| In this edition, a look at how actionable Trump’s AI action plan could be, and a peek behind Amazon’͏ ͏ ͏ ͏ ͏ ͏ |
|  | Technology |  |
| |
|
 | Reed Albergotti |
|
Hi, and welcome back to Semafor Tech. Before President Donald Trump’s inauguration two months ago, we were hearing about a “Manhattan Project for AI” that was supposed to supercharge the country’s effort to take on China in an existential race for technological superiority. Since then, we’ve had a couple of announcements of private-sector infrastructure projects that were actually underway long before he was elected, along with a head-scratching economic strategy that is ceding global power to China. As top AI companies like Google, OpenAI, and Anthropic submit their proposals for a White House AI Intelligence Action Plan, they all have something in common: They sound somewhat similar to what they were saying before Trump was elected. They advocate for more spending on energy infrastructure, government adoption of AI technology, clearer intellectual property laws, retraining for the workforce, etc. It may be time for the foundation model companies to acknowledge the elephant in the room: a master plan itself. Despite the fact that very smart people from Silicon Valley are aiding the administration — AI Czar David Sacks included — there may be no coherent policy in sight. Given the urgency of the situation, it’s surprising that the US hasn’t already embarked on ambitious projects to ramp up energy production and bolster the country’s scientific leadership. The opposite is happening as the administration makes bewildering cuts to basic research. A Manhattan Project for AI would have been a smart investment requiring serious tax dollars, but the rapidly shrinking economy, and sharp government cuts, makes that less likely. The tech industry is surely hoping that there is not only an AI plan, but action, too. Also: Read below about Amazon’s ambitious effort to boost its position as a leader in AI. ➚ MOVE FAST: Intel. Investors embraced the appointment of former board member Lip-Bu Tan as CEO after clashing with his predecessor, Pat Gelsinger, over the direction of the company. Tan has deep relationships in the semiconductor ecosystem, which will be useful as he works to turn around the company. ➘ BREAK THINGS: Intel. The chipmaker is still struggling to catch up with rivals who were faster in catching the AI wave. That has other companies circling, including TSMC, as the US administration considers ways to bolster Intel, which is seen as a national semiconductor champion.  |
|
NASA’s Goddard Space Flight Center Scientific Visualization StudioStanford University researchers are using physics-informed deep learning to determine the movements of Antarctic ice, they revealed in a study published Thursday in Science. Researchers built a machine-learning model that analyzed satellite imagery and airplane radar of the ice from 2007 to 2018, applying the laws of physics that govern ice movement. They were then able to develop new mathematical models for determining the ice’s viscosity — its resistance to moving or changing shape under stress. Previous studies relied on lab experiments to glean such information, but Antarctic ice is more complicated than what can be replicated in a lab, senior author Ching-Yao Lai told Stanford’s news site. Lai and her team confirmed that the ice shelf — floating glacial ice extending over the ocean — reacts differently to stress applied from various directions, a property known as “anisotropic.” “People thought about this before, but it had never been validated,” said first study author Yongji Wang. “Now, based on this new method and the rigorous mathematical thinking behind it, we know that models predicting the future evolution of Antarctica should be anisotropic.” The Antarctic ice sheet holds enough water to raise global sea levels by 190 feet if it all melted, according to NASA — which would decimate coastal areas and submerge major cities like New York, San Francisco, Hong Kong, and London. That level of devastation isn’t expected anytime soon, but the ice sheet still loses an average 150 billion metric tons of ice each year, NASA reported. |
|
AUSTIN, TX — Amazon’s gamble to take on Nvidia, leading to the e-commerce giant’s biggest investment ever this year, is also a bet for startup Anthropic. By itself, Amazon’s five nanometer Trainium 2 microprocessor is not as powerful as Nvidia’s latest AI chip, coveted by companies like OpenAI and xAI for its ability to train the next generation of powerful AI models. But Amazon hopes its homemade silicon, dreamed up by Annapurna Labs, an Israeli chip startup the company acquired in 2015 for $350 million, will be used to build the most powerful computer in the world — dubbed “Project Rainier.” Amazon’s success or failure is not riding on the raw power of each individual chip, but on a meticulously planned vertical integration in which entire data centers, down to each screw, copper wire and cooling fan, are engineered to squeeze every ounce of compute power from hundreds of thousands of Trainium 2 chips. “We take vertical integration to an extreme,” said Rami Sinno, director of engineering for Annapurna, during a tour of the chipmaking facility. “This concept of power and power efficiency permeates everything we do.” Rami Sinno, director of engineering for Annapurna Labs, standing in front of