← All stories
● Covered by 1 source · 1 reportMedium impact1 neutral

AI Workloads Should Run Near Data to Mitigate Rising Costs and Data Gravity

🔄 Updated 2d ago
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • AWS increased EC2 Capacity Block GPU prices by ~38% for H200 instances.
  • Data gravity impacts AI workloads more due to large datasets and frequent reads.
  • Moving 300 TB of data can cost $6,000 per copy in transfer fees.
  • Interoperability overhead and duplicate pipelines add to the total cost.

Rising Costs of Centralized AI Compute

AWS has implemented significant price increases for its EC2 Capacity Blocks, specifically for reserved GPU capacity. Prices rose by about 15% in January and another 20% in July, totaling approximately a 38% increase for H200 instances that experienced both adjustments. These price hikes highlight the growing expense of accessing remote GPU resources, especially when data needs to be moved to them.

The Impact of Data Gravity on AI

Data gravity, the principle that compute and services are drawn to large data concentrations, is increasingly affecting AI workloads. Unlike traditional analytics that might use sampled data, AI training and inference often require repeated reads of massive datasets. When these reads span different regions or environments, data transfer costs and network latency accumulate, making it more efficient to process data where it resides.

Cost Implications of Data Movement

Moving large datasets to centralized compute environments incurs substantial costs. For example, transferring a 300 TB dataset at a hypothetical rate of $0.02 per GB would cost $6,000 per copy. These transfer fees are in addition to the compute charges and can quickly escalate with frequent data movement. Real costs vary based on factors like source, destination, and service terms.

Hidden Costs Beyond GPU Rates

Beyond direct transfer fees and GPU rates, other expenses contribute to the total cost of AI workloads. These include interoperability overhead, which arises from managing cross-environment pipelines, and the engineering hours required to maintain duplicate data pipelines. These factors should be considered when determining the optimal placement for AI compute resources.

Strategic Data Placement for AI

The analysis suggests that for AI workloads, the strategic decision is whether the workload moves to the data or the data moves to the workload. Given the escalating costs of data transfer and the nature of AI's data demands, placing AI compute resources closer to the data sources is becoming the more economically sound approach. This minimizes transfer costs, reduces latency, and simplifies data management.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

Reporting from

Running AI workloads near their data sources is becoming critical due to increased data transfer costs and the compounding effect of data gravity. AWS has raised prices on EC2 Capacity Blocks for GPUs by approximately 38% for H200 instances, making cross-region data movement more expensive. This trend suggests that moving compute to data is more economical than moving large datasets to centralized compute resources for AI applications.