TheChek News Flow

Data center tweaks could unlock 76 GW of new power capacity in the U.S

Tech companies, data center developers, and power utilities have been panicking over the prospect of runaway demand for electricity in the U.S. in the face of unprecedented growth in AI.

Amidst all the hand wringing, a new paper published this week suggests the situation might not be so dire if data center operators and other heavy electricity users curtail their use ever so slightly.

By limiting power drawn from the grid to 90% of the maximum for a couple hours at a time — for a total of about a day per year — new users could unlock 76 gigawatts of capacity in the United States. That’s more than all data centers use globally, according to Goldman Sachs. To put that number into perspective, it’s about 10% of peak demand in the U.S.

If data centers were to curtail their use more, they could unlock progressively more capacity.

Such programs aren’t exactly new.

For decades, utilities have encouraged big electricity users like shopping malls, universities, and factories to curtail their use when demand peaks, like on hot summer days. Those users might turn down the air conditioning or turn off thirsty machines for a few hours, and in return, the utility gives them a credit on their bill.

Data centers have largely sat on the sidelines, instead opting to maintain uptime and performance levels for their customers. The study argues that data centers could be ideal demand-response participants because they have the potential to be flexible.

There are a few ways that data centers can trim their power use, the study says. One is temporal flexibility, or shifting computing tasks to times of lower demand. AI model training, for example, could easily be rescheduled to accommodate a brief curtailment.

Another is spatial flexibility, where companies shift their computational tasks to other regions that aren’t experiencing high demand. Even with data centers, operators can consolidate loads and shut down a portion of their servers.

And if tasks are mission critical and can’t be delayed or shifted, data center operators can always turn to alternative power sources to make up for any curtailment. Batteries are ideally suited for this since even modestly sized installations can provide several hours of power almost instantaneously.

Some companies have already participated in ad hoc versions of these.

Google has used its carbon-aware computing platform, originally developed to trim emissions, to enable demand response. Enel X has worked with data centers to tap into the batteries in their uninterruptible power supplies (UPS) to stabilize the grid. And PG&E is offering to connect data centers to the grid quicker if operators agree to participate in a demand response program.

These tweaks won’t completely eliminate the need for new sources of power. But they might turn a potentially catastrophic situation — in which half of all new AI servers are underpowered — into one that’s more easily solved.

However, recent releases of ubercheap models like High-Flyer Capital Management’s DeepSeek have raised questions about whether AI developers will need to cut prices to stay competitive. Anthropic has been one of the few AI labs that hasn’t slashed its model prices in the wake of DeepSeek.

Anthropic’s pitch suggests that it sees selling technology to businesses through an application programming interface as a bigger opportunity than competing with chatbot apps such as ChatGPT. OpenAI’s chatbot has been a breakout success with consumers as well as professionals such as programmers, marketers and lawyers, and it was generating more than $333 million per month in revenue from subscriptions as of the end of last year.

Overtaking OpenAI?

In fundraising materials Anthropic sent to investors, it said its leaders expected API revenue to hit $20 billion by 2027, which would be three times higher than the API revenue OpenAI has forecasted for that year. But even in the base case projection, Anthropic’s API revenue would beat OpenAI’s. (The rest of Anthropic’s projected revenue comes from its sales of the Claude chatbot to businesses and professionals.)

Anthropic says its technology could transform office roles such as generating or reviewing legal paperwork and automating software engineering. It cited code repository GitLab and legal search firm LexisNexis as examples of customers. Up-and-coming startups such as Anysphere, which develops the Cursor coding assistant for programmers, are also major buyers of Claude software.

The pitch comes as Anthropic seeks to raise $2 billion from investors at a pre-investment valuation of $58 billion, while OpenAI tries to raise financing at a pre-investment valuation of $260 billion.

Besides the revenue gap, another factor in the companies’ diverging valuations is Anthropic’s agreements with cloud providers that resell its AI—namely, Amazon. While OpenAI gives Microsoft 20% of its revenue through their business partnership, Anthropic previously told some investors it paid a substantially higher percentage to Amazon when companies purchase Anthropic models through Amazon.

