Berkshire Hathaway has lined up its first major acquisition under CEO Greg Abel, the Canadian executive who took over from Warren Buffett at the start of the year. The $1 trillion conglomerate agreed to pay $6.8 billion in cash for homebuilder Taylor Morrison.

Investor reaction could be summed up in the words of Abel’s Canuck compatriot Shania Twain, who famously sang, “that don’t impress me much.” Berkshire shares fell 1% on Monday, the first day of trading after the deal was announced on Sunday. But some analysts took a different view, noting the deal’s DNA — cash, real estate, an industry the company knows — has all the hallmarks of a classic Berkshire bargain.

A Greg-arious Offer

With the stock market chasing record after record, Berkshire has been highly selective with its massive cash stockpile in recent years. The company ended the first quarter with an unprecedented $381 billion in dry powder. January saw the close of the seemingly anomalous $9.7 billion acquisition of Occidental Petroleum’s chemical business, a deal struck under Buffett that represented Berkshire’s biggest acquisition since 2022.

The Taylor Morrison acquisition would be the first major, multibillion-dollar transaction under Abel, who pledged to maintain Buffett’s stoic commitment to value-hunting and said last month that he had a list of takeover targets he would move on at the right price. His former boss, who remains company chairman, emphasized that the latest deal is indeed his successor asserting himself. “I never talked to the CEO,” Buffett told CNBC’s Squawk Box. “He has launched.”

With that launch, Berkshire is expanding a well-established corner of its portfolio: It already owns a residential real estate brokerage as well as the Clayton Homes manufactured housing builder; Abel said he plans to unify Taylor Morrison and Clayton into a “combined platform.” While the market offered a humdrum reaction, it may be overlooking underlying value:

• Last year, Taylor Homes made $7.8 billion in revenue and completed 12,997 home closings.
• Essentially, Abel is betting the US housing market will soldier through affordability headwinds, with UBS analysts characterizing the deal as a “vote of confidence in the mid-long term outlook for the homebuilding industry.” UBS estimates the US housing market is “underbuilt” by roughly 7 million units.

Home Run: Berkshire’s braintrust is not alone in seeing an opportunity in homes. Last month, Tokyo-based Sumitomo Forestry closed a $4.5 billion cash deal for Nevada-based homebuilder Tri Pointe. Earlier this year, Dream Finders Homes tried to buy Beazer Homes for $700 million, an offer that Beazer’s board rejected, claiming it “significantly undervalued the company.”

Move aside, Mac Mini. Nvidia is making PCs specifically designed for running AI agents, CEO Jensen Huang announced Monday at a tech conference in Taipei.

Nvidia is introducing a new chip, the RTX Spark, to power its small-but-mighty laptops (the smallest model weighs three pounds and measures 14 millimeters). The chip, which Nvidia called “the most efficient PC chip ever built” will eventually become the backbone for 30 laptop models and 10 desktops. And to build its new fleet, Nvidia will assemble the Avengers of the PC world: Dell, Lenovo, Microsoft, HP, Asus and MSI.

Beyond the Chat

While asking ChatGPT to rate your facial symmetry from 1 to 10 is fun, the AI use-case that could actually generate GDP happens off-chat. Huang said AI agents running on Nvidia’s laptops could be told, for instance, to monitor a software engineer’s project and automatically fix issues. They’re also often used for boring tasks like reading and replying to emails or sorting through files.

But while a workforce full of AI agents isn’t going to ask for better health insurance or unlimited soft drinks in the breakroom, they rack up costs of their own. Users are struggling to find tech that can handle workloads in a cost-effective way:

• Running AI agents around the clock can consume hundreds of millions, or even billions, of tokens (units of data) a week. When users outsource the work to cloud providers like Amazon Web Services, they quickly rack up costs. Running AI locally on a powerful PC saves tokens.
• Home users scooped up Mac Minis this year as a creative and cost-effective solution for running AI agents locally, causing stores like Best Buy to sell out of the little silver hockey puck and Apple to raise its price. Nvidia’s laptops could meet that overwhelming demand if users are willing to pay a little (potentially a lot) more.

Work From Home: Nvidia’s push into PCs gets back to the company’s consumer-electronics roots. While supplying chips to data centers now makes up the majority of the company’s revenue, putting its name on the desks of AI’s most prolific users could earn Nvidia something priceless: clout.

Anthropic is getting ready to go public. But is the public ready for more Anthropic?

On Monday, the $965 billion artificial intelligence startup confidentially filed for an initial public offering. The news was hardly surprising, but comes as a new Bain & Co. survey of corporate America, seen by Bloomberg over the weekend, reveals this spring’s Great Claude-doption has been far from game-changing.

Token Disclaimers

A few months of heavy AI use has the world at a crossroads: Sure, AI can accomplish plenty of tasks, but at what cost? Last month, Uber Chief Operating Officer Andrew Macdonald said it’s getting harder to justify AI costs, adding that there isn’t a clear link between AI token spend and the creation of new customer features; in April, Uber’s technology chief said the company had already blown through its entire Claude budget for the year. Microsoft, meanwhile, began canceling employee Claude Code licenses last month in favor of its internally built GitHub Copilot CLI tool, a move that appears at least somewhat motivated by cost-cutting.

In May, Reuters reported that Starbucks canned a not-ready-for-primetime AI-powered (though non-Anthropic) inventory management tool after just nine months, and Bloomberg reported on Monday that Walmart is placing a token cap on its employees’ AI usage after high demand.

Those are just a few anecdotes, but the Bain survey of nearly 1,000 companies that pull in more than $100 million in revenue may provide enough data points to prove the trend is pretty concrete:

• About 40% of the surveyed companies achieved cost reductions of 10% or less from AI use, with most having expected much more; just 4% of surveyed companies found AI use led to cost savings of more than 30%. “The technology worked. The value didn’t arrive,” Bain researchers wrote in the report, according to Bloomberg.
• Still, 44% of respondents, the largest share, said they were approving their next wave of AI spend with funding based on targeted savings, rather than realized savings. “It is a circular bet with a structural leak,” Bain warned in the report.

On Claude Nine: Of course, there are plenty of reasons Anthropic is shaping up to be a blockbuster IPO. “The reality of AI right now is that it only works for coding,” Ali Ansari, CEO of model training firm Micro1, told Axios. Anthropic’s Claude tool remains the favorite among coders and has provided the company with a comfortable revenue floor; the company has recently begun telling investors that it’s on track to be profitable in the current quarter. At the very least, that’s more than this year’s other two expected mega-IPOs, SpaceX and OpenAI, can say.