All 11 xAI Cofounders Are Gone — What Happens to Grok, the SpaceX IPO, and the Rebuild

Ross Nordeen left xAI on March 28, 2026. He was the last cofounder. Every person who co-founded xAI alongside Elon Musk in 2023 has now left the company. Musk is the only original founder remaining.

This is not a normal pattern for a company approaching a major IPO. SpaceX — which acquired xAI in February 2026 in an all-stock deal valuing the combined entity at $1.25 trillion — is targeting a mid-2026 IPO at a $1.75 trillion valuation. The company going public has no founding team left except the man whose name is attached to everything.

Who Left and When

xAI launched in March 2023 with 12 cofounders including Musk. The departures started in 2025 and accelerated sharply after SpaceX acquired xAI in February 2026.

2025 departures (early signals):

• Christian Szegedy — former Google Brain researcher, left February 2025
• Igor Babuschkin — former DeepMind, one of the earliest technical leaders
• Greg Yang — mathematical ML researcher, left 2025

2026 departures (the accelerating exodus):

• Tony Wu — announced departure February 10, 2026
• Jimmy Ba — resigned within 24 hours of Wu, reportedly amid performance pressure on model benchmarks. Ba co-authored the 2014 Adam optimizer paper — the most-cited paper in the history of AI with 95,000+ citations
• Toby Pohlen — left late February 2026
• Zihang Dai — left early March 2026
• Guodong Zhang — left early March 2026
• Manuel Kroiss — left mid-March 2026
• Ross Nordeen — left March 28, 2026 — the last one

Nordeen was not a researcher. He was the operator — he came from Tesla where he built data centers for the Full Self-Driving system, and at xAI he was responsible for compute strategy and the hardware and software layer of xAI's data center infrastructure. He reported directly to Musk. Losing him is not losing a research paper author. It is losing the person who knew where the physical infrastructure was and how it worked.

What Musk Actually Said

On March 13, 2026, two weeks before Nordeen's departure, Musk posted on X: xAI "was not built right the first time around, so is being rebuilt from the foundations up."

This is an extraordinary public admission. The company Musk founded in 2023, raised $6 billion for, merged with his $250 billion AI brand, and is folding into a $1.75 trillion IPO vehicle — was, by his own account, not built right. He is rebuilding from foundations.

He did not specify what was wrong. The surrounding context: Grok, xAI's flagship model, has consistently underperformed against GPT-5 and Claude on independent benchmarks. Grok 3 showed improvement over Grok 2 but did not close the gap against frontier models from OpenAI and Anthropic. The tools Musk specifically called uncompetitive were xAI's coding assistants — the category where Cursor, Claude Code, and GitHub Copilot have established strong developer loyalty.

Musk also faced a separate problem: Grok was the subject of government investigations after it enabled generation of non-consensual sexual deepfake images. That controversy, combined with benchmark underperformance and the cofounder exodus, created a compounding credibility problem heading into an IPO process.

Why the Founding Team Left

The official story from Musk is that the exits were pushed, not pulled — that he was clearing out people who were not performing. He posted on X in February that the departures were "not surprising" and implied they were part of deliberate restructuring.

The alternative read from the people who left: the SpaceX acquisition changed the operating environment in ways that made xAI unappealing to researchers who joined to work on frontier AI.

The specific tension is structural. SpaceX is a hardware and engineering company with a culture of extreme execution and hierarchy around Musk. xAI was originally structured as a research lab competing with DeepMind and Anthropic — environments that give researchers significant autonomy to pursue their own directions. Post-acquisition, xAI researchers were working inside a SpaceX-dominated culture with SpaceX management above them and Musk's direct attention pulled in five directions simultaneously.

Jimmy Ba's departure specifically was tied to performance pressure — reports indicate Ba and others were given aggressive timelines to improve Grok's benchmark performance and that the pressure crossed a threshold that made staying untenable. Researchers with Ba's profile (Adam optimizer, academic prestige, multiple competing offers) do not need to stay somewhere the conditions are bad.

Meta is offering up to $300 million over four years for top AI researchers. OpenAI, Google DeepMind, and Anthropic are all aggressively recruiting. The AI talent market in 2026 is the most competitive in history. Any researcher who left xAI had offers waiting before they put in their resignation.

What the SpaceX IPO Risk Actually Looks Like

SpaceX is targeting an IPO in mid-2026 at a $1.75 trillion valuation. xAI is now a significant component of the combined SpaceX/xAI entity that will go public. The valuation assumes xAI contributes meaningful AI revenue and competitive model capability.

