Hey everyone, here’s the latest scoop straight from the source. On June 28th local time, Elon Musk took to social media with some big news: his AI powerhouse, xAI, has officially kicked off private testing for its newest brainchild, Grok 4.5, right here inside SpaceX and Tesla. According to Musk, this isn’t just another update; it’s built on a colossal 1.5 trillion parameter V9 base model. What makes it special? They’ve specifically fed it data from the hot new AI coding tool, Cursor, to sharpen its skills.
The early verdict? The performance numbers are looking insane. Musk hinted that initial tests show Grok 4.5 is already neck-and-neck with, or potentially even surpassing, Anthropic’s flagship beast, Claude Opus. And we aren’t done yet—reinforcement learning (RL) is still actively fine-tuning the model, and their custom benchmark suite, “Grok Build,” is getting more robust by the day.
But wait, there’s more! Musk also teased an aggressive roadmap: starting now, SpaceX plans to roll out brand-new models trained from scratch every single month for the rest of the year. That’s one serious pace!
For context, let’s look at the competition. Anthropic, founded back in 2021, has taken a slightly different route compared to OpenAI. While OpenAI tries to balance consumer apps (C-end) with enterprise tools (B-end), Anthropic has focused heavily on the enterprise market. Their Claude series is the go-to for businesses. Just recently, on May 28th, they released Claude Opus 4.8. Then, 11 days later, on June 9th, they unveiled the next generation: Claude Fable 5 and Claude Mythos 5. Both share the same underlying architecture, but while Fable 5 is open to the public, Mythos 5 keeps its full power locked behind closed doors for a select few trusted institutions.
Now, let’s talk about xAI itself. Established in July 2023, Musk launched it with a lofty vision: “to understand the true nature of the universe.” He assembled a “dream team” of 12 top-tier scientists from giants like Google, DeepMind, OpenAI, and Microsoft Research—including five Chinese experts. Back then, everyone thought this was the squad ready to take down the industry titans.
In February, just before SpaceX planned its IPO, xAI got acquired by SpaceX. Post-merger, the combined valuation skyrocketed to a mind-boggling $1.25 trillion.
xAI hasn’t been sitting idle. They’ve rolled out multiple iterations and secured massive funding rounds over the years. Here’s the timeline: In December 2023, they dropped the first-gen Grok, and by December 8th, the beta was open to all US-based X Premium+ subscribers. Fast forward to March 2024, they released Grok-1 with a staggering 314 billion parameters—way bigger than OpenAI’s GPT-3.5. That same year, they bagged two huge funding rounds: $6 billion for Series B and another $6 billion for Series C. By 2025, they launched the standalone Grok app and various model versions. Most recently, in January 2026, they completed a massive $20 billion Series E round, pushing their valuation to $230 billion.
However, the road hasn’t been smooth. Internal documents reveal xAI is under immense financial pressure. In Q3 2025, net losses jumped to $1.46 billion, up from $1 billion in Q1. For the first nine months of 2025 alone, cash burn hit a whopping $7.8 billion. Sure, they raised that $20 billion earlier this year, but most of it went straight into infrastructure and R&D, which ironically widened the short-term losses even further.
With the general AI market getting crowded and profit models still murky, xAI is pivoting hard toward vertical-specific solutions. Financial AI is the target. Through enterprise subscriptions and custom services, they see a clear path to monetization. Right now, xAI is betting big on financial AI to find new revenue streams and ease the bleeding from those continuous losses.
Back on March 16th, job postings on the xAI website gave us a clue. Musk’s company was launching a massive hiring drive, specifically targeting Wall Street bankers, portfolio managers, traders, and credit analysts. These folks aren’t just joining as coders; they’re being hired as “AI Mentors” for the data labeling teams. Their mission? To teach the Grok models how to think like top-tier investment bankers, effectively upgrading Grok into a “senior analyst” capable of handling complex financial scenarios.
According to those public job descriptions, the roles cover everything from traditional banking and credit analysis to crypto and stock market expertise. Their main job is training Grok on complex tasks like financial modeling, asset valuation, industry research, and trade analysis—the exact core duties of a human investment banker.
By mid-May, rumors surfaced that xAI had already onboarded several Wall Street firms to test Grok. The goal? To boost revenue before SpaceX hits the IPO stage. Big names like Apollo Global Management and Morgan Stanley have started using Grok internally, testing it alongside other AI competitors. But here’s the catch: even though some banks have signed deals, actual usage by finance professionals in daily work remains surprisingly low.
Finally, let’s dive into the wild side: space-based computing. Musk is a huge proponent of this idea. In SpaceX’s prospectus, they painted a picture where Starship launches millions of tons of computing gear into orbit. Imagine using infinite solar energy and the vacuum of space for ultra-low-cost, high-reliability compute power. Based on a formal filing with the FCC in January, SpaceX plans to deploy up to 1 million computing satellites in Low Earth Orbit (500–2000 km). These satellites will run on solar power, exchange data via inter-satellite laser links, and integrate seamlessly with the existing Starlink network. Musk calls this the first step toward a “Kardashev Type II civilization,” aiming to break free from Earth’s physical power limits for AI growth.
Progress is already underway. SpaceX has finished designing the first-gen AI satellite, the “AI1.” It features swappable compute modules, giant solar panels, and deployable liquid-cooled radiators to handle heat dissipation naturally in the vacuum of space. Musk’s timeline? They aim to achieve an annual deployment rate of 1 gigawatt of space-based AI capacity by the end of 2027, scaling up as Starship’s reusability matures. If they can launch 1 million tons of payload annually, the theory suggests they could build a 100 GW solar AI satellite cluster, potentially reaching a massive 1 Terawatt (TW) of compute power. Now that’s what I call thinking outside the box!
