Grok Needs Power. Elon Needs $12 Billion.

Elon Musk’s AI startup, xAI, is reportedly trying to raise up to $12 billion more in funding. According to the Wall Street Journal, the plan is to buy a boatload of Nvidia chips then lease them right back to itself. The funding would come through debt, not equity, which is important, because it signals Musk doesn’t want to dilute ownership but still needs serious cash to keep the engine running.

Leading the charge is Valor Equity Partners, a firm run by Musk’s longtime ally Antonio Gracias. They’re talking to lenders about raising the $12 billion, which would help xAI build a mega-sized data center to train and power Grok, the company’s AI chatbot. Grok is Musk’s answer to OpenAI’s ChatGPT, and like all AI models, it lives or dies on the quality, and quantity, of computing power behind it. That’s where the Nvidia chips come in. They’re the gold standard for AI training, and right now, they’re about as easy to find as a quiet Musk tweet.

This isn't xAI’s first fundraising rodeo. Just three weeks ago, it raised $10 billion. Half debt, half equity. Before that, the company secured another $6 billion from heavyweight investors like Andreessen Horowitz, Blackrock, and Fidelity. That puts xAI’s total capital raised at around $17 billion. And yet, that might already be gone. Reports suggest xAI has blown through most of that cash and could burn through $13 billion more just in 2025.

Earlier this month, The Financial Times reported that xAI was looking to raise more money at a valuation between $170 billion and $200 billion. Musk, of course, publicly denied this, claiming the company had “plenty of capital.” Now we’re seeing what “plenty” looks like: asking for $12 billion more, mostly to cover a growing hardware bill.

Some lenders aren’t entirely sold. They reportedly want the loan paid back within three years and are pushing for a cap on how much xAI can borrow. The concern? Massive loans for a company that’s burning cash in one of the most expensive tech races we’ve ever seen. Building AI at scale is no joke, it’s like stacking servers on bonfires of money. And with rivals like OpenAI raising $40 billion at a $300 billion valuation, and Anthropic pulling in funds at a $61.5 billion valuation, the bar for staying competitive is sky-high.

In short: xAI is in a spending sprint, and Musk is betting big. Again. Whether Grok becomes the next ChatGPT or just another expensive science project will depend on how fast (and how well) they can turn all those chips into something people actually want to use. Until then, expect more headlines, more funding rounds, and a lot more silicon.

Rapid Fire

🥇 Google DeepMind just made AI history. Its latest Gemini model, powered by new "Deep Think" tech, officially earned a gold medal score at the 2025 International Mathematical Olympiad. It solved five out of six mind-bending problems, scoring 35 out of 42 points, enough to beat out nearly 90 percent of the world’s top student mathematicians. For the first time, an AI system wasn’t just playing catch-up in math. It was setting the pace. And it did it in real-time, within the 4.5-hour competition window, using natural language from start to finish. No translations, no shortcuts, just clean, human-readable proofs.

This is more than a math win. It’s a leap in how AI thinks. The system used a method called "parallel thinking," juggling multiple solution paths at once instead of following one rigid track. That let it crank out clear, accurate proofs without needing humans to translate the questions into formal logic first. Just a year ago, DeepMind’s models needed days and expert help to crack similar problems. Now it’s doing it on the fly, outperforming nearly everyone else in the room. OpenAI ran its own tests afterward but didn’t enter the official competition. This time, Google gets the gold and the bragging rights.

🌙 Anthropic CEO Dario Amodei just gave his team a dose of brutal honesty in a leaked Slack message. On July 20, he told employees the company plans to take funding from Gulf states like the UAE and Qatar, even though he openly admitted this would enrich “dictators.” His reasoning? The AI arms race is expensive. Amodei said there’s “easily $100 billion or more” in available capital in the region and that without it, staying at the cutting edge becomes “substantially harder.” In short, if they want to keep pace with the likes of OpenAI and Google, they need the cash, ethics and optics aside.

To ease the blow, Amodei insisted the investments would be “narrowly scoped” and “purely financial” with no strings attached. Still, the memo contradicts his earlier public warnings about the dangers of authoritarian control over powerful AI. Just last year, he wrote that democracies must set the terms for AI to prevent abuse by autocrats. The message paints a clear picture of the high-stakes tradeoffs in the AI industry, where lofty values often clash with the harsh reality of billion-dollar compute bills.

