October 7, 2026
TechCrunch confirmed that Nous Research, maker of the Hermes agent, has raised a $90 million Series B and now carries a $1.5 billion valuation, alongside a launch of AI agents aimed squarely at business users. A year ago, a number like that for an agent company would have seemed speculative. Today it reads as a market correction: investors are pricing in that agents, not just chat interfaces, are where enterprise AI spend is actually heading.
What should a business reader take from this? Mostly that the agent market is getting crowded and well-capitalized fast, which is good news if you're shopping for one and bad news if you're trying to pick a durable vendor. Hermes joins a field that already includes OpenAI's Dots, which we've argued is less a software replacement than a tax on your existing stack, and a wave of well-funded agent infrastructure plays like the one behind Lambda's recent $4 billion round. The honest question for any buyer isn't whether an agent can do a task once in a demo -- it's whether it can be audited, governed, and trusted to run unsupervised inside a real workflow. That's the gap a lot of these raises are quietly betting someone else will solve.
Microsoft used the same news cycle to reveal specs and pricing for its Surface Laptop Ultra, part of a new generation of AI PCs built on Nvidia chips and a revamped Windows 11 designed to run models and agents locally. This is Microsoft trying to solve a real bottleneck: most of the agent and copilot features companies have been sold on over the past two years actually run in the cloud, which means latency, recurring inference costs, and data leaving the device. Local AI PCs promise to flip that, at least for some workloads.
For businesses evaluating fleet refreshes, the practical question is whether the premium for Nvidia-powered silicon actually translates into fewer cloud API calls and tighter data control, or whether it's mostly a marketing wrapper on a faster laptop. There's a legitimate skeptic's case here: on-device AI PCs have been promised before, and most knowledge-work AI tools still lean on cloud models for anything beyond autocomplete. Still, if local inference genuinely matures, it changes the calculus for teams worried about where sensitive data lands when they run AI-assisted tools -- a concern we take seriously enough to document in our own Security practices and Trust Center. Hardware won't fix governance problems on its own, but it does lower the cost of trying agents without shipping every keystroke to a third-party server.
Put the Nous Research raise next to Microsoft's AI PC launch and a pattern emerges: 2026 is the year the AI industry stopped asking 'can an agent do this?' and started asking 'what does it cost to run this reliably, at scale, on hardware and capital that actually exists?' That's a healthier question, but it also means business buyers should expect more vendors, more pricing models, and more hardware refresh cycles pitched as mandatory. Our own view is that the businesses who win this round won't be the ones who adopt the flashiest agent first -- they'll be the ones who treat agent adoption like any other workflow decision, with clear ownership, role-based approval built in, and a real plan for what happens when the agent gets something wrong.
If you're a business leader weighing an agent platform or an AI PC refresh this quarter: which one actually solves a problem you have today, versus one you're worried about having next year?
Sources