Good morning. OpenAI dropped GPT-4.1 on Monday — which doesn’t seem like anything worth writing home about — and at the same time said it plans to get rid of GPT-4.5, since 4.1 is cheaper.

Why 4.5 came before 4.1, I do not know.

Here’s what I do know: Nvidia is trying hard to duck around the tariffs, and NATO is adopting generative AI. Let’s get into it.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• ⚡️ AI for Good: Lightning-induced wildfires

• 🚨 NATO acquires Palantir’s AI-for-war platform

• 💰 Nvidia to produce $500 billion worth of supercomputers in the U.S. … for the first time

AI for Good: Lightning-induced wildfires

One offshoot of our changing climate — higher temperatures, less moisture — is that wildfires have become both more frequent and more intense. 

Add in the harsher weather that’s been on the rise lately, and you have a dangerous wildfire situation, one where there’s plenty of available fuel amid a surge in lightning strikes. This has led to an increase in lightning-induced wildfires. 

Researchers expect this type of wildfire to continue to increase in prevalence over time. 

What happened: A team of Israeli researchers recently assembled an (explainable) machine learning model specifically designed to predict the risk of lightning-caused wildfires. 

• The model was trained on seven years of global satellite data, from 2014 to 2020. The model was trained not just to predict the odds of a lightning strike, but to contextualize those odds against weather conditions, vegetation and local topography to predict whether a given lightning strike would cause a fire. 

• In tests on wildfire data from 2021, the model achieved a better than 90% accuracy rate, a level of precision that, according to the Bar Ilan University, “has never been seen before.”

Why it matters: Beyond validating a specifically-tuned predictive model, the researchers found that models trained to predict human-caused wildfires perform with a radically lower accuracy rate when tasked with lightning-induced predictions. The researchers further anticipate that, even over short durations, climate change is steadily increasing the risk of lightning-induced wildfires. 

We need more predictive models for each type of wildfire, according to the team, so that firefighters can respond before things get out of control. 

Big ideas. Bold strategies. AI-powered growth.

Experience it all live at ZoomInfo’s GTM 25 on May 13th at the Guggenheim Museum in New York.

This exclusive experience brings together go-to-market leaders to explore innovative GTM strategies with industry experts, including ZoomInfo Founder & CEO Henry Schuck and McKinsey & Co. Partner Todd Horst.

Connect with peers, dive deep in interactive breakout sessions, and be the first to see ZoomInfo’s groundbreaking platform innovations.

Seats are limited — reserve your spot today!

NATO acquires Palantir’s AI-for-war platform

NATO, the Europe-North American military alliance, said Monday that it has acquired Palantir’s Maven Smart System (MSS) for warfighting. 

The details: The platform purportedly brings together a wide range of data sources and a number of AI-related applications, including large language models (LLMs), generative AI and machine learning. It will be used, according to a statement, to enhance “intelligence fusion and targeting, battlespace awareness and planning, and accelerated decision-making.”

• NATO said in a statement that the acquisition was “one of the most expeditious” in NATO’s history, spanning just six months from early outlines to final acquisition. 

• It will enter into use within the next 30 days, and will be continuously leveraged as a means of adopting additional, novel technologies, “including new, emerging AI models, and modeling and simulation.” 

The terms of NATO’s contract with Palantir are unclear. 

NATO’s statement makes no mention of issues of human oversight, governance procedures, accountability protocols, or, more generally, methods of guaranteeing transparency, explainability and reliability.  

A 2022 West Point analysis of AI technologies, on the other hand, highlighted the many security vulnerabilities associated with the tech, advocating for a thoughtful, careful approach to integration, an approach that seems absent from this latest round of hype. 

The landscape: In May of last year, the U.S. Department of Defense awarded Palantir a $480 million contract for a prototype Maven Smart System. In September of last year, the DevCom Army Research Lab awarded Palantir an additional $100 million to expand the Maven system across all the branches of the U.S. military. 

Palantir began work on Maven in 2017. It was first deployed in late 2023, shortly following the onset of the Israel-Hamas war. 

Dr. Elke Schwarz, a professor of political theory at the Queen Mary University of London, told me in March that it is unclear “why military leaders, who rely on the highest degree of accuracy for their decision making, would come to rely on a technology which is, at this stage of the state of AI, likely to be quite flawed and unreliable.”

Shares of Palantir surged on the news, spiking as much as 5%. 

• A new model: In a livestream Monday, OpenAI unveiled GPT-4.1, its follow-up to GPT-4o, the multimodal model that just replaced GPT-4 to undergird ChatGPT. (These names are terrible). The model is available exclusively through OpenAI’s API. GPT-4.1 has a larger context window and, according to OpenAI, is better than GPT-4o in “just about every dimension.” But it has not been well received. Since the model achieves high performance at a reduced cost, OpenAI said it plans to deprecate GPT-4.5 in the API, with 4.1 coming to replace it.

