WSJ What’s News - Token Costs Have Execs Rethinking AI Rollouts

Episode Date: July 31, 2026

A.M. Edition for July 31. Earnings from AI hyperscalers this week showed rising demand for their cloud services, with Amazon’s CEO forecasting corporate AI use was still in its ‘early stages.’ But as EY’s Dan Diasio tells us, rising token costs are leading many businesses to question their broader AI approach. Plus, the U.S. says Hamas has agreed to a broad plan to disarm in exchange for an eventual Israeli withdrawal from Gaza. And WSJ reporter Margherita Stancati describes how the arrival of thousands of migrants into the Spanish territory of Ceuta from neighboring Morocco is piling political pressure on the center-left government in Madrid. Luke Vargas hosts. Sign up for the WSJ’s free What’s News newsletter. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Is Hamas ready to lay down arms? Washington says yes, we'll get the latest. Plus, Spain deploys its military after thousands of migrants pour into its territory from neighboring Morocco. And as their employees burn through tokens, executives rethink their AI rollouts. 82% of corporations are now concerned about the token costs that they're seeing inside their business. for many corporations, the era of unlimited AI access is ending. It's Friday, July 31st. I'm Luke Vargas for the Wall Street Journal, and here is the AM edition of What's News. The top headlines and business stories moving your world today.

U.S. officials say that Hamas and other Palestinian militant groups have agreed to a plan that would see them disarm and seed power in Gaza if Israel withdraws from the enclave and allows in humanitarian aid. The announcement is the culmination of months of diplomacy between the U.S., Israel, Hamas, and Middle East mediators, as well as the Board of Peace created by President Trump to oversee Gaza's transition. However, Journal Middle East reporter Feliz Solomon says that neither Hamas nor Israel has explicitly agreed to the plan. A member of Hamas's negotiating team told the Wall Street Journal that the group had agreed to place its weapons in a storage facility overseen by the National Committee for the Administration of Gaza, which is a Palestinian technocratic group that ultimately will be charged with taking over the administration of the Gaza Strip. They didn't say that they had agreed to fully disarm.

This is a sticking point because Israel has said that they demand the full disarmament of Hamas before they would withdraw from the Gaza Strip. A lot of the details still need to be worked out, such as when and how the disarmament would take place, how long it would take, what exactly would be handed over and what would happen to those weapons. The Palestinian police force

that is meant to actually collect and decommission the weapons from Hamas has not been fully formed. The International Stabilization Force, which is a new peacekeeping force that was established by Trump's Board of Peace, is also still its very early state. There are very few members of the ISF so far, and none of them have been deployed to Gaza.

And it could actually take many months to stand up the peacekeeping force and the police force that will actually provide the personnel to make this really happen. Still, officials said the agreement could mark a new era for Gaza. If Hamas does disarm, it would be a huge development. This has long been a sticking point. They have resisted this from the very beginning because it's resistance. is the entire ethos and identity of Hamas. Disarmament would clear the way for the rest of Trump's peace plan to be rolled out.

As long as Hamas is still present as a political and military force in the Gaza Strip, the new Palestinian technocratic committee that is supposed to come in and govern Gaza cannot come in and begin its work. And as long as that doesn't happen, reconstruction cannot fully begin. Mediators say that Israel has yet to approve the agreement, and the talks with Hamas will continue today in Egypt. Spain is deploying the military to its North African territory, Ceuta, after thousands of migrants poured across the border from neighboring Morocco yesterday. Footage showed large crowds of mostly young men climbing over fences and swimming around breakwaters, and authorities say that at least 10 people died trying to reach the tiny enclave by sea.

The crisis didn't actually start on Thursday. It started a bit earlier in recent days. Following a ruling by Spain's Supreme Court, it basically made it harder for migrants who reached the territory of Ceuta or another similar territory called Melilla to be returned to Moroccan territory. Reporter Margarita Stancati says that both countries have since reached a tentative deal to return migrants who entered Spain illegally. But the pressure on the center-left government in Madrid continues to rise. The migrant crisis in Saota is a big political problem for Prime Minister Pedro Sanchez, who has been a lot more lenient towards illegal migration than many of his European counterparts. The migrant crisis that is unfolding in Saota, combined with Sanchez's relatively pro-immigration posture,

has given political ammunition to the Spanish Premier's political rivals. So we've seen people like Elon Musk commenting on the situation on social media, calling it an invasion. We have the right-wing Prime Minister of Italy, Georgia Miloni, saying that Italy is actually considering suspending the Schengen Agreement, which essentially allows the free movement of people within Europe because of what's happening in SELDA.

Rogue AI models have struck again. Anthropic said that software it was testing managed to get out onto the internet and hack unsuspecting companies in three separate incidents dating back to April. Anthropic didn't say which companies had been hit, but all three were notified of the incident earlier this week. The news comes a week after OpenAI revealed that AI technology it was testing, had broken out of a so-called sandbox, accessed the internet, and hacked the AI company Hugging Face.

That incident has rattled security researchers and AI professionals and served as a reminder of the power and unpredictability of AI systems that act autonomously. And we're exclusively reporting that Tesla is weighing the sale or spin-off of its business in China, potentially teeing it up to merge with SpaceX. We report that after Elon Musk instructed Tesla executives in recent years to put a laser between its U.S. and Chinese businesses, some have now been told to prepare for a separation. While Tesla's operations and production in China helped to transform it into a profitable global EV brand, Musk has grown concerned about its dependence on the country for battery cells and the possibility that it could lose access to chips if Beijing invaded Taiwan.

