HARDWARE

Microsoft adopts Stargate's orphaned AI compute

Microsoft may be lapping up OpenAI’s data center leftovers. 

The company has agreed to rent data center capacity from neocloud firm Crusoe in Abilene, Texas, that was originally developed for Oracle and OpenAI, according to Bloomberg. The company will reportedly lease out 700 megawatts of capacity adjacent to the Stargate campus.

Microsoft will essentially pick up where OpenAI and Oracle left off after the two companies walked away from talks to lease the site, Bloomberg reported. The reported deal follows OpenAI and Oracle dropping plans to expand their flagship Stargate campus. 

OpenAI has faced bumps in the road in its grand plan to spearhead a historic AI infrastructure buildout. In February, the company narrowed its spending expectations for AI compute to $600 billion by 2030, down from $1.4 trillion, amid mounting concerns about its ability to generate revenue. 

However, other industry players have different outlooks. Microsoft, for instance, inked $50 billion in commitments to lease server farms in its most recent earnings report. Microsoft’s reported deal is a sign that AI-fueled data center demand is far from slowing down, Andrew Sharp, research lead at Info-Tech Research Group, told The Deep View. 

Power and grid capacity are increasingly becoming constraints on this buildout, said Sharp, driving prices up. Oracle, meanwhile, is carrying a far higher debt-to-earnings ratio than other cloud hyperscalers, Sharp noted. To put it plainly, if Oracle can’t fund OpenAI’s projects as aggressively as the hyperscalers fund their own, the so-called “Big Three” – Amazon, Google Cloud, and Microsoft Azure – will make faster gains. 

“Microsoft is buying megawatts wherever it can find them,” said Sharp. “Unclaimed power capacity is the new beachfront property, and everyone from hyperscalers to independents is scrambling for whatever's left.”

Microsoft’s bid for more power may show that OpenAI isn’t the sole indicator for the state of the AI data center buildout. OpenAI made a lot of big promises and got itself tangled in expensive commitments, unlike its primary rival, Anthropic. Thus far, it has leaned on legacy tech companies and investors to bring its vision to life. Though OpenAI has made itself a fixture in the AI space, the industry shouldn’t look at its success or slip-ups as the only bellwether. Power, compute, and ultimately influence are being shared by a handful of the leading players. Traditional powerhouses like Microsoft, Google, and Amazon have the benefits of significant capital, profitable core businesses, and existing data center footprints. 

TOGETHER WITH DRYRUN SECURITY

What happens when AI coding agents start writing real software?

We tested it. DryRun Security put Claude, Codex, and Gemini through a real development workflow to see what happened when agents built and shipped features using a normal development workflow.

The findings are clear: 143 security issues surfaced across 38 scans, and 87% of pull requests introduced at least one vulnerability. From insecure JWT defaults to weak refresh token handling and missing rate limiting, the same flaws appeared again and again.

See the breakdown in The Agentic Coding Security Report. Download the report.

POLICY

White House launches free AI literacy program

AI tools are everywhere, but they're only helpful if people know how to use them. That's the AI literacy problem — and the White House wants to fix it.

On Tuesday, the US Department of Labor announced the “Make America AI-Ready” initiative, which aims to teach American workers the basics of AI with a free course. The course takes place entirely via text message, with the intention of making it accessible to all Americans, regardless of laptop or internet access. 

“The ‘Make America AI-Ready’ initiative is designed to ensure every American worker has the chance to learn foundational skills so they can benefit from the opportunities that the AI economy presents,” US Secretary of Labor Lori Chavez-DeRemer said in the press release. 

The initiative was built through a public-private partnership between the Labor Department and Arist, an AI-powered learning platform. The course itself can be completed in seven days, with 10 minutes a day. Some course details include: 

• Delivery: “Bite-sized” daily learning content and challenges

• Content: Aligned with the five foundational areas outlined in the AI Literacy Framework, released by the Labor Department in mid-February

• Access: All interested users have to do is text “READY” to 20202

• Privacy: The phone numbers will only be used for delivering the course and not sold or shared with third parties or used for promotional purposes, according to the release.

As the course was designed to be introductory, participants who finish it will be offered additional resources to learn advanced AI skills aligned with their unique goals or interests. By the end of the seven days, the curriculum will have covered AI principles, use cases, best practices, and more. 

Ultimately, the program seeks to make AI literacy more accessible in response to a growing skills gap. Though AI-powered coding has completely shifted the roles of technical workers, AI executives and enterprises alike are eyeing a broader shift, some claiming that these tools may just be the beginning of a metamorphosis of all knowledge work. 

