Hark just raised $700 million in a Series A for an AI product it won’t describe in any detail. The startup says it’s building a “universal” AI interface and multimodal models that will work with existing products and services, then eventually ship purpose-built hardware for those systems. The first models are supposed to arrive this summer.

That’s about all we know. The company is being cagey about what makes its approach different from the dozens of other startups promising personal AI assistants, agentic systems, or new ways to interact with software. “Universal” is doing a lot of work in that pitch, and $700 million is a staggering Series A for a company that hasn’t shown anything yet.

The timing is worth noting. Hark’s funding comes the same week Google used its I/O conference to push hard on AI agents, a vision that left many observers confused about what Google was actually proposing. TechCrunch described Google’s agent pitch as “the most promising introduction at I/O” and also “the most confusing.” That’s the market Hark is entering: lots of excitement about AI that can do things on your behalf, and very little clarity about what that looks like in practice or whether consumers actually want it.

The fundraise itself is unusual. Series A rounds this size are rare, especially for companies operating in stealth. It signals that Hark either has unusually strong backing from a small number of large investors, or that its founders have track records compelling enough to command that kind of capital without showing a product. The company hasn’t disclosed its investors or valuation.

The hardware bet

Hark’s plan to build custom hardware is the most concrete thing it’s revealed, and it’s also the riskiest part of the pitch. Consumer hardware is expensive to develop, harder to distribute, and notoriously unforgiving of companies that don’t get it right on the first try. The AI hardware graveyard is already filling up with devices that promised to replace your phone or computer and shipped as expensive curiosities instead.

But there’s a logic to it. If you believe AI interfaces need to work everywhere and feel native to how people actually use technology, you can’t rely on retrofitting existing devices. You need to control the whole stack. That’s the bet Apple has made with its AI features, and it’s why companies like Humane and Rabbit tried (and mostly failed) to build standalone AI devices.

The difference is that Hark seems to be planning a two-stage rollout: software first, hardware later. That’s smarter than launching with hardware out of the gate. It means the company can test its models and interface in the real world, build some kind of user base, and figure out what the hardware actually needs to do before committing to manufacturing. It also means Hark can generate revenue and gather data while it develops the hardware, rather than burning through capital on R&D with nothing to show for it.

What “universal” might mean

The “universal” framing suggests Hark is going after interoperability, the ability to work across apps and services without requiring each one to build a custom integration. That’s a real problem. Most AI assistants today are either locked into a single ecosystem (Siri, Alexa, Google Assistant) or require developers to explicitly support them. A system that can interact with any app or website the way a human would, without needing permission or API access, would be genuinely useful.

It’s also technically hard and legally murky. If Hark’s AI is automating actions inside other companies’ products, those companies might not be thrilled about it. They could block it, sue over terms of service violations, or just change their interfaces to break compatibility. That’s what happened to early screen-scraping tools, and it’s a risk any “universal” AI platform has to navigate.

The other possibility is that “universal” just means Hark’s models are multimodal and can handle text, images, audio, and video. That’s table stakes now. Every major AI lab is working on multimodal models, and several have already shipped them. If that’s all Hark means by universal, the pitch is a lot less interesting.

The silence is strategic

Hark’s reluctance to share details isn’t unusual for a well-funded stealth startup, but it does make the $700 million figure stand out more. Investors are betting on a story, a team, or a technology demo they’ve seen behind closed doors. The rest of us are being asked to wait until summer to see if the product lives up to the fundraising.

What this tells us is that there’s still enormous appetite among investors for AI companies that promise to reimagine how people interact with software. The fact that Google’s own agent pitch landed as confusing rather than compelling doesn’t seem to have dampened that enthusiasm. If anything, it might have reinforced the sense that the incumbent platforms haven’t figured this out yet, and there’s room for a new player to get it right.

Whether Hark is that player, we’ll find out in a few months.