Of the $34.5 billion in revenue Anthropic’s management projected for 2027, nearly two-thirds would come from its API business, and a majority of those sales would come from partners like Amazon, which has committed billions of dollars to the company and is developing a data center server cluster for it.

Burning Dollars

Another factor in the valuation gap: OpenAI is likely operating more efficiently than Anthropic.

Anthropic told investors it expects to burn $3 billion this year, substantially less than last year, when it burned $5.6 billion. Last year’s cash burn was nearly $3 billion more than Anthropic had previously projected. That’s likely due to the fact that more than half of the cash burn came from a one-off payment to access the data centers that power its technology, according to one of the people who viewed the pitch.

Anthropic’s management expects the company to stop burning cash in 2027.

OpenAI, in comparison, burned just $340 million in the first half of 2024 while generating many times more revenue than Anthropic did over the same time period, according to The Information’s analysis. OpenAI projections implied cash burn could rise in the coming years.

The costs to develop AI models accounted for a major portion of Anthropic’s expenses last year. The company spent $1.5 billion on servers for training AI models. OpenAI was on track to spend as much as $3 billion on training costs last year, though that figure includes additional expenses like paying for data.

Anthropic’s costs for research and development staff reached $160 million, though that likely excludes stock-based compensation—a substantial noncash expense. OpenAI reported $1.5 billion in stock compensation in the first half of 2024.

Anthropic ended 2024 with 915 employees, 521 of whom were in research and development. The company projects that total head count will more than double to 1,900 by the end of 2025. OpenAI, in comparison, had more than 1,600 employees as of September 2024.

Many electric vehicle and battery startups are in danger of running out of money before they can generate profits. By contrast, electric aviation startups are enjoying fundraising success that resembles the industry’s party days of 2020 and 2021, when investors were throwing money at EV and battery startups.

In the latest example, Archer Aviation on Tuesday said it had raised $300 million from BlackRock, Wellington Management and other institutional investors to fund its development of hybrid electric aircraft for the military. That was on top of $430 million it raised just two months ago for the same project.

Archer didn’t ask for or need the latest money, CEO Adam Goldstein told me. Instead, he said, institutional investors reached out after the December fundraising to express interest in investing. Goldstein went for it, aiming to add manufacturing capacity at the company’s San Jose, Calif., factory (more on that below).

“I think defense tech is hot, and there are a lot of investors that are kind of knocking on our door,” Goldstein said.

Goldstein may be right that Archer struck a chord with its move to develop military aircraft, in addition to the civilian electric air taxis it hopes to commercialize later this year. (BlackRock and Wellington declined to comment.) Officials in Donald Trump’s administration have said they want the Pentagon to look at lower-cost startups for military hardware, to compete with big defense contractors such as Lockheed Martin and Northrop Grumman. After his confirmation last month, Defense Secretary Pete Hegseth said in a written statement that among his goals was “rapidly fielding emerging technologies.”

Private investors have also been exceptionally open toward air taxi startups, investing more than $2 billion in three of the most prominent companies in just the last four months: In addition to Archer’s rounds, Santa Cruz, Calif.–based Joby Aviation has raised more than $1 billion in three deals since October, and Vermont-based Beta Technologies raised $318 million in a Series C round in December.

All three startups have research and development deals with the U.S. Air Force to develop electric helicopters. But in December, a senior Air Force officer told Aviation Week of a strategic shift—the military was now also seeking hybrid aircraft because all-electric helicopters would not have sufficient flying range.

The Air Force has not yet offered any contracts for hybrid aircraft, but Goldstein said Archer and its hybrid aircraft development partner—Anduril Industries—believe there will be such a program.

Goldstein said Anduril has software that designs weapons or aircraft to meet specified performance requirements. Archer plugged in what it thought the Air Force would need in a hybrid aircraft—“a thing that can go this fast, this far, carry this much stuff, can hover for this much time, those kinds of things,” Goldstein said.

The program helped Archer’s leaders grasp what they might need to build. “Then it was like, ‘OK, we have conviction that we can go and build that,’” Goldstein said. “The thesis is that if we could build that, we think we are in a really good position to go win a contract.”