In the event of a merger, a separation could also address potential conflicts arising from SpaceX's work as a major U.S. defense contractor. Musk told investors last week that he couldn't talk about combining the companies, saying that such a move would need to be, quote, done with the appropriate process. Both Tesla and SpaceX lately have been reorganizing around AI projects. Coming up, hyperscalers are betting the farm on businesses rapidly scaling up their use of AI. But as the cost of doing so blows up budgets, is that really a given? We'll take a look after the break.

Shares of Apple have fallen off hours after it forecast lower sales growth than analysts had expected, overshadowing strong Mac and iPhone sales. That's as the AI boom has driven up Apple's costs and left it waiting in line for key device components, memory, and storage chips. One of the companies that Apple's competing with for those chips is Amazon, whose shares are up more than 10% off hours, has it reported pairing massive AI investments with accelerating cloud computing sales. Even still, CEO Andy Jassy described demand for Amazon's AI offerings

like computing power that sells to customers as being in the relatively early stages with a barbell-shaped market. On one end, he said, were generative AI apps like Claude and ChatGPT consuming tons of compute to drive rapid growth, while on the other, enterprises are using AI to cut costs and boost profitability. And between them? In the middle of the barbell is all of the current enterprise production workloads some of which are using inference in a pervasive way, but most of which aren't.

And that is going to change very significantly over time. And that will be, in my opinion, that will be the largest absolute segment. Well, that's how the hyperscalers see the market for their AI offerings. But where are their clients actual priorities? Dan Diazio is the global AI consulting leader at EY, which was just published its big AI Pulse Survey. Dan, the corporate leaders that you surveyed seem to know that the software relationships that they've gotten used to are changing, particularly how they're priced. And notably we heard this week in Microsoft's earnings on Wednesday, the company is moving from a seats-based pricing to a seats-plus consumption model. And we've heard isolated reports about businesses really struggling with this shift towards token-based pricing. Essentially, they've been kind of blowing through their AI budgets way faster than they accounted for. What are you hearing? I mean, I think for many corporations, the era of unlimited AI access is ending. Companies are now starting to put a price on intelligence in their organization,

which means they're making decisions around who should have access. Where do you start to apply these premium frontier AI models? Which departments might require more tokens, like the R&D department, might get a larger allocation than, say, in the Human Resources Department. and they're also starting to look forward to where they might be able to build a proprietary strategy. We saw that 82% of corporations are now concerned about the token costs that they're seeing inside their business and nearly every one of them says it's causing their business to start to rethink what their approach to AI scale out has been.

Now, it's not all just downside. Two-thirds of companies we see are putting controls in place to be able to monitor. monitor token usage so they can start to use that data to make allocation decisions. Just to be clear, is it the cost of AI token usage that's really the issue here? Or in some cases, is it the predictability around what those costs are going to be? I mean, it's not always clear how many tokens a specific AI use case is going to cost. And I can only imagine that when you scale that up to the enterprise level, that sort of variance could be massive. There's really three components to that total cost.

One is the amount of tokens you're consuming with a specific activity. So the higher computationally intensive areas like building out software or maybe doing a detailed study require more tokens. There's also the cost of the model that you're producing. Some of the latest and greatest frontier capabilities that are pushed out by the frontier AI firms, those latest and greatest models might be five times more expensive than some of the less capable models that are in their portfolio. So the model that you'd pick actually has a pretty big outcome on cost. And then the third bit is it does come down to the lack of predictability.

For many times, a specific task might take, let's say, 20,000 tokens. But because of some configurations on data, it might get stuck and retry again and again. And that could start to really increase the meter to several hundred thousand tokens to be able to perform that task. Finally, Dan, the sort of trillion-dollar question then flowing off of this whole discussion is, and I don't think trillion dollar is an overstatement there, is how are businesses altering their AI rollout in response to these AI costs? Especially were they to pump the break, that would have a big impact on all of this capital spending by big tech firms and how wise that spending looks several years on from now. The really interesting thing that I found from this study is that while nearly all of the companies are starting to rethink their strategy, we saw twice as a lot of the company. we saw twice as many companies that still indicate that they are accelerating versus slowing down. And that acceleration comes in the form of better prioritization of where they're investing their efforts.

And many companies are starting to rethink their ambition and their strategy to make sure that where they originally saw a business case that maybe didn't factor in what this token spend might be, they're reevaluating that to say, is that really what we should be doing to be able to power our business with AI or their opportunities to be able to move more towards revenue generation or growth activity. Dan Diasio is the global AI consulting leader at EY. Dan, thanks so much for being with us on What's News. Thanks, Luke. And before we go, heads up.

We've got a special bonus episode coming later today. In the next What's News and earnings, we'll be looking at worries about private credit at the country's biggest private investment firms. That'll be here in your What's News feed today at midday. And that's it for What's News. This Friday morning, today's show was produced by Hattie Moyer. Our supervising producer was Daniel Bach, and I'm Luke Vargas for The Wall Street Journal. We will be back tonight with a new show.

Otherwise, have a great weekend, and thanks for listening.