What's Nvidia GTC got to do with PR? Well, a whole lot as it turns out.

Jensen Huang didn't come to GTC this year to talk about chips. He came to declare the new battleground, and that battleground is inference. Huang's message was that we've hit an "inference inflection point" where AI value is created not in training, but in how models generate tokens, reason, and do work.

If you work in communications, it's easy to scroll past a GPU keynote. But what Huang described has a direct impact on how brand authority now works - the inference layer (the point at which AI systems generate answers) is where brand perception will be built.

For the past decade, the game was visibility: SEO, earned media, paid, social – all designed to intercept a buyer mid-journey.

That changes now.

When a B2B buyer asks ChatGPT, Perplexity, or Gemini "which platforms should we use for competitive intelligence?" they're not just browsing, but delegating some of the research and thinking. And the shortlist they get isn't based on your latest campaign. In PR, only being as good as your last piece of work is not the case any longer. It's based on a model's inference across months and years of authoritative signals: media coverage, analyst commentary, publications, backlinks, social – the patterns across all of it.

That is your influence infrastructure. It's the signal models draw on during inference - and every PR needs to get an understanding of it. We have an advantage in PR - that's because earned media is one of the most durable signals that survives inside the inference layer.

The Britannica signal: you need to be part of the answer or you're invisible

Britannica and Merriam-Webster's lawsuit against OpenAI highlights a deeper issue: inference can extract value without attribution. These sources built authority over centuries. AI used it to generate answers, and they got nothing.

Now apply that to your brand. If your thought leadership and coverage shape AI answers, but your name never appears, what was it all for?

Don't flood the internet

The PR industry has been having this debate internally too - in a broader sense than LLMs. When Sir Martin Sorrell declared on BBC Radio 4 that "there's no such thing as PR anymore," PRCA CEO Sarah Waddington CBE, pushed back. And the PRCA followed through: in February 2026, the PRCA published a new definition of the profession, prompted by a recognition that too many still understand PR primarily as media relations or publicity. The new definition repositions it as "a strategic discipline focused on trust, insight, and long-term success."

The direction is right. But it still leaves the operational question unanswered: in an era where AI systems are the primary research tool for B2B buyers, what does building trust actually look like at the inference layer? That's the gap Agentcy addresses.

Inference-era communications: what actually works

The comms teams that will win at the inference layer aren't the ones sending the most press releases. They're the ones building what we call an influence infrastructure — a durable, consistently reinforced signal across every channel that shapes how models understand "who matters in this category."

The PRCA's new definition of PR, focused on trust, insight, and long-term value, is directionally right. But the real question is operational: What does trust look like when AI is the intermediary? The winners won't be the loudest, but the most consistently legible to models.

That means:

• Earned media that creates citation patterns, not just volume
• Analyst and third-party validation
• Consistent narrative across channels
• Monitoring your AI footprint, not just media coverage

The GTC takeaway for comms leaders

Jensen Huang didn't mention communications. But he didn't need to. He described inference as the central value-creation layer of AI – where answers are generated and decisions are shaped. That's the biggest shift in brand authority since PageRank.

The companies that understand this will own the narrative. The ones that don't will be invisible.

As an industry, we need to stop measuring coverage volume and start diagnosing influence infrastructure.