• Apple’s catch-up plan: Bloomberg’s Mark Gurman reported Monday that Apple plans to begin analyzing on-device user data, comparing it to the synthetic data it uses to train its AI models. This comparative approach allows Apple to better fine-tune its training approaches without training models on user data.

• Maybe just believing in AGI makes AGI exist (Arg Min).

• Former Google CEO tells Congress that 99% of all electricity must be used to power superintelligent AI (Futurism).

• I tested the AI that calls your elderly parents if you can’t be bothered (404 Media).

• DeepMind slows down research releases to keep competitive edge in AI race (FT).

• Unemployment fears hit worst levels since Covid as tariffs fuel inflation outlook, Fed survey shows (CNBC).

Nvidia to produce $500 billion worth of supercomputers in the U.S. … for the first time 

Nvidia said Monday that it plans to begin manufacturing AI-enabling supercomputers in the U.S. for the first time. 

The details: The semiconductor giant has already commissioned more than a million square feet of manufacturing space across Arizona and Texas to accomplish this goal. TSMC’s new plant in Phoenix, Arizona has already begun producing Nvidia’s next-generation Blackwell chips; at the same time, Nvidia is building plants in Houston and Dallas in partnership with Foxconn and Wistron, respectively. 

• The company expects both plants to ramp into mass production within the next 12 to 15 months. 

• Through partnerships with TSMC, Foxconn, Wistron, Amkor and SPIL, Nvidia said it intends to produce some $500 billion worth of AI supercomputers in the U.S. over the next four years. 

This push into U.S. manufacturing, according to Nvidia, will create hundreds of thousands of jobs. 

“The engines of the world’s AI infrastructure are being built in the United States for the first time,” CEO Jensen Huang said in a statement. “Adding American manufacturing helps us better meet the incredible and growing demand for AI chips and supercomputers, strengthens our supply chain and boosts our resiliency.”

The move mirrors a recent pledge by Apple to invest $500 billion in U.S. manufacturing over the next four years. 

Shares of Nvidia, down 17% for the year, were flat Monday. 

The landscape: The timing of the announcement is no coincidence. Earlier this month, President Donald Trump imposed sweeping tariffs across the globe, including a 32% tariff against goods from Tawain, home, of course, to TSMC. Though he has since paused some of these duties, the U.S. remains in the midst of a mounting trade war with China. 

And though Trump said over the weekend that chips, smartphones and computers will be exempt from his reciprocal tariffs, he is reportedly planning on issuing additional tariffs that would include semiconductors as soon as this week. 

Nvidia’s announcement comes shortly after the company narrowly avoided a new round of U.S. export controls that would have prevented it from selling its older H20 chip in China. The chip was spared after Huang reportedly promised to invest in manufacturing in the U.S. 

While the pledge is now in writing, the project is sure to come against a number of challenges, namely that, though Nvidia’s chips can be fabricated in Arizona, they still might need to be shipped to Taiwan for advanced packaging, according to Reuters. Add to this the lack of available and highly skilled chip workers in the U.S., and Nvidia is facing an uphill battle. 

It isn’t clear if chips produced in Arizona will be more expensive than those produced in Taiwan. It is additionally unclear how many chips Nvidia’s U.S. factories might be able to produce, when fully operational, compared to Nvidia’s total production and demand numbers. 

Nvidia didn’t respond to a request for comment. 

Right now, everything is fuzzy. 

There is a dramatic unpredictability to the tariffs, creating an environment of persistent uncertainty even as the U.S., according to some CEOs, teeters on the edge of a full-blown recession. 

The implications of such an event on the AI ecosystem could be severe. 

From Nvidia, this is a clear step to insulate itself from that chaos. 

But it might be coming too late. 

Microsoft, for one, has said that it plans to slow down its data center push. So, while the hyperscaler still plans to drop $80 billion in AI infrastructure (which prominently includes chips) this year, it seems likely that its AI-related capital expenditures for 2026 will be lower.

Right when Nvidia’s U.S. factories are slated to enter into full production, the feverish build-out that pushed Nvidia to such heights might well begin to slow down, at which point — tariffs aside — Nvidia may experience the other side of its current demand problem: not enough of it.

🤔 Your thought process:

Selected Image 2 (Left):

• “The fake one was too dreamy with photoshopped vibes. And people's feet apparently don't exist.”

Selected Image 1 (Right):

• “The height of the wave seems too high that close to shore. Further out it seems calm.”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!

Here’s how you use GenAI:

31% of you use generative AI for work. 24% use it for search. 16% use it for therapy/general life stuff.

14% of you don’t use it.

Work:

• “Use AI for summary of large chunks of content, creating outlines and buyer personas, and writing meta text (all of which aren't perfect and need to be fact-checked and edited every time).”

Life stuff:

• “I use it for suggestions for physical therapy to better my condition which requires searching through very wide range of exercises available. It’s my buddy for conversation. I also use for suggestions. Advise if I have to make choices for options where there are pros and cons. I want to start for technical analysis of stocks which I do regularly where several indicators are referred to.”

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.