With the new money raise, Goldstein wants to invest in production lines to make composite materials for the hybrid aircraft’s body in Archer’s factory. He is also considering investing in a company making next-generation silicon-based batteries that can deliver a significant, short burst of power for takeoff and landing.

Archer declined to say whether the investment would be in its current battery supplier—Molicel, a Taiwanese company that many industry hands regard as one of the world’s best manufacturers of high-power batteries. Molicel did not immediately respond to an email.

But Goldstein said Archer is testing the battery maker’s cells, with the objectives of building a production line in Archer’s factory, as well as stockpiling an inventory of cells as “a kind of natural hedge” in case the U.S. gets into a conflict with China, which now supplies the vast majority of the world’s batteries. “In the end, the supply chain all leads back to one country in batteries,” he said. “It looks very promising, something we’d like to invest in, that we’d like to make it work, and that’s what we’re going out and doing.”

Ultimately, Archer and its investors are betting on the new administration. “This is exactly in line with what the administration talks about, which is, ‘How do you go build lower-cost unmanned systems to help complement or replace some of the existing systems and build, like, the future of the programs?’” Goldstein said. “I’m pretty bullish on what’s possible to go build.”

Simon Kohl aims to capitalise on his experience with Nobel-recognised AlphaFold protein prediction system

A scientist who helped build DeepMind’s protein prediction programme, which won the company’s co-founder a Nobel Prize, has raised $50mn to launch a start-up to discover new drugs.

Simon Kohl has founded Latent Labs to capitalise on his experience working on the AI-powered AlphaFold2 system, which solved an important problem for life sciences researchers by allowing the prediction of the 3D structure of proteins based on their chemical composition. 

Latent Labs, which has offices in London and San Francisco, will work with the pharmaceutical industry to design synthetic proteins, which could be used in drugs such as antibody treatments.

Kohl said generative AI, which creates new content, had the potential to change how drugmakers operate by making biology “programmable”.

“In a perfect world, the dream of purely computational drug discovery comes to life,” he described. “You tell me, this is the sort of disease you’re going after. Maybe there’s a target protein that you have in mind. And at a push of a button, we can generate candidates that meet all of the criteria you care about, which de-risks the steps that come after, shaves time off the process, and at the end of the day, will yield better drugs, faster.”

Many other start-ups are also trying to create drugs using artificial intelligence and most plan to partner with pharmaceutical companies on any drugs they do discover.

Latent Labs is pursuing a different model: it wants to be a service provider to pharmaceutical companies, providing a generative AI platform that they can use. Its aim is to work with drugmakers, many of which have built their own AI teams, to use the technology to cut the long and expensive process of discovering and developing drugs.

DeepMind has also spun out its own drug discovery company, Isomorphic Labs. Kohl said Isomorphic was following the more common business model of partnering with pharma companies, and was taking a broader approach beyond just proteins, including looking for the small molecules that form the basis of many drugs.  

The fundraising was co-led by Radical Ventures, which specialises in AI software companies, and European healthcare venture firm Sofinnova Partners. Other investors include Jeff Dean, Google’s chief scientist, and Aidan Gomez, one of the authors of the research paper that led to the creation of ChatGPT. 

Aaron Rosenberg, a partner at Radical Ventures who also used to work at DeepMind, said Latent Labs had hired an “all star” team from companies including Microsoft, Google, and Altos Labs, a biotech focused on longevity.

He said that the company’s model, which was already “amazingly capable” would improve dramatically. “This is a whole new world of computational biology.”

Apptronik, which makes humanoid robots, raises $350M as category heats up

Apptronik, a University of Texas spin-out that was quietly building humanoid robots before it became quite so fashionable, on Thursday announced a $350 million Series A round of financing. B Capital and Capital Factory co-led the round, which also featured participation from Google, whose DeepMind division is partnering with Apptronik to deliver embodied AI for bipedal robots.

“What 2025 is about for Apptronik and the humanoid industry is really demonstrating useful work in these applications with these initial early adopters and customers,” CEO Jeff Cardenas tells TechCrunch. “And then true commercialization and scaling happening in 2026 and beyond. That’s what this raise is designed to do.”

