Deciphering University of Chicago’s Ill-Timed, Inscrutable Anthropic Partnership
Amid layoffs and a $140 million budget deficit, the school won't yet say how much it's spending on Claude Enterprise.
Microsoft’s AI CEO, Mustafa Suleyman, says that Anthropic is “extremely expensive, and I think many people are urgently looking for alternatives.” In fairness, Suleyman is not an unbiased observer: he wants his own company’s AI models to rival those of Anthropic, Google, and OpenAI, the latter of which recently renegotiated its longstanding partnership with Microsoft. But Suleyman does have a point: Anthropic’s tools are proving to be unsustainably pricey.
The reason is that Anthropic’s Claude Enterprise plan—which is designed for businesses that want to integrate Claude Chat, Claude Code, and Claude Cowork—is now usage-based, rather than a flat-rate. Businesses loved the old subscription model, where Anthropic charged monthly for each licensed employee, with a set amount of AI usage per person (a rate limit). Businesses could pay more for a “premium” tier, which allowed for an even higher rate limit. But that system wasn’t sustainable—Anthropic was subsidizing too much of the cost per user, especially after Claude Code really took off. Recently, Anthropic started billing Enterprise clients based on AI usage, period. No more subsidies; it’s on each organization to establish and monitor their own usage caps.
Businesses have been caught off-guard by the billing change, as I wrote about last week. The chatter from CEOs and CFOs has been “oh shit, this isn’t good,” and “please, please, please stop tokenmaxxing, I beg of you.” (Tokens are units of text/data processed by AI models; as Anthropic explains, on top of a fee for each Enterprise user, “Every token your team consumes is billed separately at standard API rates.”)
So I was taken aback when I saw the University of Chicago’s June 2 announcement, penned by President Paul Alivisatos, that the school “is partnering with Anthropic to provide Claude Enterprise for all academics and staff starting in July, and for all students before the fall term.” UChicago has approximately 7,500 undergrads, 9,000 graduate students, 600-something postdocs, plus 20,000 staff and faculty. Alivisatos noted that students and staff are not required to use Claude Enterprise, and it’s safe to assume many people will choose not to indulge in these AI tools. But what happens if even a few thousand students and staffers become semi-prolific Claude users?
The timing of the announcement is even more bizarre when you factor in UChicago’s budget deficit, which has decreased but is still reportedly a gaudy $140 million. “Early deficit reductions came heavily from restricting spending growth, including early retirement incentives for faculty, hundreds of layoffs and a pause on doctoral admissions for many graduate programs,” according to the Hyde Park Herald.
How on earth, then, does a Claude Enterprise deal pencil out? I sent a series of questions to UChicago, and presumably was not the only reporter to inquire about what appears, at first glance, to be a fiscally ruinous decision. The university has since published an FAQ page about its Claude Enterprise integration—I’d describe it as moderately helpful in a very roundabout way. From the FAQ:
Are there usage limits, or additional costs associated with usage?
The University’s Claude account will provide generous usage for everyday work. Standard seats offer approximately twice the usage limit of a personal Claude Pro plan. Premium seats, which are comparable to a Claude Max plan, offer more usage limits and are intended for heavier users working in Claude Code. Limits will reset on a rolling basis, and actual capacity varies with the length and complexity of work.
(The “premium seats” aren’t free, by the way; the FAQ says that “upgraded premium access will be available at a reduced cost.”)
Best as I can tell, UChicago is not subject to the unlimited, token-based billing system that Anthropic is now requiring of other organizations. It appears UChicago has signed up for a capped, hybrid setup—something akin to how Anthropic sold its Claude Enterprise program to businesses until recently. One can only assume that Anthropic is subsidizing a chunk of the costs for UChicago. I don’t know for certain, because a UChicago spokesperson responded to my questions by referring me back to the above FAQ.
I additionally asked about the financial parameters of this arrangement, and pointed out that other universities have disclosed details about their AI partnerships. California State University, for instance, partnered with OpenAI for $16.9 million, and recently announced an extension for $13 million per year. That partnership, while also controversial, reportedly included 500,000 licenses (the CSU system is much larger than the total UChicago population). The UChicago spokesperson wrote back: “On cost—we will provide further details to the university community soon.”
That is not a very reassuring answer, especially for UChicago faculty warily eyeing the aftershocks of early retirement packages and layoffs. The best-case scenario, I suppose, is that this deal is “discounted” and Anthropic is metering Claude Enterprise usage, which eliminates the possibility of an army of undergrads “vibecoding” their eyeballs off until UChicago’s budget deficit quadruples in a month’s time. But even if that’s the case, it doesn’t mean UChicago is making a prudent financial decision. What are the terms? How much is UChicago paying Anthropic for an essentially subscription-based version of Claude Enterprise? And where is the money coming from? There have been passing mentions in news reports about a previous $50 million donation to UChicago that was earmarked for AI research, but no evidence yet that the donation will help fund Claude Enterprise.
As for Anthropic: this is bad business. It’s not as if their new strategy of token-based billing for Claude Enterprise is popular. Quite the opposite, in fact. Anthropic made the switch because it had no choice. No matter how much money Anthropic is getting from UChicago, it won’t be nearly enough.
But that’s not the point, is my guess. Anthropic is striking deals with universities for the same reason that Google cornered the market on K-12 schools and passed out its products like candy: the actual goal is to acquire lifelong users. The more young adults you can get to embrace Claude, the better. The difference, though, is that Google was already incredibly profitable when it gambled on the K-12 “market.” Anthropic certainly talks a big talk, but its medium-term future—let alone its ability to cultivate lifelong users—is far from assured. It’s making a risky bet here. The University of Chicago is too. So it goes in the age of AI.
Here’s what else we’re reading this week:
More Anthropic news, from the Financial Times: “Anthropic is helping the US National Security Agency deploy its powerful Mythos AI model for offensive cyber operations, embedding engineers inside the agency.” Interesting that “offensive cyber operations” aren’t part of Anthropic CEO Dario Amodei’s arbitrary red lines.
Seven in 10 Americans oppose data centers being built near their homes, according to a new poll from Heatmap Pro. In Monterey Park, a city that’s part of Los Angeles County, residents voted overwhelmingly in favor of a ballot measure that forbids the construction of data centers within city limits.
Still lots of votes to tally, but Silicon Valley’s favorite mayor, Matt Mahan, insta-conceded the California governor’s race on Tuesday night. His campaign resembled a Blue Origin rocket launch. As of Thursday afternoon, he’s got 4.1% of the vote, which puts him in sixth place. Accounting for endless Big Tech funding and a lackluster slate of competitors, Mahan’s performance is especially embarrassing. In Santa Clara County, which includes San Jose (where Mahan is the mayor), he’s currently pulling in just 10% of the vote.
Speaking of Santa Clara: Rep. Ro Khanna cruised to a primary win, despite some uninspired saber-rattling from Y Combinator President Garry Tan, who backed entrepreneur Ethan Agarwal. As of Thursday afternoon, Khanna has 58.9% of the vote. Agarwal has 6.2%.
President Trump issued a long-rumored AI executive order on Tuesday. It is watered down and functionally useless. In short: it asks large AI companies to voluntarily submit their new products for government review, a process intended to “gauge what threats the products may pose to sensitive financial, national security and other computer systems,” Politico reported. Unsurprisingly, tech executives are thrilled with Trump’s hands-off approach. Congrats to them, and to David Sacks.
OpenAI was compelled to clarify that it does not currently have an employee-funded Political Action Committee, and as an organization, does not donate to political candidates. The company’s president, Greg Brockman, has provided truckloads of cash to Leading the Future, a super PAC that advocates against AI regulation.
Kevin O’Leary, aka Mr. Wonderful, aka Man Who Coasted Off His Marty Supreme Cameo For Longer Than He Deserved, has graciously offered to halve the size of his controversial data center project in Utah. His new plan would require roughly 20,000 acres of land, which is larger than the square mileage of Manhattan.