Good morning. Can IBM engineer another comeback?
Big Blue, founded in 1911 and the oldest tech company in this year’s Fortune 500, has certainly had its fair share. But the tech planets seem to be aligning—data, AI, un peu de quantum—for another inflection point.
My Fortune colleague Sharon Goldman has the scoop on what CEO Arvind Krishna has been up to since his appointment to the top job in 2020. The company’s shares are higher than they’ve ever been—in my lifetime, anyway—so something’s surely clicking. Be sure to give the story a read.
Today’s tech news below. —Andrew Nusca
Want to send thoughts or suggestions to Fortune Tech? Drop a line here.
Meta goes nuclear
Meta has agreed to buy power from Constellation Energy’s Illinois nuclear plant in a deal that will fuel the Facebook and Instagram parent’s power-hungry AI operations.
The 20-year agreement for 1,121 megawatts from Constellation’s Clinton plant will begin in mid 2027 when a 10-year state subsidy for the facility expires.
How much power is 1,121 megawatts, you ask? Estimates vary greatly, but roughly enough power to support half a million homes, depending on their HVAC use.
The agreement includes investment to increase output of the facility and may lead to a new second reactor, for which Constellation already has federal approval.
Meta didn’t share financial details about the deal. Constellation, it’s worth noting, is the largest U.S. nuclear operator.
For those keeping score at home, Meta is far from the first tech giant to strike a nuclear power deal in a bid to support AI ambitions (and curb spiraling AI emissions).
Late last year, Microsoft signed a similar deal for power from Pennsylvania’s Three Mile Island nuclear plant; Amazon and Alphabet have invested in small-scale nuclear reactors; OpenAI is maneuvering to do business with Oklo, Sam Altman’s “fast reactor” startup. —AN
Why Nvidia is trashing billions of dollars’ worth of AI chips
Nvidia’s blockbuster quarterly earnings last week came with a big dose of negative news: A $4.5 billion write down on chips originally destined for customers in China that will ultimately go undelivered.
The company had developed the so-called H20 chips, which are less powerful than its top-of-the-line semiconductors, to comply with Biden administration regulations against sending technology to foreign adversaries that could help their AI efforts.
The Trump administration, however, went a step further in early April and banned exports of even second-tier chips.
The details around the decision to write the value of the chips to zero—rather than sell them to other customers—wasn’t explained.
But the H20 was built specifically for the China market under the old export rules. Because of the chip’s design and limited capabilities, it may be difficult (and costly) to use in other countries.
Even if it could sell the chips by cutting their price, Nvidia would risk damaging its image as a seller of top-tier innovation.
So what will happen to the H20 stockpile? Supply chain experts interviewed by Fortune suspect it will be discarded.
“These chips will meet the ‘cool lava lamp’ you got from your aunt in a landfill somewhere,” Alan Amling, a professor of supply chain at the University of Tennessee’s Haslam College of Business, told Fortune. “With so many other growth opportunities, the opportunity cost of repurposing, retesting, and requalifying these chips was obviously too high a bar.” —Alexandra Sternlicht
Yoshua Bengio: AI is engaged in deception, cheating, and lying
One of the ‘godfathers of AI’ is warning that current models are exhibiting dangerous behaviors as he launches a new nonprofit focused on building “honest” systems.
Yoshua Bengio, a pioneer of artificial neural networks and deep learning, has announced LawZero, an organization focused on building safer models away from commercial pressures.
In a blog post announcing the new organization, Bengio said LawZero had been created “in response to evidence that today’s frontier AI models are growing dangerous capabilities and behaviours, including deception, cheating, lying, hacking, self-preservation, and more generally, goal misalignment.”
The nonprofit is building a system called Scientist AI designed to serve as a guardrail for increasingly powerful AI agents.
AI models created by the nonprofit will not give the definitive answers typical of current systems. Instead, they will give probabilities for whether a response is correct.
Bengio told The Guardian that his models would have a “sense of humility that it isn’t sure about the answer.”
In the blog post announcing the venture, Bengio said he was “deeply concerned by the behaviors that unrestrained agentic AI systems are already beginning to exhibit—especially tendencies toward self-preservation and deception.”
He cited recent examples, including a scenario in which Anthropic’s Claude 4 chose to blackmail an engineer to avoid being replaced, as well as another experiment that showed an AI model covertly embedding its code into a system to avoid being replaced.
“These incidents,” Bengio said, “are early warning signs of the kinds of unintended and potentially dangerous strategies AI may pursue if left unchecked.”—Beatrice Nolan
More tech
—1,000 people have left CISA. Departures from the nation’s top cybersecurity agency have skyrocketed during the second Trump administration.
—Cohere raising $500 million. The Toronto AI startup would be valued at $5.5 billion.
—Rippling vs. Deel continues. Deel alleges its rival directed one of its employees to “pilfer” assets by posing as a customer.
—Qualcomm patches critical flaws. Dozens of chips are affected; three vulnerabilities may be in use by hackers.
—HPE beats Q2 estimates. Quarterly revenue was $7.63 billion; it expects reduced tariff impacts for the rest of the year.
—Social media ban…banned. A federal judge rules that Florida cannot enforce a law that requires social media platforms to block kids from using their platforms.
—Photoshop comes to Android. The iconic Adobe app is now available in beta in the Google Play store.
Endstop triggered
