Axios AI+

January 29, 2025
Meta and Microsoft report earnings today, with both companies likely to face heightened scrutiny over their planned billions in AI research and infrastructure spending. Today's AI+ is 997 words, a 4-minute read.
Situational awareness: Microsoft and OpenAI are investigating whether makers of China's DeepSeek model used data output from OpenAI's models in violation of OpenAI terms of service, Bloomberg reported.
1 big thing: Trump's funding freeze chills AI research
The tech industry has pushed hard to partner with the federal government on everything from AI research to semiconductor manufacturing, but President Trump's move to "temporarily" pause all federal grant funding threatens to slow or derail that work.
Why it matters: While many big tech companies are flush with cash, the industry relies on the government for a range of research funding, institutional coordination and other support.
State of play: The president's move to freeze a wide array of federal grants — announced Monday but then put on hold by a federal judge late Tuesday — created confusion about both its breadth and its constitutionality.
- The uncertainty hit the tech industry in a variety of ways, throwing a number of projects into limbo.
Tech projects that are funded through congressionally approved grants include efforts to boost U.S. chip manufacturing and to "rip and replace" telecommunications gear from Huawei and other Chinese firms now used by rural cellular networks.
- "That funding is now paused, leaving our rural small businesses in the dust and our telecommunications networks at risk," the office of Sen. John Hickenlooper (D-Colo.) said in a statement Monday, adding that Trump's move "appeases China by allowing them to continue having a hold in our rural communication networks."
- A federal judge on Tuesday temporarily blocked the administration's funding freeze with a stay until Monday. But uncertainty at agencies and among grantees remains widespread.
Zoom in: The National Science Foundation and a host of tech companies big and small are investing in AI research.
- One big driver for the National AI Research Resource, an NSF project, was a sense of urgency to outcompete China.
- On Monday, NSF told grantees they must comply with Trump's various executive orders.
- "In particular, this may include, but is not limited to conferences, trainings, workshops, considerations for staffing and participant selection, and any other grant activity that uses or promotes the use of diversity, equity, inclusion and accessibility (DEIA) principles and frameworks or violates federal anti-discrimination laws," NSF said.
Between the lines: The Trump administration is targeting policies related to DEI, nongovernmental organizations, "the green new deal" and more.
- That could cover a host of NAIRR-backed projects that have to do with environmental sustainability or bias.
Among those is a project meant to ensure that large language models and image-generating models target digitally underrepresented groups and reduce bias against senior citizens.
- The project recognizes that AI models are more attuned to the digital behaviors of younger generations and could leave senior citizens feeling underserved.
Another project is meant to study AI energy efficiency and performance to reduce energy costs and drive policy decisions around resource management.
- Project leads emphasize the importance of environmental sustainability, noting that OpenAI's GPT-3, for example, consumed over 500 tons of carbon during its training process alone.
- But University of Pennsylvania professor Benjamin Lee, one of the project leads, said the project's focus on the environment is secondary and its work might be safe, since AI energy demands continue to be relevant to the new administration.
- "My perspective is I'm waiting and seeing to see what specifics will come down the pipeline," Lee said.
Yes, but: Another researcher on the same project — Carnegie Mellon University assistant professor Emma Strubell — described feeling "extremely concerned, particularly for our work on environmental sustainability."
- Strubell said of the administration's moves: "Likely deliberate ambiguity to confuse, discourage, disorganize."
2. DeepSeek's aftershocks continue
China's DeepSeek, which topped the IOS App Store on Sunday and helped wipe out more than $600 billion in market capitalization for Nvidia Monday, continues to upend the tech landscape.
Why it matters: The advent of DeepSeek — which nearly matches the performance of advanced U.S. models at lower cost — hit all the flashpoints of AI, from geopolitics to market bubble fears and climate change to the open source debate.
Here's the latest: This morning, misinformation tracker NewsGuard released findings that DeepSeek's chatbot "failed to provide accurate information about news and information topics 83 percent of the time."
- DeepSeek provided "non-answers" 53% of the time and said that its knowledge cutoff was October 2023. NewsGuard also says the bot relayed the Chinese government's position without being asked to do so.
Yesterday, DeepSeek was the topic of the first question in the first press briefing of Trump's new administration (asked by Axios' Mike Allen).
- Trump's press secretary said the White House is "looking into" the potential security implications of the technology.
DeepSeek further roiled the AI energy use debate this week, as some celebrated the Chinese model's lower electricity use while others claimed that its popularity will only increase energy demands overall.
Before DeepSeek, the best-known open source AI models came from Meta — although open source experts have long debated what level of transparency an AI maker needs to provide to qualify as open source, and whether Meta meets it.
- Over the weekend, Yann LeCun, Meta's chief AI scientist, posted on Threads that DeepSeek owed its performance in part to Meta open source projects like PyTorch and Llama.
- But a day later, The Information reported that Meta's engineers were scrambling in "war rooms" to respond to DeepSeek.
Go deeper: Ina also talked about DeepSeek and its implications on Tuesday's episode of NPR's "Here & Now."
3. Training data
- OpenAI launched a version of ChatGPT designed for government agencies. (CNBC)
- Meanwhile, OpenAI is seeing a surge of revenue from its $200-per-month ChatGPT Pro product, which is now outpacing the enterprise version sold to large businesses. (The Information)
- The Navy banned its service members from using DeepSeek due to "security and ethical concerns." (CNBC)
4. + This
AI-fueled writing suggestions are popping up everywhere, from Gmail to Facebook to Apple's operating system. While some appreciate the help, many see it as an unwanted intrusion into their tone and style, a critique especially well put in this blog post by Jenny Lawson — one of my favorite authors.
- Also, you should totally check out her books, starting with "Let's Pretend This Never Happened" — and I recommend the audiobook with her narration.
Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter and Matt Piper for copy editing it.
Sign up for Axios AI+





