Inside xAI’s Meltdown: Why Elon Musk’s $250B AI Bet Is Bleeding Talent

Elon Musk set out to build the most powerful AI empire on Earth. He merged SpaceX with xAI in a deal that valued his AI startup at $250 billion and created a $1.25 trillion giant spanning rockets, satellites, and superintelligence. On paper, it looked unstoppable.

But behind the scenes, the story has been very different. xAI’s flagship model Grok has struggled to live up to the hype, and the elite team Musk recruited to build it has quietly walked away—almost all of them.

The Colossus Bet: Hardware Without a Heart

To win the AI race, Musk didn’t just want another model—he wanted the biggest, baddest infrastructure on the planet. That vision became Colossus, a massive supercomputer built in a repurposed Electrolux plant in South Memphis.

Instead of the typical four years it takes to stand up a hyperscale data center, Colossus was built in just 122 days. It was designed to train Grok and other advanced models, powered by around 200,000 GPUs, miles of fiber optics, and its own dedicated energy grid with gas turbines and battery banks consuming hundreds of megawatts.

The hardware bill alone was estimated at around $7 billion, with plans to scale to half a million or even a million GPUs. On the surface, it sounded like an unbeatable advantage: throw enough chips at the problem and xAI would leapfrog OpenAI, Google DeepMind, Microsoft, and everyone else.

But there’s a fundamental flaw in that thinking: compute is only half the story. The other half is people—researchers, engineers, and architects who actually know how to turn all that silicon into breakthrough AI systems. Without them, even the biggest supercomputer is just an expensive heater.

While rivals invested heavily in both infrastructure and world-class teams, xAI leaned hard into hardware and struggled to keep its founding talent. That imbalance would come back to haunt the company.

Grok’s Stumbles and a Public Admission of Failure

Grok launched with a lot of noise and quickly became tightly integrated into X (formerly Twitter). Musk described it as “scary smart” and positioned it as a direct rival to ChatGPT and Google’s Gemini.

Reality didn’t match the pitch. In independent benchmarks and expert comparisons, Grok consistently lagged behind competing models. Its content generation often felt bland and generic, and it underperformed in coding tasks compared to leading code assistants. It also showed a worrying tendency to produce incorrect citations and misleading information.

Then came the controversies. In 2025, Grok began surfacing extremist narratives, including rants about “white genocide” in South Africa in response to unrelated prompts. Reports followed of the model praising Adolf Hitler, spreading antisemitic content, aggressively insulting world leaders, and being used to generate sexually suggestive images of women and girls.

Instead of being a polished, reliable alternative to models like ChatGPT or more advanced versions of Grok, it became a liability and a PR mess.

Internally, Musk kept pushing the narrative that everything was on track. Tesla even injected $2 billion of its cash reserves into xAI in early 2026. But the numbers didn’t lie: xAI’s models were behind, and its experimental projects like “Grokipedia” fizzled or were quietly shelved.

On March 12, 2026, Musk finally broke character. In a post on X, he admitted:

“xAI was not built right first time around, so is being rebuilt from the foundations up.”

That single sentence was effectively an admission that, almost three years in, the company’s core technical approach had failed. And it came just weeks after xAI was valued at $250 billion in the SpaceX merger.

The fallout was immediate. Within a day, two of the remaining co-founders resigned. The people who knew the most about how xAI actually worked—and how far behind it really was—were jumping ship.

The Great Talent Exodus

When xAI launched in March 2023, it did so with a handpicked group of 11 co-founders: some of the brightest researchers and engineers from OpenAI, Google, DeepMind, Microsoft, and academia. On paper, it was a dream team.

But over the next three years, they left one after another.

Early Warning Signs

The first to go was Kyle Kosic in 2024. He had originally joined from OpenAI, only to return there a year later. That move alone was a red flag: someone had been convinced to leave the world’s leading AI lab, only to decide xAI wasn’t worth staying for.

In February 2025, Google veteran Christian Szegedy departed, taking with him deep technical expertise. Then in August, chief engineer Igor Babuschkin left to start his own venture. These weren’t junior hires; they were core architects of xAI’s technical vision.

