The Morning Download: When the Lights Go Out on AI

Good morning. Artificial intelligence is experiencing a capacity crunch, a warning sign for the current boom, as demand grows faster than companies can access the computing power to meet it.

Just as more firms make adoption mandatory and more users rely on AI tools for everything from coding to peppering their attorneys with questions (more on that below), strains are showing.

Here's The Wall Street Journal:

The reliability of core services on the internet is often measured in nines. Four nines means 99.99% of uptime—a typical percentage that a software company commits to customers. As of April 8, Anthropic’s Claude API had a 98.95% uptime rate in the last 90 days.

“That is not normal,” said Amir Haghighat, co-founder and chief technology officer at Baseten, an AI inference startup. “Think about AWS, databases, RDS or Stripe—these need to be very resilient with a very high uptime. But that is not the world we live in when it comes to AI. That’s not the quality of service that you want to be getting from the company that’s providing intelligence for your application.”

The crunch is rippling across the industry:

• Anthropic has begun rationing computing during peak weekday hours, drawing user complaints that limits are hit within minutes.
• OpenAI scrapped its Sora video app partly to free up compute for higher-priority products. Token use on its API surged from 6 billion per minute in October to 15 billion in late March.
• GPU rental prices have spiked sharply. Nvidia's most advanced Blackwell chip now costs $4.08 per hour to rent, up 48% in two months.

Historically, the Journal notes, price increases have been among the only ways to address a supply crunch. That's a prospect that cuts both ways: painful for AI providers fighting to hold users, and equally unwelcome for enterprises trying to build the business case for mission-critical deployment.

Have you experienced the capacity crunch? Has it changed how you think about your AI strategy? Let us know. 

AI chatbots are making legal bills bigger, not smaller. Clients are now flooding their lawyers with “pages and pages of stuff" from emails and questions to patent applications, FT reports. One litigation partner described a client "barrage" so overwhelming his firm now only responds to material points at selective intervals. Patent attorneys report absorbing extra costs reviewing flawed AI-generated filings and are now raising fixed fees accordingly.

The New York Times looks at the global AI arms race where the rush to build can trump any reasonable, measured understanding of the risks—and there are many when it concerns weapons designed to move faster than human reason. The availability of the technology means startups and their investors are playing a larger role, helping call the shots. 

So far, according to the NYT, the only major accord on AI weaponry between China and the U.S. is a 2024 nonbinding pledge to keep humans in the loop over using nuclear weapons.

The AI weapons race extends beyond kinetic attacks to cyberattacks as well.

The WSJ reports that National Cyber Director Sean Cairncross is convening officials across agencies to identify security weaknesses in critical infrastructure and bolster government systems against exploitation by AI.

The move comes in response to the release of more powerful AI models, most notably Anthropic’s Mythos.

Cairncross is working with the Office of Science and Technology Policy and the National Security Council and last week officials convened a call with executives including Anthropic’s Dario Amodei and OpenAI CEO Sam Altman. Chief executives of the biggest banks have also met with Washington officials.

The activity comes as the White House engages in a series of mixed signals around AI deployments, encouraging light-touch regulation while actively trying to ban Anthropic from federal agencies.

Yet, it seems like the U.S Defense Department’s effort to stifle Anthropic has not hurt its rep in the enterprise.

Almost one-third of U.S. businesses paid for its tools in March, reflecting a rise of more than six percentage points from February, the FT reports, citing data from Ramp.  Rival OpenAI, still leads in business adoption with 35%, but growth has been flat, according to Ramp.

Apple's smartglasses move. Apple is getting into the smart specs game with plans to introduce a product in late 2026 or early 2027. Like Meta’s Ray-Ban wearables, Apple’s smartglasses will boast a camera for snapping photos and shooting video and will be synced with a smartphone, in this case the iPhone, for further functionality, Bloomberg reports.

OpenAI CEO's home attacked. A person was taken into custody after allegedly throwing a Molotov cocktail at OpenAI Chief Executive Sam Altman’s home and making threats at the company’s San Francisco headquarters, a spokeswoman for the company said Friday. The SF Standard reports on a second possible attack Sunday.

OpenAI on Monday signed a lease for an 88,500 square foot space in London's King’s Cross area, CNBC reports.

Philippines orders Meta to stop the 'fake news.' The Philippines government is warning Meta Platforms that the continued spread of false, “panic-inducing” content on its platforms, including misleading claims around exaggerated oil price increases, could violate the country's laws, WSJ reports.

President Trump’s announcement that the U.S. military would blockade the Strait of Hormuz sets up a risky new showdown that could draw American forces into a prolonged struggle to control the strategic chokepoint while compounding the global economic damage caused by the conflict. (WSJ)

Prime Minister Viktor Orbán suffered a landslide defeat in Hungary’s election on Sunday, ending the 16-year rule of a politician who had become a standard-bearer for populist right-wing leaders worldwide. (WSJ)

President Trump lashed out at Pope Leo XIV following the pontiff’s public condemnation of the war in Iran, accusing the leader of the Catholic Church of being weak on crime and catering to liberals. (WSJ)

The Internal Revenue Service has shed thousands of enforcement workers since President Trump returned to office.  Lawyers say they see more taxpayers and tax-shelter promoters eager to cut corners or cheat. (WSJ)

The WSJ Tech Council brings together CIOs, CTOs and CISOs advancing innovation and shaping the future. Join this trusted community where tech executives connect with peers to explore emerging trends and gain the perspective they need to stay ahead of disruption.

Request Information

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

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