Forget San Francisco. One of the biggest winners of the AI boom is in Boise, Idaho, the land of potatoes—and of memory chipmaker Micron.

Micron's sales tripled year-over-year to $23.9 billion in its fiscal second quarter, thanks to red-hot demand for its chips. Gross profit margins swelled to 74.4%, versus 36.8% a year ago. That's what happens when a construction boom of AI data centers causes a global shortage of memory chips and a spike in prices.

Micron is one of top three global producers (along with Samsung and Hynix) of the high-bandwidth memory that's paired with Nvidia's GPUs to train and run AI models. Companies can't get enough of the stuff. And as Micron shifts production to HBM memory instead of the less lucrative DRAM and NAND memory used in PCs and smartphones, prices are rising all around. The "memory crunch" is a curse for consumers who pay more PCs and phones, but a blessing for Micron.

As it scrambles to capitalize on all the demand, Micron is building more chip fabrication facilities. The company's capital expenditures this year will top $25 billion—higher than expected—and executives said that capex in the next fiscal year will increase by another $10 billion, according to Bloomberg. Investors didn't seem to like that, and Micron's stock fell 4% after hours. Then again, the stock is up 62% this year. And in an industry like memory chips, with a long history of severe boom-and-bust cycles, can you blame investors for taking a little profit in the boom times? —AO

Some companies are finding that AI coding tools are introducing risks that outpace the productivity gains they promise. Last week, Amazon convened a meeting after a series of outages affected its website and app. Internal documents viewed by both CNBC and the Financial Times originally cited "Gen-AI assisted changes" as a factor in a "trend of incidents"—a reference that was later deleted ahead of the meeting.

And the issues aren't limited to Big Tech. Software engineer Alexey Grigorev recently watched an AI agent destroy his entire production database—wiping out years of course data after a setup error caused the system to delete the live environment instead of cleaning up duplicates. He later said he had "over-relied on the AI agent" and removed safety checks he should have kept in place.

Some studies suggest AI code is just not ready to be rolled out at scale without proper safeguards. For example, a CodeRabbit analysis of open-source GitHub pull requests found that AI-authored code contained roughly 1.7 times more issues than human-written code. The consequences, CodeRabbit's VP warned, aren't always as visible as an outage. Across the industry, senior engineers are also bearing a disproportionate "correction tax," fixing AI output at a rate that eats into time saved.—Beatrice Nolan

Robinhood began rolling out its long awaited 'Social' feature to select users Wednesday.

“Customers will be able to follow other Robinhood traders, swap strategies, discuss market moves, and trade with clarity. Plus, they can trust that every customer profile belongs to a real person, verified through KYC,” said the company in a blog post, announcing the beta launch of the tool.

Robinhood is taking a go-slow approach to the feature, which it announced last year, in part due to regulatory uncertainty over copy trading. The offering won’t be as free-wheeling as the European copy-trade popularized by eToro—Robinhood users won’t be able to automate their trades to copy a Social user, but can do so manually. And Robinhood is building the initial version of its Social offerings around a relatively small community of users with verified identities, and encouraging thoughtful discussion. The company anticipates rolling out the Social feature to all users by the end of the year.—Jeff John Roberts