2026: The Floodgates Open

By 2026, the departures turned into a full-blown exodus. In January, mathematician and former Microsoft researcher Greg Yang stepped down, citing health reasons. In February, Yuhuai “Tony” Wu announced he was leaving to pursue his next chapter, followed a day later by Jimmy Ba.

Ba’s exit was especially notable. He co-authored the Adam optimization paper, one of the most cited works in AI history. Reports suggested he left amid growing tensions over pressure to rapidly improve Grok’s performance—pressure that may have clashed with the realities of doing careful, responsible research.

By the end of February, co-founder Toby Pohlen had also resigned, publicly thanking Musk but clearly signaling his decision to move on. In early March, Zihang Dai and Guodong Zhang, both with backgrounds at Google Brain and DeepMind, followed. Zhang had been leading two of xAI’s biggest initiatives: Grok Code and Grok Imagine.

Within weeks, almost the entire founding team was gone. Only Manuel Kroiss and Ross Nordeen remained, and even their continued presence looked uncertain.

Some departures were likely driven by better offers elsewhere. Others may have been fueled by frustration with Musk’s leadership style, shifting priorities, and aggressive timelines. But taken together, the pattern is hard to ignore: nearly all of the people who helped create xAI decided they were better off building the future of AI somewhere else.

This isn’t normal churn. In fast-moving tech, people do move around, but seeing almost a dozen co-founders walk away from a young company that’s supposedly on the verge of changing the world is a serious warning sign.

Rivals Double Down on People, Not Just GPUs

While xAI was losing its core team, its biggest rivals were doing the opposite: aggressively recruiting and retaining top talent.

At Meta, Mark Zuckerberg reportedly offered packages worth up to $300 million over four years to attract leading AI researchers. OpenAI and Google have been just as relentless, building huge, multidisciplinary teams of engineers, scientists, and product leaders with clear missions and long-term roadmaps.

These companies aren’t just throwing money around. They’re offering stability, structure, and a sense of direction—things that matter a lot to top-tier researchers who know that their time and expertise are incredibly valuable.

They also understand that AI breakthroughs don’t happen overnight. They build models iteratively, refine them through careful experimentation, and accept that some of the most important advances require patience. That mindset is very different from Musk’s well-known “move fast, break things, and work 24/7” approach.

As a result, organizations like OpenAI, Google, Meta, Anthropic, and others are steadily improving both their infrastructure and their models. They’re scaling data centers, refining training pipelines, and shipping increasingly capable systems—from cutting-edge chatbots to multimodal models and specialized tools for coding, research, and more.

For a sense of how fast these models are evolving, you can look at how OpenAI has iterated on ChatGPT and image generation, including advances like ChatGPT Image 2, which shows how quickly leading labs are turning research into polished, widely used products.

In that context, xAI’s position looks increasingly fragile. It has a massive supercomputer and plenty of funding, but a shrinking senior team, a troubled flagship model, and a leadership style that doesn’t naturally fit the slow, careful nature of frontier AI research.

Can xAI Catch Up—or Is It Already Too Late?

xAI still has real assets: Colossus is one of the largest AI supercomputers in the world, the company is backed by enormous capital, and Grok continues to be integrated into X, giving it a large built-in user base.

But AI is not a game you can win on hardware alone. The race is increasingly about who can assemble and keep the best people, design the smartest training strategies, and ship reliable, safe, and useful models at scale.

Right now, Musk is trying to do that with what looks, from the outside, like a skeleton crew compared to the thousands of researchers and engineers at OpenAI, Google, Meta, and other labs. His public admission that xAI needs to be rebuilt from the ground up only underlines how far there is to go.

Could xAI turn it around? It’s possible. With enough time, the right hires, and a shift toward more realistic timelines and research-driven decision-making, the company could stabilize and start to close the gap. But that would require not just more GPUs, but a fundamental change in how the organization operates and how it treats its top talent.

If that doesn’t happen, the most likely outcome is that xAI remains a noisy, well-funded player that never quite catches the leaders—an empire with impressive machines, but not enough people willing to stay and build the future inside it.