Welcome to Eye on AI. Beatrice Nolan here.
I’m in Paris this week at one of Europe’s largest tech conferences, VivaTech, where concerns about “sovereign AI” are front and center.
The U.S. government’s move to abruptly shut down foreign access to Anthropic’s powerful Mythos-tier models last week has been a stark example of Europe’s technological dependence on America—and what happens if the U.S. decides to pull the plug. Conversations on stage and off have been focused on what AI sovereignty actually means and how Europe can achieve meaningful AI independence.
The other question on the table: AI spend. After two years of near-unrestrained experimentation, companies are starting to seriously question how much they’re spending on AI—and what they’re actually getting back.
Corporate leaders across industries are now being confronted with ballooning AI costs as employees attempt to obey directives for widespread AI adoption. Uber recently burned through its entire 2026 AI budget in four months, and its COO said AI spend is getting harder to justify. Last month, one consultant told Axios that a client burned through half a billion dollars in a single month after failing to cap AI usage for employees.
Even inside companies building the technology, there’s a growing sense that the free-for-all phase might be coming to an end—at least for some customers. Peter DeSantis, SVP at Amazon, said this kind of cost shock is a normal phase of new technology adoption, with companies now moving from experimentation to figuring out how to control usage and budgets.
“Just like every technology, when we first launched the cloud, some of our most successful customers were delighted by the agility… but many of them woke up one day, and they’re like: ‘Wow, we’re spending a bunch of money,” he told Fortune. “Anytime there’s a new technology, I think there’s work to be done to figure out how to efficiently use it and how to budget for it, and I think we’re going through that.”
Buyers also say they are becoming more selective. Philippe Rambach, Schneider Electric’s chief AI officer, told Fortune the focus internally had shifted toward a more deliberate focus on matching use cases to cheaper, fit-for-purpose models.
“On the solutions that we build, we are very cautious to use the right model; you don’t always need to use the latest frontier model. Quite often you can use relatively cheap models,” he said.
“The question of the cost of AI is becoming more and more important,” he added. “We need to have that under control. We need to measure it. We need to include that in our business case, business plans, and decisions.”
In other words, not every task needs GPT-5-level or Mythos firepower, and not every employee needs access to the most cutting-edge technology. This cuts against how many companies have actually rolled AI out, with many handing out licenses liberally and encouraging heavy AI experimentation.
Recently, there’s been lots of anecdotal evidence that this pressure to “just do it” when it comes to using AI has had the perverse, but predictable, result of employees using it for almost everything—including things as trivial as checking the weather. The result is increasing costs for companies. Many businesses also still haven’t figured out where AI actually delivers meaningful ROI. As one executive told Axios, people tend to automate what they dislike, not what creates value.
Now companies are shifting from exploration to optimization—ensuring AI delivers actual business value, or at least that AI experimentation doesn’t break the bank.
Or, as Rambach put it, companies may just be “coming back to reality.”
With that, here’s more AI news.
Beatrice Nolan
bea.nolan@fortune.com
@beafreyanolan
Tokens are getting cheaper, but companies are spending even more on AI as a result, top economist warns — Sasha Rogelberg
Brinker’s CIO spent years rebuilding restaurant tech. Now, the Chili’s operator is ready to explore more AI — John Kell
Cohere CEO on G7 leaders’ choice: sovereign AI or digital serfdom — Aidan Gomez (Commentary)
Citi, Ford, and Experian share their strategies for scaling AI agents — Alexei Oreskovic
ChatGPT's market share falls below 50%. ChatGPT's share of the AI assistant market has dropped below 50% for the first time, falling to 46.4% by the end of May, according to Sensor Tower's State of AI Report for 2026. Gemini now holds 27.7% of the market while Claude holds around 10.3%. The report also found that Anthropic leads the field on subscription conversion, with 13% of its users paying for a plan. OpenAI's Pentagon deal in February—which sparked backlash and calls for boycotts at the time—triggered a measurable spike in ChatGPT uninstalls, suggesting brand trust matters to AI users. Also in the report, Global AI app spending is on pace to exceed $4.2 billion in the first half of 2026. Read more in TechCrunch.
