NVIDIA + JAM : Unlocking Tens of Billions in Annual Savings
How the world’s leading AI company could cut operating costs by up to $60B annually with JAM.
Savings Breakdown
🗄️ Save 100x on Storage:
Jam is rated to save between 1.1x and 100x on exabyte-scale storage by combining lossless AI compression with global deduplication.
⚡ Save 90% on Energy:
By streaming at disk speed and offloading RAM, JAM cuts hardware requirements and power use by 90%.
🌐 Save 100x on Networking:
Updates and checkpoints shrink up to 100×, turning 50GB into ~500MB. Reduce the cost of digital twins. Save time with faster rollouts and guaranteed 50% lower bandwidth costs at scale.
Exploding Data, Rising Costs
Exponential Data Growth
AI/ML workloads doubling every 18 months.
Infrastructure Spend
1 Exabyte of storage = ~$6B per year.
Idle Time
Expensive GPUs sit idle due to I/O bottlenecks.
Jam Makes Infrastructure Smarter
Without Jam
- Dataset training requires a GPU.
- Learning 1TB requires 22 days and 1,460,000 watt hours.
- Servers require over 1TB of RAM to process big data.
- Site-to-site data-transfers and backups are still a bottleneck over existing backbone infrastructure.
- Raw datasets requires more storage
- Costly storage, compute, and bandwidth expansion
- Compression ratios of 1.5x to 3x with gzip/ZSTD
- Datasets require labelling and tagging, and often need to be filtered down for domain specific uses.
With Jam
- Dataset training requires a SSD.
- Learning 1TB requires 88 minutes and 18 watt hours.
- 1000× lower memory use, less hardware required.
- 100× smaller updates & patches (e.g. 50GB → 500MB), reduce ISP network use by orders of magnitude.
- 10–50% smaller storage footprints, JAM archives are designed for cold storage and hot retrieval.
- Seamless integration — no infra changes
- JAM content can be further compressed via gzip/ZSTD,
keep your existing pipelines in place or mix-and-match. - Train on ANYTHING, infer on ANYTHING, cross-domain learning without labels.
Annual Savings at NVIDIA Scale
Conservative (1 EB/year): ~$6B saved
Aggressive (10 EB/year): ~$60B saved