Cloud infrastructure spent decades catering to humans. You click. You scroll. You stream. Predictable patterns. Boring, in a good way. AI agents? Different beast. They wake up. Spin up a dozen sub-agents. Hammer databases, tear through documents, ping APIs. Then—poof —they’re gone. Leaving nothing but a trail of burst traffic.
AWS noticed. On Thursday they rolled out a new version of OpenSearch Serverless. It’s a fully managed search and vector database built specifically for agentic workloads. The promise is simple: scale instantly when agents go wild. Drop to zero when they stop.
Old infrastructure assumes a human behind the screen. This is a lie now. The tech industry is realizing its systems were designed for an audience that no longer holds the steering wheel.
The shift is happening quietly but aggressively. Machine-generated traffic isn’t the minority anymore. Cloudflare reports that bots made up 31% of all HTTP traffic in the last six months. A quarter of that? AI crawlers, search tools, assistants.
“Non-human traffic will exceed human traffic…” sometime in early 2027, Lai Yi Ohlsen at Cloudflare told TechCrunch.
Think about that date. Not far off. Google’s I/O conference last week teased a future where you delegate life tasks to AI—booking flights, researching buys. But enterprises are moving faster than consumers. Companies are deploying agents internally. Behind the scenes, machine-to-machine traffic is eating bandwidth.
Cloud providers have a problem. Systems built for human whims can’t handle autonomous agents constantly invoking tools and retrieving data.
That’s why Tia White at AWS calls the timing “straightforward.” Agents aren’t experiments anymore. They are in production.
“They spike without warning. They go idle without notice,” White said. “Enterprise needs search that keeps up…”
Without paying for the empty air in between.
Here’s the technical fix: decouple compute from storage.
Old serverless models still required one instance to stay alive. Storage and compute were glued together. You paid for a car sitting in a parking spot whether you were inside it or not. AWS’s new version separates the two. Compute scales to meet the burst. Then it vanishes. You pay $0.
“It’s like a metered parking spot,” the logic goes. Use it. Pay. Leave. Go.
Developers don’t need to lift a finger to manage this. OpenSearch Serverless integrates with Vercel and Kiro. You deploy backends. They handle the infra.
Everyone is racing to solve this. Databricks and Snowflake are rebranding as memory systems for enterprise data. Microsoft updated Azure to share memory between agents during those violent traffic spikes. Cloudflare just dropped its own infra updates last month. Same goal: persistent environments, instant scale.
The pattern is clear. As more companies unleash agents, the old internet structure will break under the weight of machine intent. We will have to redesign everything to accommodate them.
Will it make agents cheaper? Probably. Easier to scale? Undoubtedly. But we are watching the ground shift beneath us, layer by server layer. Who knows where the road ends now?