🍎 Apple just dropped iOS 26 beta 4, and it’s a visual course correction. After the “Frosted Glass” backlash from beta 3, Apple decided to bring back more of the original Liquid Glass aesthetic. Navigation bars in core apps like Photos, Music, and the App Store now let more background color shine through. It’s not as see-through as beta 2, but it’s a solid middle ground, transparent enough to feel fresh, not so much that you lose your bearings. You’ll also spot subtler touches like dynamic tints in Notification Center and darker lock screen backgrounds when scrolling through alerts. Translation: Apple’s still obsessing over the details, and it shows.

The other big news? AI-powered notification summaries are back. You’ll get bite-sized rundowns of news and entertainment stories, but with a disclaimer from Apple that basically says, “Hey, this might mess up the meaning of the original headline. Double-check, okay?” So, a handy feature with a built-in shrug. The beta also introduces new dynamic wallpapers, and companion updates dropped for iPadOS, macOS, watchOS, tvOS, and visionOS. Expect the public beta later this week. If Apple sticks the landing, iOS 26 could be its most stylish (and slightly smarter) update since iOS 7.

Tech Radar

OpenAI has been confirmed as the mystery customer behind Oracle’s jaw-dropping $30 billion-a-year cloud computing deal, one of the largest in tech history. The scale is hard to overstate: OpenAI will rent about 4.5 gigawatts of power capacity from Oracle’s data centers. That’s roughly a quarter of all data center power currently operating in the U.S. and enough energy to power millions of homes. The Abilene, Texas site alone is expected to host 400,000 Nvidia GPUs, hardware valued near $40 billion. While the deal won’t hit Oracle’s books until 2028, it already signals a major leap for Oracle’s cloud ambitions. For context, Oracle brought in just over $10 billion from data centers last year. This deal triples that.

The agreement is part of OpenAI’s ambitious “Project Stargate,” a $500 billion infrastructure initiative backed by Oracle, SoftBank, and others. So far, about $50 billion has been raised. Oracle is expected to build out massive new campuses in states like Texas, Michigan, Ohio, and New Mexico as part of the expansion. While Microsoft Azure remains a major cloud partner for OpenAI, this move solidifies OpenAI’s multicloud strategy, with Google Cloud and CoreWeave also in the mix. Beyond the headline numbers, the deal shows just how much raw power is needed to keep AI at the bleeding edge, and how Oracle, once also-ran in the cloud, is suddenly front and center.

Recently Deployed

Silicon Valley startup xLight just raised $40 million to build a prototype that could change the game in chipmaking. The company is working on a next-generation laser for extreme ultraviolet lithography, or EUVt: the tech behind today’s tiniest, fastest chips. EUV machines are the most expensive and essential tools in a semiconductor fab, and right now, Dutch company ASML is the only supplier in the world. xLight’s laser, built on the kind of particle accelerator tech used in U.S. national labs, promises to be more energy-efficient and ten times better than existing options. The goal is to strengthen U.S. leadership in chip tech and compete with China, which is ramping up its own EUV development.

The company is backed by big names, including Playground Global and former Intel CEO Pat Gelsinger, who now chairs xLight’s board. Gelsinger called it a chance to fix past mistakes, like letting U.S. laser tech fall into foreign hands. xLight is sourcing components from American labs and plans to keep its supply chain rooted in the U.S. and allied countries. With export controls already keeping EUV machines out of China, this project could help the U.S. stay ahead in the chip race. The stakes are high, but so is the payoff. If xLight delivers, it could be the spark that keeps Moore’s Law alive.

That’s a wrap.

Between Musk’s chip binge, OpenAI’s $30B cloud flex, and DeepMind’s mathlete moment, one thing’s clear, AI infrastructure isn’t just a tech story anymore. It’s the economy. Whether you’re raising billions to lease your own GPUs or solving Olympiad-level problems in your sleep, the future is being built (and burned through) one data center at a time.

Oh, and if you thought ethics might slow any of this down, ask Anthropic. Turns out, values cost more than $100 billion.

We’ll be back tomorrow with more big bets, flashy updates, and eyebrow-raising moves from the tech world’s most caffeinated players. Until then, keep your servers cool and your summaries smarter than Apple’s AI.