Every few years, someone promises to abolish the hard drive. The latest is Space, a San Francisco startup coming out of stealth with $2.4m in pre-seed funding and an unusually literal pitch: a distributed filesystem that lets people and AI agents work across petabytes of live data while using, the company claims, zero local disk space.

The round was led by a16z’s Speedrun accelerator, with Golden Ventures, Northside Ventures and a dozen angels from companies including Parsec, Sentry, Superwhisper and Modem.

The idea is to collapse the cloud and the local drive into a single shared layer that sits just above the operating system.

Files appear in the Finder and open in the native apps you already use, but nothing is fully downloaded; when an application or an agent asks for data, Space says it streams only the exact byte ranges needed to finish the task.

In theory a video editor could open a project larger than their laptop, and an agent could traverse an entire organisation’s files while reading only the fragments relevant to its job.

That last point is where the founders are placing their bet. Most agentic tools today make a user upload an entire file to a chatbot even when the task needs a single page or frame, which is slow and wasteful, much as giving agents their own native primitives has become its own cottage industry.

• “Humans and agents are working across more data than any local disk can hold,” said co-founder Jason Zhao. “They should not have to wait for files to be copied or ingested before work can begin.”*

The timing fits a broader scramble to build the plumbing beneath the AI boom, from VAST Data’s multibillion-dollar storage empire to a rash of startups selling the unglamorous data layers agents need to function.

Space’s contribution is to argue that the humble filesystem, the oldest abstraction in computing, is the natural home for that access layer, precisely because every app, device, person and agent already speaks it.

The snag is that a cloud drive which streams only what you need and mounts like a local disk is not a blank page.

Tools such as LucidLink have given creative teams on-demand access to enormous media projects for years, and Dropbox, Box and Google Drive have each bolted on selective sync.

Space’s wager is that those were built for humans syncing files one device at a time, whereas the coming surge of demand is agents crawling them at machine speed, and that a filesystem rebuilt natively for both beats a sync tool with AI stapled on.

Space is in private beta with roughly 100 users and teams onboarded, and claims an organic audience of more than 80,000 across platforms, a following the founders trace to a previous company where shifting terabytes of video each month became an organisation-wide bottleneck.

“Space is disrupting one of the oldest assumptions in computing: a file has to exist on your device before you work with it,” said Jonathan Lai, a general partner at a16z, casting the product as a single mechanism that can serve human creators and AI agents at once, an instinct visible across the firm’s wider bet on agent infrastructure.

The founders’ longer-range pitch is grander still: a “Space Computer” in which the physical machine becomes a window onto effectively unlimited storage and compute.

Whether a small team can deliver anything resembling infinite computing on $2.4m is, obviously, the open question, and the graveyard of “the disk is dead” promises is well populated.

Yet the observation underneath the hype is sound: AI agents have quietly turned data access, not storage, into the real bottleneck, and the filesystem is a shrewd place to fight over it. Space has picked a genuinely large problem.

The harder part, as ever, is building the boring, reliable machinery that makes petabytes feel local.

Get the TNW newsletter

Get the most important tech news in your inbox each week.