There's a lot happening with xAI and Grok right now.

A merger, an aggressive rebuild, and a lawsuit that goes to trial next month.

Let me break it all down.

My video if you prefer watching:

The Merger Nobody Fully Absorbed

On February 2nd, SpaceX acquired xAI in the largest corporate merger in history.

Combined valuation: $1.25 trillion. xAI alone was valued at $250 billion.

That's not the interesting part. The interesting part is why.

SpaceX has a planned IPO in July 2026. When that happens, xAI's financials become public for the first time. xAI burned through $7.8 billion in cash in just the first nine months of 2025. The enterprise business is a fraction of what OpenAI, Anthropic, and Google have built.

Folding xAI into SpaceX before the IPO gives the combined entity a better story to tell investors. Grok and Colossus add AI credibility to SpaceX's space infrastructure. Starlink gets AI optimization. Starship gets simulation tools.

But it also puts a hard deadline on fixing everything with xAI.

"Not Built Right"

Last week, Musk posted on X: "xAI was not built right first time around, so is being rebuilt from the foundations up."

That's an extraordinary thing to say publicly about a $250 billion company. But it explains everything that's happened since.

Ten of the twelve original co-founders have departed. The remaining leadership is a new team, hired fast, all reporting directly to Musk. The biggest hires:
Andrew Milich and Jason Ginsberg joined from Cursor, the AI development environment that went from zero to $2 billion in annual revenue under their watch. They're now heading engineering and product for Grok's coding tools.

Devendra Singh Chaplot, a founding engineer from Mira Murati's Thinking Machines Lab and ex-Mistral co-founder, joined to work on model training. That almost certainly means Grok 5.

At a conference last week, Musk publicly admitted xAI is "currently behind on coding" and pledged to catch up to and exceed rivals by mid-2026.

The new team has months, not years, to pull this off before the IPO.

What Grok Actually Is Right Now

The current flagship is Grok 4.20 Beta, rolling out through March. Three variants:

Standard. Fast, non-reasoning, good for everyday tasks.

Reasoning. Slower, thinks through problems step by step.

Heavy. The interesting one. Up to 16 specialized AI agents collaborating in real time. Each agent has a specialty: coding (the coding agent is named Lucas), math, physics, creative writing. You give it a complex task and it spins up a team.

The numbers are solid.

Output speed doubled to 265 tokens per second. Instruction-following jumped 29 points to 82.9%. Hallucination rate dropped to 22%, the lowest xAI has ever measured. Context window goes up to 2 million tokens in some configurations.

Where Grok genuinely leads: search.

It's number one on Search Arena, and its real-time access to X/Twitter data is something no other frontier model has. If you want the most up-to-date information on anything happening on X, Grok wins.

Where it lags: raw coding benchmarks.

SWE-Bench sits in the 72 to 75% range, competitive but not leading. Claude 4, Gemini 2.5, and o3 are ahead. That's what the Cursor hires are there to fix.

The Supercomputer Behind It All

Colossus is xAI's AI training cluster in Memphis, Tennessee. They built the original 100,000 GPU system in 122 days back in 2024. Colossus 2 launched in January 2026 as the world's first gigawatt-scale AI cluster, over 200,000 GPUs scaling toward 555,000, running at roughly 2 gigawatts of power.

In March 2026, xAI announced a $659 million building expansion at the same site.
Grok 5 is actively training on Colossus 2 right now. Multiple Grok model variants are training in parallel simultaneously.

What's Coming: Grok 5

Grok 5 is reportedly a 6-trillion-parameter Mixture-of-Experts model. The architecture means it activates only the relevant portions of its parameters for any given task, so massive capability without proportional compute costs.

Native video understanding is expected. Musk has floated a 10% chance of inventing new physics or technology by 2026.

The Video Generation Nobody's Talking About

While everyone focuses on the text models, Grok's video generation is quietly leading every major leaderboard.

Grok Imagine launched January 28th. It generates 10-second, 720p clips with audio. Extensions up to 30 seconds that preserve the style and music of the original. It's currently ranked first on Video Arena, Image-to-Video, and Editing benchmarks.

Quality, cost, and latency are all leading. This is probably xAI's most underrated product right now.

The Lawsuit

Musk is suing OpenAI. The claim: OpenAI owes him up to $109 billion, calculated from his original charitable donations mapped against OpenAI's current $500 billion valuation. Microsoft is also on the hook for up to $25 billion under the same reasoning.

The judge, Yvonne Gonzalez Rogers, was not impressed. She called the damages report "weak" and said the jury will understand the economist "is pulling these numbers out of the air."

She still allowed it in, noting that startup math works differently than traditional finance.

Her comment on the antitrust claims: "I'm not sure that phase two is going to go anywhere, frankly. There is plenty of competition in this industry."

She may also treat the jury's verdict as advisory, meaning she makes the final call herself.
Trial starts April 27th. Musk is a potential witness.

The Honest Summary

xAI is in a genuinely interesting position. The compute infrastructure is world-class. The video generation is leading. The real-time X data advantage is real. And Grok 5 on Colossus 2 could be a genuine step-change if it delivers.

But the enterprise business is thin, the coding product is behind, 10 of 12 co-founders are gone, and the company burned almost $8 billion in nine months. The SpaceX IPO timeline creates a hard clock on all of it.

Musk is swinging for the fences and this year will be insane for him an xAI.

Until Next Time,

Wes “We didn’t even talk about the space datacenters” Roth