The IPO narrative requires investors to believe that:

1. Grok will close the benchmark gap against GPT-5.4 and Claude
2. xAI's data center infrastructure (the part Nordeen built and understood) will scale effectively
3. The rebuild Musk announced in March will produce a competitive product before the IPO filing

None of these require the original founding team to be present. The research and infrastructure they built still exists. But institutional investors doing diligence for a $1.75 trillion IPO will ask why every founder except the CEO left in the 6 months before the filing.

The standard investor question: if the founding team built something great, why did they leave before the payoff? Founders who believe in the company they built stay for the IPO. Founders who believe the company is heading the wrong direction — or that the returns will not materialize — leave.

Musk's response to this narrative is the "not built right" framing. He is preemptively arguing that the founders who left were associated with the flawed first version, and the rebuild is a fresh start that the IPO will fund. Whether institutional investors accept this depends heavily on what the rebuilt Grok actually demonstrates before the S-1 is filed.

What the Rebuild Actually Means

"Rebuilt from foundations" is not a specific technical claim. But the surrounding evidence points to a few things:

Infrastructure: Nordeen's departure is the most concrete loss. He built the physical infrastructure that runs xAI's training and inference. The "rebuild from foundations" almost certainly includes rebuilding or significantly restructuring the data center architecture. xAI announced the Colossus cluster in Memphis — 100,000 H100 GPUs — in 2024. Whether Colossus was "built right" or is part of what needs rebuilding is unknown.

Model architecture: The Grok model family has not demonstrated the benchmark trajectory that GPT-5.x or Claude 3.x showed. A foundation-level rebuild suggests xAI is reconsidering the model architecture, training approach, or data pipeline rather than iterating on the existing Grok framework.

Team composition: Post-departure, xAI is recruiting heavily. The company is hiring engineers from SpaceX's culture — execution-focused, willing to work under Musk's direct leadership style — rather than academic researchers. The profile of the team being rebuilt is different from the profile of the team that left.

What This Means for Grok and Developers

Grok is available via API on xAI's developer platform. Grok 3 is the current production model. Developers building on Grok need to understand the risk profile:

The rebuild creates uncertainty. A company that publicly admits its product was "not built right" and is rebuilding from foundations is not in a stable development cycle. API stability, model versioning, and capability roadmaps are all uncertain during a rebuild.

Grok's pricing is aggressive. xAI has undercut OpenAI and Anthropic on API pricing to gain market share. The question is whether pricing reflects sustainable unit economics or a pre-IPO customer acquisition strategy that changes post-listing.

The talent departure matters for model quality. Researchers like Jimmy Ba (Adam optimizer), Tony Wu, and Greg Yang were working on Grok's training methodology and model architecture. Their exits mean the team running the rebuild is not the team that built the existing models.

For developers evaluating the Grok API: the pricing and context window are competitive. The uncertainty is in the roadmap — the rebuild timeline, the IPO process, and whether the new team produces a model that justifies the $250 billion xAI valuation embedded in the SpaceX IPO.

The Broader Pattern: AI Lab Founder Departures

xAI is not the only lab seeing founder exits. The pattern across the industry:

• OpenAI: Sam Altman fired and rehired (2023), Ilya Sutskever left (2024), Greg Brockman resigned (2024), John Schulman went to Anthropic (2024)
• Google DeepMind: multiple senior researchers left to found Isomorphic Labs, Inflection, and others
• Anthropic: itself founded by OpenAI departures — Dario Amodei, Daniela Amodei, and six others

The difference with xAI is timing and completeness. OpenAI, despite losing founders, retained substantial research leadership below the founder level. xAI lost all 11 non-Musk cofounders within approximately 18 months of its founding — and the remaining leadership structure has not been publicly clarified.

Key Takeaways

• All 11 xAI cofounders are gone — Ross Nordeen, the last, left March 28, 2026. Musk is the only original founder remaining
• Nordeen's specific loss: compute strategy and data center infrastructure — the operational backbone of xAI's training capacity
• Musk publicly admitted xAI "was not built right the first time" and is being rebuilt from foundations — an extraordinary pre-IPO admission
• SpaceX IPO risk: combined entity targeting $1.75T valuation in mid-2026 — institutional investors will scrutinize why every founder left before the payoff
• Jimmy Ba's departure is the benchmark signal — he co-authored the most-cited paper in AI history (Adam optimizer, 95K citations) and left amid performance pressure on Grok
• AI talent market context: Meta offering $300M/4yr packages; OpenAI, DeepMind, Anthropic all aggressively hiring — departed founders had immediate alternatives
• Developer implication: Grok API remains live and competitively priced, but rebuild uncertainty means the roadmap is opaque; evaluate accordingly