about the AV data explosion for years. It's time to put some real numbers behind it.
Proud to see Beamr taking the lead on this initiative. The more voices we bring in, the more valuable it becomes for the entire industry.
🚦 Introducing The State of AV Data 2027.
How much data do AV teams collect, keep, and train on? And what does it cost?
We're building an industry-wide benchmark to find out.
Launching at CES 2027.
📊 Work with AV data and have 5 minutes? Share your view:
https://t.co/xS0Fqzh235
#StateOfAVData #ADAS
Yesterday’s selloff had several causes. One of them is the future of data centers: UBS, as reported by Reuters, estimates that as much as 30% of planned US data-center capacity additions in 2028 could be at risk.
Demand for AI computing, storage and massive volumes of visual data keeps growing anyway. So what happens when AI demand keeps growing, but the infrastructure needed to support it becomes increasingly scarce?
Beamr CEO published a statement on that: Beamr has always been an infrastructure accelerator - reducing video data volumes to accelerate transfers, lower costs and improve efficiency. When AI infrastructure becomes scarce, Beamr’s GPU-accelerated technology becomes a capacity multiplier: doing significantly more with the infrastructure you already have.
Full statement -> https://t.co/OhY9yCsWgR
Beamr CEO Sharon Carmel in an interview with The Wall Street Transcript:
-> Outlines Beamr's expectations of approximately 50% revenue growth in 2026.
-> The urgent need: "Physical AI systems need enormous amounts of visual data... it is growing exponentially, and the cost of moving, storing, and processing it is becoming a major limitation."
-> The ML-Safe solution: "Cutting the size of the video by up to 50% while preserving everything that the machine learning process needs."
-> The scale of the opportunity: "A relatively small number of production customers can have a very substantial impact on our revenues."
The full interview -> https://t.co/1RPbMNYBTX
Missed it on Bloomberg TV Saturday? Beamr CEO Sharon Carmel interview airs on CNBC today at 11:30 a.m. ET.
Sharon talks about the validation of Beamr’s technology for autonomous vehicles, our path to commercialization, and how we offer a structured program for each customer to design, test, and validate compression on their own models and data, and then build a deployment plan that fits their needs.
@RedChip
Watch a clip from Beamr CEO Sharon Carmel interview airing on Bloomberg TV this Saturday, October 24, at 7:00 p.m. ET, and on CNBC this Sunday, October 25, at 11:30 a.m. ET.
-> What ML-Safe compression means for autonomous vehicles
-> How Beamr's technology was validated by NVIDIA
-> Beamr Blueprint AV: helping AV teams evaluate and validate compression on their own models and data
-> And the research discovery that compression can improve AI model accuracy
Read @RedChip press release about the interview: https://t.co/JlSHgSFIfc
We’ll be at @NVIDIAGTC Berlin, October 20–22, to walk through what ML-safe video data compression actually requires across the pipeline: from capture at the edge, through detection, depth, and embedding.
Lossless or Content-adaptive - the right mode depends on the model, the dataset, the quality criteria, and the pipeline stage. Knowing where to compress, and how far, is the work.
Managing autonomous vehicle pipelines? Schedule a meeting with us here: https://t.co/fQ9xdmmSb0
@NVIDIADRIVE
Tommy Flanagan, the editor of Faultline, looked at our ML-Safe work and how we are validating it for the autonomous vehicle market - the storage and networking challenge and what it takes to make massive video datasets smaller without changing what the models see.
Worth reading: https://t.co/Eh3LTLF8yP
Curious about our ML-Safe webinar? You can watch it here: https://t.co/zIraOgHIea
.@BeamrVideo is bringing Ai upscaling to live sports broadcasting and streaming with Nvidia Video Super Resolution combined with Beamr’s content-adaptive bitrate (CABR) encoding, running inside existing production and delivery chains.
https://t.co/sQY3VDbvis
Our CTO Tamar Shoham spoke with @AutoSens_ TV about ML-Safe compression:
-> Why we can't ignore the elephant sitting on the servers, and how to identify the information that matters least to the model.
-> Why lossless and content-adaptive compression must be safe for machine vision models.
-> How we benchmarked ML-Safe compression across 2D and 3D object detection, depth estimation and reasoning models.
https://t.co/hsLLXEp6k5
On stage at the @dSPACEglobal User Conference, talking about ML-Safe Lossless compression.
12-bit Bayer in, 12-bit Bayer out, zero data loss, in real time. Running on the GPU's video encoder that is already in the data logger (47% smaller than raw in our initial tests).
Test it on your video data: https://t.co/8PmjGSBS9a
@AutoSens_ Europe, Barcelona, live from the floor. ML-Safe Lossless in the spotlight, on stage and in our booth.
Here? Talk to us at booth 131.
Not here? Visit: https://t.co/h4asntoLth
Lossless compression for RTMaps: 12-bit Bayer in, 12-bit Bayer out, in real time.
It runs on the GPU already in the data logger, while the compute cores keep running without extra load.
Capture more in the vehicle, and move it faster and at lower cost.
RTMaps users will get two modes to choose from:
-> Lossless where you need the outputs bit-exact. Initial tests showed 47%.
-> Content-adaptive, for up to 50% smaller files, confirmed against the model's own accuracy metrics.
Beamr and @INTEMPORA will demonstrate it this week:
-> @AutoSens_ Europe in Barcelona, September 22–24, stand 131
-> @dSPACEglobal User Conference West Coast, September 23 in Santa Clara, California
Visit https://t.co/8PmjGSBS9a