a rack of Trainium chips. Reed Albergotti/Semafor.If the plan works, it won’t just be a win for Amazon, but also for Anthropic, the AI company behind the Claude AI chatbot. It has become a favorite among professional software developers and “vibe coders,” whose only gripe with the tool is its rate limits that cut users off in order to keep costs under control. Anthropic is Amazon’s most important customer and has agreed to use Rainier to train the next version of Claude, making it more performant and cost effective, providing more coveted “tokens” for Claude’s users. Anthropic, boosted by an $8 billion investment from Amazon on its way to a $60 billion valuation, used Google Tensor Processors and Nvidia GPUs to train the previous versions of Claude models. Two people familiar with the matter told Semafor that the company’s agreement to use Amazon’s custom chips is separate from Amazon’s decision to invest in the company. However Anthropic came to its decision, it is a win for Amazon; luring a leading foundation model company away from Nvidia is not easy. Since 2006, Nvidia has been improving and adding functionality to Cuda, a powerful software program that allows AI researchers and other programmers to run nearly any machine-learning algorithm or AI model on Nvidia GPUs. Because of Cuda’s head start, competing with Nvidia is incredibly difficult. |
|
 The amount Emirati state-owned investment firm MGX invested into Binance, marking the UAE’s bet on “a more crypto-friendly world order,” Semafor’s Kelsey Warner wrote. The news comes shortly after Trump signed an executive order establishing a US crypto reserve, and his family’s representatives reportedly discussed taking a stake in Binance’s US arm. |
|
 In this moment of media fragmentation and the rise of niche communities, are there still ways to reach mass audiences? This week, Ben and Max bring on legendary marketing executive, Frank Cooper, who’s always been at the center of big cultural shifts from his time at Def Jam in the 90s, AOL in the 2000s, and BuzzFeed in the 2010s. Throughout his career, he’s also been seen as the culture translator for big corporations, as the CMO at PepsiCo and the CMO of Visa. They talk about Frank’s unique career, who and what he thinks still moves people in mass — like Post Malone at the Louvre — and what he makes of this particular moment in the culture. He also shares stories from his time working in hip hop, what he’s learned from LL Cool J, and how Snoop Dogg became the world’s most marketable star. |
|
One cloud computing company is trying to cut through the economic noise to become a herald of the artificial intelligence IPO scene despite a soft market and few attempts by competitors. CoreWeave filed an IPO plan last week, and it expects to list on the public market later this month, The Information reported. CoreWeave’s attempt comes during a massive slump in the IPO market. In the last three years, the number of tech companies entering the public market hit decade-lows. And while the overall IPO count crept upwards since 2022, it’s far from the bustling activity that marked years prior. Tumbling US stocks in recent weeks haven’t instilled confidence in the market either. Major stock indexes, including tech ETFs, are largely down on tariff and inflation concerns since Trump took office for the second time.  CoreWeave, which calls itself an “AI hyperscaler,” rents out GPUs, CPUs, and other computing equipment to tech companies so they don’t have to purchase and maintain their own. Its IPO would represent one of the first in recent years from a major company supporting the expansion of AI, making it a bellwether for both the tech industry and how the public market perceives AI in the years ahead. |
|
Nicoco Chan/ReutersNo profit? No problem. Chinese regulators are calling to ease the IPO approval process to make it easier for innovative but unprofitable tech companies to go public, Chinese media site Yicai reported. There’s no outright ban on unprofitable companies listing on the public market in China, but the approval process has become stricter in recent years. Under the proposed changes by the country’s main securities regulator, unprofitable companies with at least 4 billion yuan ($552 million) in market value could access the science- and technology-focused equities market, similar to the US Nasdaq. That listing criteria had effectively been paused due to heightened oversight, with companies not utilizing this pathway since 2023. Such a move could give AI startups greater access to the capital markets at a critical time in the technology race with the US. It could also help revive China’s downturned IPO market, which, like the US, has decreased in the number and value of IPOs in recent years. |
|
 Todd Korol/ReutersThe CEOs of two of the world’s largest energy companies touched on a similar theme at the oil and gas industry’s annual CERAWeek conference: Humanity must not choose between renewables and fossil fuels, Amin Nasser of Saudi Arabia’s Aramco and Sultan Al Jaber of the UAE’s ADNOC said. “We can all feel the winds of history in our industry’s sails again,” Nasser said, as Semafor’s Mohammed Sergie reported. The overall message wasn’t exactly “drill, baby, drill,” but Al Jaber concluded with: “Let’s make energy great again.”
|
|
| |  | | You’re receiving this email because you signed up for newsletters from Semafor. |
|