By Benjamin Guggenheim

Not a subscriber? Sign up here to get this newsletter in your inbox.

• AI data centers experience massive fluctuations in power usage that degrade traditional lithium-ion batteries. Flow batteries offer a promising alternative.
• The House passes a suite of children’s online safety bills Monday evening, but big differences remain between what the House and Senate want. A potential federal framework on AI hangs in the balance.
• A new analysis by The Post finds political bias in a range of AI chatbots.

This is Benjamin Guggenheim. Welcome to WP Intelligence’s AI & Tech Brief, where we examine the transformative technology of artificial intelligence at the intersection of innovation, policy and power. Get in touch at Benjamin.Guggenheim@washpost.com or on Signal at ben_guggenheim.88.

The Lead Brief

AI data centers are creating demand for a new kind of battery that can handle huge fluctuations in power consumption. Enter flow batteries.

As WP Intelligence Lead Energy Analyst Kathryn Clay outlines in a new report, AI workloads present a unique challenge to our century-old power grid. Unlike large factories, power demand by data centers can rise or fall by hundreds of megawatts every second as thousands of graphics processing units (GPUs) enter different phases to run computations, communicate and store model training progress.

Spurred by the expansion of the electric vehicle market, lithium-ion batteries have dominated other battery technologies and are regarded as the lowest-cost storage option.

But lithium batteries also degrade quickly with repeated charge and discharge cycles — making them poorly suited for data center use. As a stopgap measure, software engineers have responded by generating “dummy loads” to smooth out energy loads when power consumption is low.

However, this extraordinary tactic wastes an enormous amount of energy. And, even with the dummy load tactics, the grid may be strained to the point of triggering a blackout event.

A potential solution: Unlike traditional batteries, flow batteries store energy in liquid rather than solid form. Flow batteries aren’t as energy-dense as lithium batteries — meaning that they don’t pack as much energy per weight or volume — and are currently more expensive.

However, industry proponents say that flow batteries can handle a nearly unlimited amount of loads and last 25 to 30 years — making them a promising contender to replace lithium batteries for use in AI data centers, where energy density doesn’t matter as much.

For instance, Quino Energy, a Harvard University spin-off led by CEO Eugene Beh, recently secured $12.6 million from the Department of Energy to commercialize a specific kind of organic flow battery technology. And the Chinese multinational conglomerate Tencent announced a grant last week for Quino to build a flow battery to support a renewably powered microgrid in the Maldives.

“This represents the first commercial deployment of organic flow battery technology,” said Beh. “Flow batteries will play a defining role throughout the AI boom.”

A flow battery in some ways resembles a small chemical plant, as shown in this diagram. Unlike many other battery types, in a flow battery, power and energy can be engineered separately. The amount of energy they can store is determined by the size of the liquid tanks, and the amount of power they deliver is determined by the size of the processing stack.

The bankability question: Kathryn points out that the reason lithium batteries have come to dominate has less to do with technological superiority and much more to do with economics and timing. Manufacturers invested hundreds of billions in lithium supply chains and battery factories during the EV boom, making them bankable to operators looking for predictable returns.

However, manufacturing costs for flow batteries are still high. Kathryn recommends that data center developers explore flow batteries as a potential grid-support asset for large facilities, with an eye toward life-cycle economics rather than up-front battery costs alone.

The Legislative Chessboard

The House passed a suite of bills on Monday evening aimed at addressing children’s online safety. The Kids Act was put on the chamber floor under a fast-track procedure requiring a two-thirds majority vote.

The package passed 267-117.

The legislation’s passage is a big win for House GOP leadership, including Majority Leader Steve Scalise (R-Louisiana) and House Energy and Commerce Chair Brett Guthrie (R-Kentucky), who were eager to put a flag down on the lower chamber’s stance on the subject — as Sen. Marsha Blackburn (R-Tennessee) negotiates a package with the White House that would include a limited preemption of state AI laws and a different set of children’s safety bills.

While the Kids Act empowers parents to monitor their children’s use of apps and creates guardrails around chatbot use, there’s still a big gap between what Senate lawmakers are looking at and what the House just passed.

The differences: Chief among the differences is a “duty of care” provision in the Senate’s Kids Online Safety Act, which was left out of the House bill, and requires tech companies to proactively prevent and mitigate harms to young users. Blackburn has said any package that does not include duty of care is a nonstarter.

There are also big disagreements between Democrats on language in the Kids Act that would preempt states’ laws on children’s safety. The top Democrat on the Energy and Commerce committee, Frank Pallone Jr. (D-New Jersey), said in a statement that the language explicitly establishes a floor for guardrails and allows states to create stronger online safety laws.

But the top Democrat on the Senate Commerce committee, Maria Cantwell (D-Washington), said in a press call with reporters Friday that the language could make it harder for parents to bring cases against Big Tech companies.

The package includes “preemption language that tilts the scale and makes it harder for parents to bring cases that hold social media companies accountable for dangerous design features,” said Cantwell, referencing a landmark case this spring in which a jury in Los Angeles found YouTube and Meta negligent for causing mental health harms.

“Under the language in this bill, tech companies would have to argue the case was preempted, and a judge could have stopped that case before it was even brought,” Cantwell said.

Bias in chatbots

The Post tested a range of chatbots for political bias and found that most have clear political leanings on hot-button issues, Kevin Schaul writes in a new analysis.

Using political questions designed by researchers, The Post queried a range of chatbots and found that ChatGPT answered every question exclusively with left-leaning arguments. Google’s Gemini mostly presented both sides of the political arguments, while Elon Musk’s Grok cited the most conservative views (though still presented left-leaning arguments 40 percent of the time).

As Kevin writes, chatbots can pick up political biases in a variety of ways. Most are trained on text from all over the internet, but bias could also come from workers hired by AI companies to rate chatbot responses or from the constitutions that guide chatbot responses.

Ceren Budak, a professor at the University of Michigan, says that the data shaping chatbot views tends to reflect the views of wealthy and industrialized people in democratic societies.

Take a look: You can view a comparison of chatbot responses at the bottom of the report on political questions regarding defunding the police, free speech, affirmative action and the death penalty.

OpenAI spokesperson Liz Bourgeois told The Post that ChatGPT was designed “to be objective by default and help people explore ideas from different perspectives” — and that the company works to “measure and reduce political bias.” Bourgeois says the company was unable to replicate the Post’s findings. (The Washington Post has a content partnership with OpenAI.)

SpaceX, which offers the Grok chatbot, did not respond to a request for comment from The Post.

This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently from The Washington Post newsroom. Learn more about WP Intelligence.