Two AI stories broke this week that tell you a lot about where this industry is right now. One's about money pouring in. The other's about the con artists trying to ride the wave. Let's get into both.
Nvidia Eyes a $30 Billion Stake in Perplexity
Nvidia is in talks to invest in Perplexity at a valuation north of $30 billion, according to The Information. That's a jump of more than 50% from the roughly $20 billion mark Perplexity hit in its last raise about a year ago.
The number that jumps out to me isn't the valuation though. It's the revenue. Perplexity's annualized revenue has gone from under $250 million at the start of the year to over $750 million now. That's real growth, not just hype riding on a headline.
Why it matters: Nvidia doesn't write checks for fun. When the biggest chip maker on the planet wants equity in an AI search company, it tells you Nvidia sees Perplexity as more than a customer buying GPUs. It's a bet that AI search and AI agents are going to be a real business, not just a feature bolted onto Google.
Robert's take: I like this move a lot more than another mega round for a chatbot nobody can tell apart from the rest. Perplexity actually built something people pay for and use every day. Still, a 50% valuation jump in a year off one round of talks is the kind of number that makes me want to see a couple more quarters of revenue before I get too excited. Growth is real. Whether it's already a $30 billion business is a different question.
OpenAI Catches Russia Using ChatGPT to Fake a Think Tank
OpenAI banned a batch of ChatGPT accounts tied to Russia after finding they were running a covert influence operation. The setup was sneaky. A fake think tank, a website full of copied academic papers, and a made up "sovereignty index" built to make Russia look good.
The operators used VPNs to get around OpenAI's block on Russian access and prompted the model in Russian to write social posts, then had it scrub any language that might give away where the posts came from. The stuff got pushed out on Substack, Telegram, X, Facebook, and LinkedIn.
Why it matters: the actual reach was small. Most of these posts barely got any views. But OpenAI made the right call by pointing out that audience size isn't the point. The point is the machine they built. A fake research outfit with a real looking website is exactly the kind of thing that gets picked up and cited later by someone who never checks the source.
Robert's take: this is going to keep happening and it's not really an OpenAI problem, it's an everybody problem. Any model good enough to write convincing prose can be pointed at building fake credibility for whoever's paying. The good news here is OpenAI actually caught it and published the writeup instead of burying it. That's the bar every AI company needs to clear, not just building guardrails but being straight with the public when someone finds a way around them.
Two different stories, same underlying theme. AI is now real money and real infrastructure, which means it's also a real target for people looking to game the system. Keep an eye on both sides of that coin.