I don’t have permission to write the file yet. Here’s the article I’ve written for you:

HEADLINE: Profound raises $180M at $1.8B valuation seven months after last round

DESCRIPTION: The AEO startup’s rapid-fire fundraising signals continued investor appetite for AI infrastructure, while new players tackle agent safety and enterprise customization.

Profound just raised $180 million at a $1.8 billion valuation, seven months after closing a $96 million Series C. That’s $276 million in less than a year for a company building what it calls “AEO” infrastructure, though the acronym remains undefined in public reporting.

The velocity matters more than the label. Investors are still writing nine-figure checks for AI infrastructure plays, even as public sentiment toward data centers and AI deployment turns sharply negative. A New York Times poll released this week found 61% of likely voters oppose data center construction to power AI, with only 14% strongly in support.

That disconnect between capital markets and public opinion is creating strange incentives. Companies are raising at speed, but they’ll eventually need to operate in communities that don’t want them there.

Agent safety gets $40 million

Two former AI safety researchers have a different problem to solve. Artificial Intelligence Underwriting Company raised $40 million in a Series A led by Ribbit Capital to build what amounts to guardrails for AI agents.

The founders include an early Anthropic employee and the former COO of METR, an AI safety organization. Their pitch: AI agents are being deployed without adequate controls, and someone needs to provide the infrastructure to keep them from going rogue.

It’s a reasonable bet. As agents move from demos to production, companies will need ways to monitor and constrain their behavior. The question is whether this becomes a compliance checkbox or actual safety infrastructure.

Ribbit’s involvement is notable. The fintech-focused firm doesn’t typically lead AI safety rounds, which suggests they see this as an enterprise infrastructure play with insurance-like economics rather than a purely technical solution.

Enterprise AI splits into custom and commodity

Salesforce and Nvidia launched Koa, a reasoning model built on Nvidia’s open-weight Nemotron and trained specifically for sales, marketing, and customer support. It’s exactly the kind of thing that should worry OpenAI and Anthropic.

The big AI labs are selling general-purpose models. Salesforce is building a model that already knows how to write follow-up emails and route support tickets. For an enterprise buyer, the choice is obvious.

This is the real competition: not who has the biggest model, but who can ship something that solves a specific business problem without six months of prompt engineering and fine-tuning. Nvidia’s involvement gives Salesforce the compute and model infrastructure. Salesforce brings the customer data and years of CRM workflow knowledge.

Meanwhile, Meta is bundling its AI tools into a new subscription called Meta One, combining expanded access to Meta AI with premium features across Facebook, Instagram, and WhatsApp. It’s part of a broader push to turn AI access into a revenue stream rather than a feature.

The timing is deliberate. Meta just announced camera-free smart glasses (codenamed Luna, possibly launching at Connect next week) after a PR crisis around its camera-equipped Ray-Bans. The Luna glasses will reportedly include six microphones for AI chat instead of video recording capability.

That’s a tacit admission that the privacy concerns are real enough to warrant a separate product line. It also suggests Meta sees voice-based AI interaction as valuable enough to ship without the camera tech that caused the backlash.

What it means

The funding environment remains strong for AI infrastructure, but the paths are diverging. Profound is raising at speed with minimal public detail about what it actually does. AIUC is building safety infrastructure that might become mandatory. Salesforce is customizing models for specific enterprise workflows.

None of these companies are trying to build AGI. They’re building the scaffolding around other people’s models, which might be the better business anyway. The labs are burning billions on compute. The infrastructure companies are raising smaller rounds to solve narrower problems.

Jensen Huang said this week that AI doesn’t need regulation because “safety can be engineered by each AI product maker.” That’s self-serving, but it’s also creating a market. If every AI product maker needs to engineer safety, someone has to sell them the tools to do it.

OpenAI, Anthropic, and Google DeepMind have been in talks on AI safety for weeks, even as the Trump administration dismisses safety concerns in favor of keeping pace with China. China, meanwhile, is skeptical of any deal that keeps US companies ahead.

The result is a fragmented approach: some companies are trying to coordinate on safety standards, others are saying safety is someone else’s problem, and investors are funding both the infrastructure and the guardrails.

It’s a strange way to build an industry, but it’s the industry we have.