JPMorgan blocks Hong Kong staff from using Claude. The bank has removed Anthropic's Claude models from the list of AI tools available to its Hong Kong employees, the Financial Times reported, following a similar decision by Goldman Sachs. The move stems from the wording of Anthropic's licensing terms, which exclude use in Greater China—a category that some banks have interpreted to include Hong Kong. Western AI models are banned in mainland China, but international firms have historically been able to access them in Hong Kong through global contracts. The restrictions are raising concerns about Hong Kong's revival as an international financial centre. Read more in the Financial Times.
A co-lead on Google Gemini leaves for OpenAI. Noam Shazeer, a co-lead on Google's Gemini model and co-founder of Character.AI, has announced he is joining OpenAI. Shazeer has been at Google since 2000, leaving briefly to found Character.AI before Google rehired him in 2024 as part of a deal that gave it non-exclusive access to the startup's technology. He is a significant figure in the development of modern AI—a 2017 paper he co-authored is widely credited with laying the groundwork for today's large language models. His move is the latest in a series of high-profile talent shifts across the major AI labs. Read the full post on X.
AI chiefs clashed over governance at the G7. The heads of Anthropic, OpenAI, and Google DeepMind all told G7 leaders that international AI governance is urgently needed — but largely disagreed on what it should look like, according to reporting by Rowland Manthorpe from Sky News. Anthropic CEO Dario Amodei pushed for a US-led coalition of democracies to control access to the technology and isolate adversaries, including China. OpenAI CEO Sam Altman warned against concentrating power in the hands of the few, arguing that once guardrails were in place, the priority should be human liberty. Google DeepMind CEO Demis Hassabis proposed a technical standards body supported by the leading labs—something he indicated he had already been working on. Altman pushed back on that too, cautioning that the labs themselves should not have too much control. Read more from Rowland Manthorpe at Sky News.
Apple plans camera AirPods for 2027. Apple is planning to release its first AI-focused wearable—a pair of camera-equipped AirPods—in late 2027, Bloomberg reported. The earbuds will use computer vision cameras to feed visual context to Siri, letting users ask questions about their surroundings rather than capture photos or video. The product is scheduled to land alongside a second-generation foldable iPhone and a 20th-anniversary iPhone model, as part of what Apple internally describes as its biggest-ever product wave. The devices will be among the first major releases under incoming CEO John Ternus, who takes over on September 1. Read more in Bloomberg.
That's the percentage of Americans who want the legal right to interact with a human rather than an AI in medical, legal, educational, and government settings. A new Johns Hopkins University survey of more than 2,000 U.S. adults, conducted in April and May, found strong support for the so-called "right to a human" across party lines. Across various sectors, 79% of respondents said they want a human option for medical care, 76% for legal proceedings, and 74% for education.
This strong support for rules around AI extends beyond human-interaction rights. Some 75% of respondents said they want to be told when they're interacting with AI, 73% want to ban AI from using individuals' faces and voices, and 68% want labels on AI-generated images and video. As AI pops up more and more in daily life, the technology is also facing a wave of wider backlash and protests. The study found a surprising amount of cross-demographic consistency on the desire for more regulation of the technology.
"What was surprising to us in this new poll was that daily users of AI, and people who view AI positively, also want regulation," Christopher Honey, a computational cognitive neuroscientist at Johns Hopkins who worked on the survey, said.
June 17-20: VivaTech, Paris.
July 6-11: International Conference on Machine Learning (ICML), Seoul, South Korea.
July 7-10: AI for Good Summit, Geneva, Switzerland.
Aug. 4-6: Ai4 2026, Las Vegas.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。