Here is one way to measure how fast the AI game can change: Groq, the startup founded in 2016 to build chips and take on Nvidia, is now worth about half what it was last September.
On Monday, August 17, 2026, the company will announce a $350 million funding round led by Disruptive, a Dallas investment firm.
The round values Groq at $3.5 billion. A Groq representative says Nvidia is also expected to invest, though Nvidia has not responded to a request for comment.
For a private company, a valuation is just the price the newest investor agrees to pay. When that price drops by half, it is a signal that investors are pricing in a lot more risk.
The drop did not come from nowhere.
Months after Groq hit $6.9 billion in September, Nvidia signed a licensing agreement with Groq and brought on Jonathan Ross, the startup's founder and CEO, along with several other important staff members.
Licensing pacts like this let a bigger company take a startup's people or technology without buying the whole business. Groq is not the only one going through it.
Meta Platforms and Google have also pulled talent from AI startups this way instead of taking over the companies. It is an unusual way to hire, and more than one tech giant has used it recently.
For a deep-pocketed tech giant, it is a fast way to bring in experience without the cost of a full acquisition. For the startup, it can mean losing the people who built the company.
With its founder gone, Groq has changed what it does. It now runs data centers that rent out computing power for AI software, a job called inference.
In plain terms, inference is the step where a trained AI model actually answers a question or makes a decision. It is different from training, which is the long, heavy process of teaching a model how to respond.
It is the part of AI that keeps working long after training is done, and that takes a constant supply of computing power. That is why the money keeps flowing.
In June, Groq announced a $650 million raise to fund the data-center push. That round reset the company's valuation, but Groq did not disclose the new number at the time.
Some of the latest funding will go toward lifting total data-center capacity beyond 200 megawatts by next year. Alex Davis, Groq's executive chairman and the founder of Disruptive, put the strategy in simple terms.
"Inference will without a doubt become the largest and most critical layer of AI infrastructure," he said. He added: "We will be focused on supporting the most important model makers."
The company did not crash into a wall.
It made a deal, lost its leader, and changed its business model. All of that happened in about a year.
The same forces show up in public AI stocks: a licensing deal, a founder heading out the door, a rival taking an ownership stake. Any one of those can move the numbers.
Groq is spending real money to prove its new story works, and the next year will show whether it does. For your portfolio, the bigger point is that there is no settled AI pecking order yet.
The winners next year may not look like the winners from last year. A market that can change this fast still has room for surprises.