Smooth Operator

Hello, and happy Sunday! Silicon Valley came to Washington to announce a $500 billion funding effort aimed at boosting AI infrastructure in the United States. Alex Duffy explores what we know—and what we don’t know—about the initiative known as “Stargate.” Then, check out everything we published this week at Every.

ICYMI: The next cohort of our course How to Write With AI, taught by Evan Armstrong, starts on February 13. You’ll learn how to fashion expert prose and get your writing to resonate with your intended audience while harnessing the latest AI technology. Head to the course website to learn more and sign up.—Kate Lee

Release notes

Stargate, unicorns, and autonomous agents 

This week, the Trump White House announced a $500 billion private sector AI infrastructure investment called the Stargate Project. Funded by SoftBank, OpenAI, Oracle, and the Emirati investment firm MGX, the project counts Microsoft, Nvidia, and British semiconductor company Arm as partners. Its goal: improve AI infrastructure in the United States.

Company executives, many of whom gathered at the White House to announce the deal on January 21, said they will invest this money over the next four years with an initial sum of $100 billion, which is already being used to build the first of 10 500,000-square-foot data centers. 

While $500 billion is a staggering sum, the money comes from private firms, not taxpayers. And it’s not all new: Some of the companies, such as Microsoft, had already committed funds toward building out AI infrastructure. Now it’s under the Project Stargate umbrella, getting a presidential endorsement as a driver of AI innovation in the country. It is also unclear exactly what the funds will be used for, and critics—including Elon Musk—have publicly questioned whether the number is, in fact, exaggerated. However, two things so far are certain:

1. The Stargate Project will advance the development and adoption of AI in the United States.

2. Despite what Elon Musk claims, Microsoft CEO Satya Nadella says the company is good for its $80 billion investment.

This announcement came at a critical moment, as AI gains mainstream adoption and rapidly improves. Meta’s Mark Zuckerberg recently boasted that the company will roll out AI agents capable of acting as mid-level engineers later this year, while three days ago OpenAI launched its agent Operator, which promises to accomplish online tasks for users autonomously. There’s an unmet market need for helpful AI agents. Devin, a much-hyped $500-per-month “collaborative AI teammate,” often falls short, failing 70 percent of real world tasks in one test. But even the CEO of Anthropic, Dario Amodei, who is relatively reserved and hype-averse, believes AI that's ”better at most things than most people” is coming in the next two to three years.

Many people will benefit from these improving models and tools. For individuals, a recent survey conducted by newsletter writer Lenny Rachitsky revealed that the most widely adopted tool across 6,500 tech professionals was not Slack or Gmail, but ChatGPT. For industry, companies across the board are beginning to implement agents that are helping their employees improve the effectiveness and efficiency of problems such as customer support, reviewing contracts, or equity research.

But in order to keep training and running the top AI models, developers need better infrastructure—namely, chips and data centers. In that way, Project Stargate should push forward AI development in the U.S., at a time when competition from China is, for the first time, staring us in the face.

On January 20, a Chinese company called DeepSeek released its open-source R1 model. This model caused quite a stir because it is a language model about as good as OpenAI's flagship model o1 while sporting a price tag 30 times cheaper, and was trained by a much leaner team possibly without access to top-of-the-line chips. While impressive, R1 specializes in deep reasoning rather than quick answers. Still, R1 shows that despite the U.S. government’s best efforts to cut off China from high-powered chips, the country is still finding ways to forge ahead with cutting-edge AI development. And for developers anywhere relying on language models for hard problems, switching to R1 today could save money and improve the quality of responses.

In the U.S., Google announced the latest update to its “thinking” model, debuting at number 1 in the Chatbot Arena, a website dedicated to comparing and testing the best AI chatbots. These new “thinking models” from Google, OpenAI, and now DeepSeek break from the trend of needing bigger models for better responses. Instead, they take longer to generate an answer, but the additional reflection improves the quality of the response. However, the size of major AI systems has doubled every six months for 15 years‚—and that was before another $500 billion was invested with Project Stargate. Better and bigger at the same time compound to exponential progress with opportunities to capitalize aplenty.

Clay is another company using AI tools internally to solve real problems by providing data useful for marketing and sales teams. This week, Clay (which Evan profiled last summer) raised $40 million, more than doubling last year's valuation from $500 million to $1.3 billion. Also this week, legal AI startup Harvey is reportedly set to double its valuation to $3 billion, and AI code editor Cursor raised a $150 million Series B funding round to try and automate code.

A $500 billion infrastructure investment, a breakthrough open-source model, and billion-dollar valuations for AI-first companies—it's tempting to call this a watershed moment for AI. But perhaps that misses the point. It is not about individual breakthroughs anymore. It’s about the steady drumbeat of progress across every front: infrastructure, models, applications, and adoption. The foundation for the future is being built.—Alex Duffy

At Every, we’re supporting partners with AI training and implementations. If your team is working on an AI strategy for 2025 and would like to collaborate, reach out to [email protected].