imgix.

Here is a slightly strange business. imgix, a San Francisco company that has been around since 2011, will take an image you already own, sitting on a server you already pay for, and it will resize it, crop it, compress it, convert it to a modern format, and hand it to your users faster than you could. And the interface for all of this - the entire product, really - is a URL. You append ?w=800&auto=format,compress to an image address and the picture that comes back is smaller, sharper, and correctly sized for whatever screen is asking. That is the whole trick. It sounds too simple to be a company, which is usually a sign that it is a good one.

The counterintuitive part is that most engineering teams, left to their own devices, will build this themselves. They will write a service that generates seven thumbnail sizes every time a user uploads a photo, store all seven, and then discover a year later that they need an eighth size for the new phone, at which point they have to regenerate everything. imgix's pitch is that you should not do this. You should keep one original image and let imgix produce every variant on demand, on the fly, from the URL. This is the sort of insight that is obvious in retrospect and worth a lot of money in practice.

The origin story is appropriately literal. Chris Zacharias was YouTube's second full-time web developer - he built the early video player and the file uploader - and at some point he noticed that YouTube's video thumbnails were larger than they needed to be. This is the kind of thing that bothers a certain type of engineer in a way it does not bother normal people. Zacharias left, tinkered, reportedly built the early version out of his living room, and picked up some of his first clients making a website for members of the comedy group The Lonely Island. In 2011 he raised seed money from Y Combinator and SV Angel; in 2014 imgix opened a proper San Francisco office and raised a Series A from RRE Ventures. The legal entity has operated under the name Zebrafish Labs.

The customer list is the sort that makes the product self-explanatory. Unsplash, the free-photo giant, runs its imagery through imgix and grew from roughly 20 million images a month to 350 million in a year without a delivery problem. Kickstarter serves more than 1.4 billion images a month, generated on demand from single origin files rather than pre-baked in every size. Eventbrite processes seven image sizes per event and, having handed the problem to imgix, no longer maintains any image infrastructure of its own - which is really the point. The company also counts Porsche, YETI, and Japan's Nikkei among tens of thousands of customers, and it says it delivers on the order of billions of image requests every day. It does all of this with a team of roughly forty to fifty people spread across the US, Canada, Germany, and Japan, which is a remarkable amount of internet to sit underneath with so few employees. That is what a good abstraction buys you.

Around 2023 imgix did the thing every software company was expected to do, which is add AI, but it did so in a notably undramatic way. Instead of building a separate product, it extended the URL. Background removal, object removal, generative fill, super-resolution upscaling, text-to-image - all of it became more parameters on the same query string that developers had already learned years earlier. Then in 2025 the company shipped video using the exact same grammar: adaptive delivery, smart cropping, watermarking, auto-generated previews, image-to-video, and, in its Q2 2026 release, captions generated from a video's own audio and translated into nearly 100 languages. The lesson, if you want one, is that when your core abstraction is chosen well, you can bolt on entire new media types and AI features without asking anyone to relearn anything. Most companies would kill for that.

imgix charges on consumption. It moved to a credits-based model so that every customer - not just the big ones - gets access to images, video, and AI on the same platform, with a free tier covering up to 1,000 origin images and full API access. In practice most customers spend somewhere between $500 and $5,000 a month, driven mostly by bandwidth, the volume of source images stored, and the number of transformation requests; the largest enterprises spend well past $10,000. It is a usage-based infrastructure business, which means it grows when its customers grow, which is a comfortable place to sit. The competition - Cloudinary, Cloudflare Images, ImageKit, Fastly's optimizer, Akamai - is real and well-funded, and imgix's answer has consistently been the same: the simplest interface wins, and the interface is a URL.