The Austin-based startup had raised a relatively modest $28 million combined prior to this round. Cardenas says the previous goal was to generate more revenue than money raised – a goal he says the eight-year-old startup achieved during that time. That revenue came by way of pilot deals — including with Mercedes and GXO Logistics — and by selling robots outright. For now, however, the goal of generating more revenue than fundraising is going to have to go on-hold for a while.

Apptronik’s humanoid work dates all the way back to 2013, three years prior to its founding. It was then that members of the University of Austin at Texas’s Human Centered Robotics Lab competed in the NASA-DARPA Robotics Challenge, an effort that centered around a humanoid robot called Valkyrie. Since then, the space agency has maintained a partnership with Apptronik as the company has readied its own generation of humanoids, including its current humanoid, Apollo.

Cardenas points to that decade-plus of humanoid experience as a primary differentiator between Apptronik and competitors like Figure, 1X, and Tesla. Boston Dynamics and Agility Robotics have long histories, as well, but Apptronik is a seasoned veteran in the category compared to much of the competition.

That history may explain why Google’s DeepMind AI team has been working with Apptronik to build robot behavioral models. Their “strategic partnership” is similar in nature to others in the industry. Last week, Boston Dynamics announced a tie-up with The Robotics & AI institute. That followed a similar deal between the Spot-maker and Toyota Research Institute that’s aimed at improving how robots learn.

All are indicators of a much bigger trend, including OpenAI’s multiple deals in the space. The ChatGPT-maker has invested in both 1X and Figure. Last August, Figure announced that it would further leverage OpenAI models to develop natural speech conversations for its own 02 robot, though last week, the outfit announced new plans; going forward, it will move all of its AI development in-house.

“We found that to solve embodied AI at scale in the real world, you have to vertically integrate robot AI,” Figure CEO Brett Adcock told TechCrunch last week. “We can’t outsource AI for the same reason we can’t outsource our hardware.”

Apptronik may ultimately make the same choice, but for now, a Google DeepMind partnership makes far more sense for the startup than the additional funding required to build bespoke humanoid AI models in-house. “We believe that right now, Google is at the top of the game, and building some of the best models in the world,” Cardenas says.

Scaling and production are the magic words for Apptronik’s A round. Apptronik’s current headcount is just north of 170 people, and it’s planning a 50% increase over the next year.

Still, Cardenas is pragmatic about timelines in a category that can be fast to overpromise and underdeliver. Cardenas tells TechCrunch that Apptronik has yet to move beyond the pilot stage with any of its partnerships. For all of the excitement around humanoids, it’s still key for companies to take a measured approach to the category, addressing things like safety concerns and reliability prior to scaling the technology in a meaningful way.

In the meantime, the company has a handful of ongoing pilots, including Mercedes, which makes for a natural choice. Automotive manufacturing has been the leading use case for these kinds of pilots, requiring tote moving and other manual tasks on the factory floor. Boston Dynamics has similarly been working with its parent company, Hyundai; Figure has deployed robots with BMW; and Tesla’s Optimus will eventually get to work on the company’s own EVs.

Bringing it all back home

Like many of its rivals, Apptronik is also looking for ways to put Apollo to work outside of factories and warehouses. The day could arrive when these robots come home to help with groceries, cook, fold laundry, and other tasks buyers might want to offload onto an automaton. Cardenas is even more excited about age tech as an important avenue for advanced robotics. As the population ages and more older adults prefer to live independently, humanoids could eventually help.

“The holy grail for me is [age-tech],” says Cardenas. “As humans,” he says he asks himself, “where could we apply this technology that improves the human condition?”

The holy grail will have to wait, however.

For now, Apptronik, like most humanoid manufacturers, is focused on industry. Factories and warehouses are a good first step, as corporations have the money and other resources required for pilots. Scaling manufacturing for these projects will continue to drive the price point down, but as it stands, the systems are far too expensive for the home – or even care facilities – to be a practical route. Apollo’s target price is below $50,000, says Cardenas. But Apptronik isn’t there yet.

“We’re in the window where the economics now make sense,” says Cardenas. “And we know how to get to much more affordable systems.”

Gapping down
In reaction to earnings/guidance:

Other news:

Gapping up
In reaction to earnings/guidance:

Other news: