The Morning Download: Groq’s Founder Foresees Wild Days Ahead

What's up: Get ready for another record year for chips; Hong Kong calling; Big AI deals closed out 2025; What are companies actually doing with AI? Our reporters talked it out.

Happy New Year and welcome back. We’re starting 2026 by picking up a big AI story that closed out last year. On Christmas Eve, Nvidia announced that it had struck a $20 billion non-exclusive licensing deal to use Groq’s AI inference technology. Groq’s chips, known as language processing units, are designed to run previously trained AI models.

That process, known as inference, is taking off and is expected to dominate the use of AI starting this year. Groq founder Jonathan Ross, who previously started development of the tensor processing unit chip at Google, joined Nvidia as chief software architect.

From the WSJ’s coverage: “Groq had been projecting revenue of $1.4 billion in 2026, up from about $500 million this year and about $90 million in 2024, people familiar with its financials said.”

I met with Ross in mid-December. Here are edited highlights of our conversation, in which he offered insight into why AI is making it easier for people to experiment with wild ideas.

WSJ Leadership Institute: Many people are concerned about the threat of excess capacity in AI computing. Are you?

Ross: In Silicon Valley, we normalized around the idea of labor scarcity. Every Silicon Valley company is starved for talent. They don't necessarily have all the talent that they want, they want to get as much of the good talent as they get, right? And so I think, as we start to realize that AI is more comparable to the labor market, we're going to stop treating it like selling widgets. We're going to start treating like, what are the rules on labor markets and how does that work? I'm going to start applying that logic to AI. The basic rule of the labor market is that labor gets put to use.

WSJ Leadership Institute: What are the implications of that AI labor?

Ross: The dynamics have changed. It used to be if you invested in a startup, that startup had a prototype three to six months later. Now, if you don't have that like a week or two after funding, because you're using AI code assistance, people are like, what's wrong with you? You're going to die. The expectation for speed, it's crazy … It doesn't take nearly as many people to do something, so more things are going to be built.

Airbnb is an example of an aberration of a company. There weren't like 50 people going, “You know what? We really need to rent out our house to strangers” … We're going to start seeing a lot more of those really sort of crazy ideas because you don't need to convince a lot of capital to come to your startup in order to build. So people are going to get to try wilder ideas … If you look at the rate of startups being funded, we have never seen anything quite like this.

WSJ Leadership Institute: What’s next for competition in the AI chip market?

Ross: One of the things that happened when creating that TPU at Google is everyone got a false sense of how easy it was to build an AI chip. If I went to you and I said, “You know what? Google search is great. I am going to build my own search instead,” you would be like, “You’re nuts.” Building a chip is harder. And so you’re going to see a lot of people dropping off. The TPU is a bit of an aberration.

AI is evolving so quickly, that it’s a challenge to synthesize all of the changes. The WSJ Leadership Institute team got together in the waning days of last year to unpack the forces shaping corporate AI adoption and what to expect in the year ahead. Here are highlights of the conversation among Isabelle Bousquette, Belle Lin, Tom Loftus and myself. Click here for the full conversation.

AI adoption is real, but uneven: CEOs say they’re “all in” on AI," yet deployment is often vague, incremental, and uneven across organizations. At the same time, AI is quietly boosting efficiency rather than radically reshaping how companies operate.

ROI matters: The payoff is measured less in dollars than in time saved and headcount avoided. Leaders see value in productivity and “cost avoidance,” even if near-term financial returns are hard to pin down. "The benefits they see are often expressed in nonfinancial terms. It’s like, well, of course it’s worthwhile. It’s saving us time and that’s really useful."

‘Agentic,’ in the wild: We all agreed that there is a lot of hype and fully autonomous, multi-step agents are still rare.  Although, as we reported earlier, companies like Walmart and Bank of New York are pushing ahead with early implementations.

Nvidia stands apart: Nearly every AI discussion circles back to Nvidia and this one was no different. My colleagues cited the AI chip maker's learning culture and its leadership as the gold standard for operating in the AI era.

More on what to expect from 2026 below.

Expect supply chain challenges.

Goldman Sachs estimates that Nvidia will sell $383 billion in GPUs and other hardware in the 2026 calendar year, an increase of 78% over the prior year, the WSJ reports. Analysts polled by FactSet estimate that the combined sales from Nvidia, Intel, Broadcom, AMD and Qualcomm in 2026 will top $538 billion.

But expect rising demand to exacerbate shortages in certain components. Memory chips supplies, already tight, are expected to get tighter in 2026 as more data centers switch from AI training to AI inference workloads, which are more likely to be “memory-bound,” or constrained by having enough accessible memory capacity, WSJ reports.

“We’re significantly short of our customers’ needs and it’s going to persist for a while,” Sumit Sadana, chief business officer of memory-chip maker Micron Technology, told the Journal. 

Hong Kong stocks in 2025 posted their best performance since 2017 in percentage terms, thanks to a tech rally fueled by artificial intelligence, WSJ reports.

Alibaba Group and Tencent Holdings rose 73% and 43.65%, respectively. Baidu, which operates the AI chatbot Ernie Bot, gained 59% over the year. Shares of SMIC, China’s largest contract chip maker, more than doubled in Hong Kong this year, while Moore Threads and MetaX, two Chinese AI chip startups that IPO'd last month, ended the year up more than 400% from their listing prices.

Chinese AI model developers MiniMax and Zhipu have already secured IPO slots for January 2026.

On Friday, shares of AI chip designer Shanghai Biren Technology surged nearly 76% in their trading debut Friday, Bloomberg reports, while Baidu said its AI chip unit Kunlunxin filed a listing with the Hong Kong stock exchange, Reuters reports.

With stock-based compensation at about $1.5 million per employee, on average, across its workforce of roughly 4,000, OpenAI is paying employees more than any tech startup in recent history, according to financial data it has shown investors.

That figure is more than seven times higher than the stock-based pay Google disclosed in 2003, before it filed for an IPO in 2004 and about 34 times the average employee compensation of 18 other large tech companies in the year before they went public, according to a WSJ analysis of data compiled by Equilar.

Related? High-profile tech investors are criticizing a proposed ballot initiative from a California healthcare union that would impose a one-time, 5% tax on the assets of those with net worths above $1 billion who lived in the state as of Thursday, WSJ reports. Consultants allied with Gov. Gavin Newsom also are running a campaign to oppose the measure.

The WSJ says Meta's Manus acquisition ranks among the most prominent U.S. tech purchases of an Asian-developed AI product. Nvidia’s licensing deal with AI chip startup Groq is a sign of growing competition for top talent and technology around AI inference.

Meta Platforms has agreed to acquire Manus, a Singapore-based AI startup founded by Chinese entrepreneurs, in a deal valued at more than $2 billion. Manus earlier made waves after debuting an AI agent that performs deep research and builds custom websites using models from firms including Anthropic and Alibaba.

The WSJ says the $20 billion licensing agreement announced last week gives Nvidia access to Groq’s specialized inference-chip technology, and prevents a GPU alternative from becoming a serious competitor.

Despite health hazards, nicotine pouches have become a go-to stimulant for a subset of tech workers who claim the products help them focus. WSJ reports that two nicotine startups—Lucy Nicotine and Sesh—earlier this year made branded vending machines for Palantir Technologies.

OpenAI has sharpened focus on improving the accuracy of its audio AI models as it develops a device that responds to spoken commands, the Information reports. 

British renewable-energy startup Octopus Energy will spin off Kraken Technologies, an AI utility management platform used by global energy retailers, at a valuation of $8.65 billion, WSJ reports.

Louis V. Gerstner Jr., the first outsider to head International Business Machines and credited with steering the company away from what looked like a possible demise in the early 1990s and refocusing it on services, died Saturday. He was 83 years old.

Condo owners are struggling with the worst market in more than 10 years. Prices for U.S. condominiums posted their biggest annual decline since 2012 earlier last year. Single-family home prices have also slowed, but they are still up from the year-ago levels. (WSJ)

President Trump threatened to intervene if Iran cracks down violently on ongoing protests, putting more pressure on Tehran as it tries to contain discontent with its spiraling economy. (WSJ)

Zohran Mamdani became the mayor of New York City on Thursday, declaring he will govern “expansively and audaciously” and promising to make good on his pledge to lower the cost of living for New Yorkers. (WSJ)

This February 10–11, technology leaders will gather in Palo Alto for The WSJ Technology Council Summit to explore the realities of enterprise AI, the evolving role of tech leadership and the urgency behind building meaningful, business-driving AI strategies. Join the Technology Council and be part of the conversations shaping the future of corporate innovation.

Request Information

The WSJLI CIO Journal Team is Steven Rosenbush, Isabelle Bousquette and Belle Lin.

The editor, Tom Loftus, can be reached at thomas.loftus@wsj.com.