With sweeping change in the way, those who can use AI effectively stand to gain significant advantages in productivity and learning, while those without access to or understanding of it risk being left behind.

AI literacy is a critical skill beyond being able to leverage it in the workplace. A broader concern is that, without the ability to critically evaluate AI-generated content, misinformation becomes a serious threat. As a result, AI literacy should be a priority, and this initiative is a promising step toward making it accessible. That said, challenges remain with this approach, namely, whether there is sufficient public demand for enrollment, and whether the course content is comprehensive enough to serve learners across varying skill levels. It’s why many AI firms are pushing for AI literacy courses in K-12 school curricula, seeking to address both issues.

TOGETHER WITH TWELVELABS

How To Review & Approve Video Content In Seconds, Not Weeks

As the old saying goes, "time is money" and that's especially true when it comes to reviewing video content for approval. The faster you approve, the sooner it gets out into the world, the quicker your ROI starts to take effect. Simple, right? Sure... unless you're dealing with 60,000+ creators like AffiliateNetwork does.

They needed a way to verify and validate thousands of videos fast, all without growing headcount or hitting bottlenecks. Rather than settle for general AI solutions, they turned to TwelveLabs, whose temporal reasoning and unique architecture allowed them to check for compliance in seconds using natural language rules (like "show the logo for 2+ seconds"). The end result is a lean team that serves 60,000+ creators with zero bottlenecks.

Want to do the same for your organization? Check out TwelveLabs right here.

GOVERNANCE

OpenAI funds $1B to fix AI's downsides

OpenAI has billions in funding earmarked for “scaling AI for everyone.” Now, it’s reaching out a hand. 

On Tuesday, the company's non-profit arm announced that it would spend $1 billion on AI-related philanthropic efforts through the OpenAI Foundation, part of its previously announced $25 billion commitment to “curing diseases and AI resilience.” OpenAI said it would focus the funding on several different efforts, including life sciences, economic impact and community programs. 

Additionally, the company tapped several new leaders for this work, including Wojciech Zaremba, a co-founder of OpenAI, as head of AI resilience; Jacob Trefethen from Coefficient Giving as head of life sciences and curing diseases; and Anna Makanju, former VP of global impact at OpenAI, as head of AI for civil society and philanthropy.

OpenAI will also invest in initiatives to protect society from the biggest risks of AI, including “biological threats” and data-driven research and evaluation on AI’s impact on youth.

“Advanced AI models will also present new challenges that are already surfacing, and we need to be prepared to identify these challenges and develop solutions to address them,” OpenAI said in its announcement. 

The funding skyrockets the philanthropy that OpenAI’s non-profit division has done thus far: According to Bloomberg, the OpenAI Foundation gave away $7.5 million last year. And in December, the OpenAI Foundation said it would award over $40 million to more than 200 nonprofits across the country. 

However, OpenAI’s giving mood follows Anthropic co-founder pledges to donate billions to campaigns dedicated to “effective altruism,” or philanthropy. In other words, donate to causes where their money can do as much societal good as possible.

OpenAI and other tech giants that donate to these causes are essentially aiming to solve problems that their models are causing. Rather than slowing down the development of their models, they’re addressing problems as they arise: mental health impacts, biological threats, model safety, and more. Additionally, funding efforts in life sciences and disease cures are another attempt to push the narrative of developing ultrapowerful systems in the name of human progress. So we have to keep in mind that this still serves the AI industry's larger utopian narrative. 

LINKS

• Arm, Meta introduce “AGI CPU” for agentic infrastructure

• OpenAI to scrap Sora AI video platform to focus on enterprise

• Amazon’s Zoox to launch robotaxis in Austin and Miami later this year

• Halter, a startup making AI-collars for cows, raises $220 million

• Internet Watch Foundation report 260x rise in AI child abuse videos

• Beehiiv launched a way for users to manage their accounts from AI platform 

• Andrew Bosworth to lead Meta’s initiative to become AI-native

• Claude: Users can now use Claude on their computer to complete tasks. 

• Figma: The design platform enabled support for AI agents on the Figma canvas

• ChatGPT: An update makes it easier for users to find, resume and build on files uploaded to the chatbot

• Luma: The AI lab launched Uni-1, a new model that can think and generate pixels at the same time

• OpenAI: The company released  prompt-based safety policies developers can use to create age-appropriate protections for teen

• Nvidia: Senior Research Scientist, Efficient Deep Learning

• TSMC: Senior AI Research Scientist

• Meta: AI Research Scientist, Media - Meta Superintelligence Labs

• MongoDB: AI Staff Developer Advocate

GAMES

A QUICK POLL BEFORE YOU GO