Amazon S3’s standard storage price hasn’t dropped in a full decade. Simon Willison pointed this out in a short post, noting that the service “used to drop in price reasonably often.” Then the cuts stopped. The last one came in December 2016, and standard storage still costs $0.023 per GB-month today.
Some background: S3 is the service a big chunk of the internet uses to store files, and a lot of AI training data sits there too. So its price history tells you something about cloud economics in general.
📉 The Numbers
Here’s the S3 standard storage price history Willison put together:
- March 2006: $0.150/GB-month (launch)
- November 2010: $0.140
- February 2012: $0.125
- December 2012: $0.095
- February 2014: $0.085
- April 2014: $0.030
- December 2016: $0.023
- Today: $0.023
That’s seven prices in ten years, then nothing for the next ten. Over the first decade, storage got about 85% cheaper.
The biggest drop is the April 2014 one, a 65% cut in a single move. That wasn’t a slow slide down the cost curve. Google had just slashed its own cloud storage prices, and AWS answered right away. Most of the savings customers got came from a price war, not from engineering.
🧊 Why the Cuts Stopped
So why has the price been frozen since? Here are a few forces that likely explain it:
- The market settled into an oligopoly. Once AWS, Microsoft Azure and Google Cloud split the market between them, none of them had much reason to start another price war.
- Switching costs lock customers in. Moving petabytes out of S3 is slow and expensive, partly because of egress fees (the charge for moving data out of AWS). Customers who stay put don’t push hard on price.
- AWS cut prices sideways instead. Instead of lowering the standard rate, Amazon added cheaper tiers: Infrequent Access, Glacier, Glacier Deep Archive and Intelligent-Tiering. Price-sensitive customers get discounts if they rework how they store data. Everyone else keeps paying $0.023.
- Hardware savings got slower. Hard drives still get cheaper per GB, but not at the pace they did in the 2000s. AI data centers are also competing for the same drives.
There’s also the inflation angle. A flat price since 2016 means S3 has gotten cheaper in real terms, by roughly a third. But that’s inflation doing the work. AWS didn’t cut anything.
🤖 What This Means for AI Pricing
This is where the story matters for anyone building with AI. Model prices today look a lot like S3 prices between 2006 and 2016. Per-token costs keep dropping, sometimes by huge amounts, as OpenAI, Anthropic, Google and a pile of open-weight labs compete for developers.
S3’s history suggests those cuts won’t go on forever. Commodity infrastructure tends to follow the same pattern:
- Early competition drives steep, regular cuts.
- A few big price wars cause the largest drops.
- The market consolidates and prices flatten.
- Discounts move into tiers, commitments and “optimized” products.
We can already see the fourth step starting in AI. Batch APIs, prompt caching discounts, cheaper “mini” and “flash” models, and reserved-capacity deals are the AI version of Glacier and Intelligent-Tiering.
🛠️ Practical Takeaways
If you’re budgeting for AI or cloud infrastructure, here’s what to take from this:
- Don’t assume prices will keep falling. Plan around today’s prices. Treat future cuts as a bonus, not something your business model depends on.
- Use the tiers. Lifecycle rules on S3 or batch and cached inference on AI APIs are where the real savings are now.
- Stay portable. Price wars only help you if you can switch. Abstraction layers and multi-provider setups keep you in a position to benefit.
- Watch for new competitors. Cloudflare R2 and Backblaze B2 have made egress fees a selling point. New pressure like that is what gets prices moving again.
Ten years without a price change isn’t failure. It’s what a mature market looks like. The open question is how quickly AI inference gets there too. You can find Simon Willison’s full price table in his original post.