While Microsoft didn’t offer a specific forecast for its AI capital expenditures for the next quarter or coming year, the company’s chief financial officer, Amy Hood, said that the company’s total spend will “increase sequentially, and we now expect the fiscal year 2026 growth rate to be higher than fiscal year 2025.”

Tech companies are making these ambitious plans for more capital spending under the assumption that demand for AI will only continue to grow. But some analysts are raising concerns that the AI market is a bubble and will eventually burst.

Those worries are being fueled by announcements about enormously expensive, multi-year data center projects and staggered investments. Last month, Nvidia said it would invest “up to $100 billion” in OpenAI, provided that the ChatGPT maker builds and deploys at least 10 gigawatts of AI data centers using Nvidia’s chips. OpenAI, meanwhile, said just yesterday that it was planning to develop 30 gigawatts of computing resources worth $1.4 trillion.

Microsoft has committed to putting a total of $13 billion in OpenAI, and it continues to use the company’s frontier AI models, but took a $3.1 billion hit in net income this quarter due to losses from that investment. Microsoft said that the ongoing nature of its partnership with OpenAI will result in increased volatility. Going forward, Hood said, the company will exclude any impacts from its OpenAI investment in its financial outlooks.

Microsoft CEO Satya Nadella told analysts there are two “critical” things to consider about how the company views its capital expenditures. The first is that it is finding ways to make its fleet of data centers “fungible,” or interchangeable, meaning they can be easily modified to meet changing customer demands in the future. The second is that the company is expecting to continually modernize its infrastructure.

“It’s not like we buy one version of Nvidia and load up for all the gigawatts we have. Each year, you buy, you ride Moore’s Law, you continually modernize and depreciate it, and you use software to grow efficiency,” Nadella said.

Mark Moerdler, a senior research analyst covering global software at Bernstein, says that Microsoft is “building capacity in tranches over time and can shift resources, which gives them a lot of protection.” But, he added, “Is there an overall AI bubble? [It’s] possible, and that they did not answer.”

Renewable energy sources like wind and solar generation now account for over 20% of electricity in the U.S., and keep growing after large-scale production has more than doubled since 2000.

Still, high-profile power failures illustrate persistent challenges from the lack of available capacity to provide enough energy at times of need, said Chiara Lo Prete, an associate professor of energy economics in the John and Willie Family Department of Energy and Mineral Engineering at Penn State.

The issue isn’t insufficient generation, but an unreliable ability to deliver ample power when customer use spikes, particularly where renewable resources and natural gas dominate power production, Lo Prete said. To better support the clean-energy transition, she and colleagues at a Washington, D.C.-based nonprofit recently studied 11 electricity market design proposals under consideration by grid operators. These designs put forward different approaches to guide energy generation and sources, as well as use across every sector of the energy market.

The proposals, yet to be tested in the market, range from a modest variation on current market designs to a complete overhaul. Researchers organized proposals into five categories from least to most dramatic, including concepts for long-term contract auctions and a two-pronged approach combining long- and short-term markets.

“Market structures should allow utility operators to recover both fixed and variable costs so they foster greater system reliability overall,” Lo Prete said.

Findings published in the journal Energy Economics spotlight key questions confronting utility decision makers and can shape more research into adjusting electricity markets. Lo Prete explained that forecasting overall demand—expected to see historic growth of 25% by 2030 and 78% by 2050—will be especially difficult as transportation electrifies and more data centers come online.

Mandatory “forward contracts,” or advance obligations by distributors to purchase specific amounts of electricity from power generators, could help support investments in resources that are instrumental in meeting decarbonization objectives, she said.

Lo Prete noted the February 2021 system failure in Texas that left more than 4.5 million homes without power; rolling outages in California in August 2020; and near-blackouts, also in the Golden State, in September 2022. In each instance, the underlying problem was a lack of accessible energy in the moment of greatest demand, she said.

Such situations have led grid operators to weigh the market approaches reviewed by researchers in their study, Lo Prete said. Reforms on the table would attempt to accommodate ongoing shifts in power generation, whether through longer-term auctioning of future electricity supplies, more centralized resource planning or other mechanisms like so-called “swing contracts.” They seek to ensure the availability of power production capabilities for dispatch in future operating periods.

“When the markets were restructured in the late 1990s, the energy system was very different from the one we have today,” Lo Prete said.

At that point, the system centered on thermal power plants driven by fossil fuels and nuclear energy. Utility markets today aren’t structured to integrate and sustain the renewable sources and large-scale electricity storage that have taken root since then.

Still, maintaining a range of power generation is vital, as older facilities like coal-powered plants contribute less to the power supply but remain important to consistent service, Lo Prete said. Last year, coal accounted for 8% of primary energy consumption nationally, down from 23% in 2000, according to a congressional report.

For their study, Lo Prete and her research partners at Resources for the Future (RFF) examined market proposals to assess energy affordability, efficiency, energy adequacy and other factors. Lo Prete, a faculty associate of the EMS Energy Institute and the Institute of Energy and the Environment and a Wilson Faculty Fellow at Penn State, completed a sabbatical at RFF ahead of the paper’s publication.

Among their conclusions, researchers found the organization of regulatory oversight makes it more difficult to incorporate clean-energy policy into electricity markets. Those “forward contracts” requiring specific electricity purchases could promote energy storage and power systems’ overall ability to fulfill customer needs, they found.

At the same time, the authors said it was tough to make recommendations or endorse one proposal over others, in part because the concepts were in different stages of development. Researchers cited specific concerns over inadequate investment incentives in current energy markets.

The authors also urged cooperation among energy-market researchers, encouraging them to make proposals accessible to broad audiences and facilitate input and feedback from those constituents. Communication will help researchers understand concerns and possible points of confusion, they said.

At Resources for the Future, contributing to the paper were Karen Palmer, senior fellow and director of the Electric Power Program, and associate fellow Molly Robertson.

Microsoft’s Azure cloud platform, its widely used 365 services, Xbox, and Minecraft started suffering outages at roughly noon Eastern time on Wednesday, the result of what Microsoft said was “an inadvertent configuration change.” The incident—which marks the second major cloud provider outage in less than two weeks—highlights the instability of an internet built largely on infrastructure run by a few tech giants.

Microsoft’s problems specifically originated from Azure’s Front Door content delivery network and emerged just hours before Microsoft’s scheduled earnings announcement. The company website, including its investor relations page, was still down on Wednesday afternoon, and the Azure status page where Microsoft provides updates was having intermittent issues as well.

Microsoft described in status updates on Wednesday that it went through a process of sequentially rolling back recent versions of its environment until it could pinpoint the “last known good” configuration. At 3:01 pm ET, the company said it had identified and pushed this stable configuration and that “customers may begin to see initial signs of recovery. We are currently recovering nodes and routing traffic through healthy nodes.”

A Microsoft spokesperson said in a statement, “We are working to address an issue affecting Azure Front Door that is impacting the availability of some services. Customers should continue to check their Service Health Alerts.” The company did not immediately respond to questions from WIRED about the nature of the configuration change that caused the outage.

In addition to occurring on Microsoft’s earnings day, the outage comes nine days after Azure rival Amazon Web Services suffered a massive outage that impacted sites and services around the world. Major cloud providers, often called “hyperscalers,” standardize and often improve baseline security and reliability for their customers, but problems and outages can cause them to become single points of failure for large populations of critical digital services

“Even Azure’s outage status page is down,” says Davi Ottenheimer, a longtime security operations and compliance manager and a vice president at the data infrastructure company Inrupt. “Another configuration change error—we are in the age of integrity breach more so now than ever.”

Azure blocked customers from making configuration changes to their instances while it worked to address the issue. The company said in a status update at 3:22 pm ET that it expects “full mitigation” of the situation by 7:20 pm ET.

“Organizations may think they’re insulated by their choice of cloud provider, but dependencies run deeper,” says Munish Walther-Puri, an adjunct faculty member at IANS Research and the former director of cyber risk for the city of New York. “When key partners rely on other hyperscalers, exposure multiplies. As AI becomes the next layer of critical infrastructure, these outages demonstrate the brittleness of our digital backbone.”