Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchWith 10 of 12 original co-founders gone and Grok falling behind rivals, Elon Musk is aggressively recruiting Cursor engineers to rebuild xAI — the most expensive AI talent crisis in Silicon Valley history.
AI, startups & growth insights. No spam.
TL;DR: xAI, Elon Musk's AI company, is undergoing its most significant restructuring since founding. Nearly every original co-founder has departed. Grok — the flagship model — has fallen behind Claude, GPT-4o, and Gemini in independent evaluations. In March 2026, Musk acknowledged the AI coding initiative was "not built right the first time" and brought in executives directly from Cursor, the rival AI coding tool, to restart the effort. This is happening against a backdrop of $30 billion in total capital raised, a 300-megawatt Memphis supercomputer called Colossus, and a company that merged with X in April 2025 — yet still cannot close the product gap with its better-capitalized rivals. The story of xAI is, increasingly, a story about the limits of capital and supercomputing when the people who built the thing keep walking out the door.
When xAI incorporated in March 2023, Elon Musk assembled a founding team that read like a greatest-hits compilation of frontier AI research. The roster included researchers from DeepMind, OpenAI, Google Brain, and Tesla's Autopilot team — people with publication records at the top venues in machine learning and direct experience building the systems that became GPT-4 and AlphaCode.
The original xAI co-founding team included Igor Babuschkin, a DeepMind veteran who had led significant work on large language models and code generation; Tony Wu, who came from Google DeepMind; Jimmy Ba, a University of Toronto professor known for the Adam optimizer paper cited in nearly every modern neural network training run; Christian Szegedy, a Google research scientist behind the Inception neural network architecture; Yuhuai (Tony) Wu, a researcher who had contributed to reasoning models; Zihang Dai, known for Transformer-XL work from Carnegie Mellon; Guodong Zhang and Greg Yang, both with strong theoretical backgrounds in deep learning; Kyle Kosic; Jansen Wong; and several others brought in during the first year to round out the technical leadership.
By early 2026, the majority of those original co-founders had left xAI.
The departures unfolded in waves rather than all at once, which is why the full picture took months to become clear. Some co-founders stayed through Grok 1 and Grok 2, departing quietly after those launches without press releases or public statements. Others left following internal disagreements about product direction, model architecture choices, or the pace of decisions. The consistent thread in reporting from multiple outlets has been a working environment shaped almost entirely by Musk's personal priorities — which shifted frequently between Grok's LLM capabilities, the AI coding product, integration with X's social graph, and the political dimensions of the company's output.
Igor Babuschkin — perhaps the most visible technical co-founder — was still active enough to respond publicly to benchmark criticism as late as February 2025, defending Grok 3's evaluation methodology when an OpenAI researcher accused xAI of publishing misleading results. The fact that xAI's most prominent technical co-founder was spending time in public disputes about benchmark integrity rather than building product said something about where the company's energy was going. His subsequent departure removed one of the few people at xAI with the institutional credibility to push back on Musk's product decisions from a position of technical authority.
The attrition is not unique to xAI — nearly every major AI lab has experienced significant turnover as the talent market heats up and researchers get competing offers from well-funded rivals. What distinguishes xAI's situation is the rate and seniority of departures. Losing a few early researchers is normal. Losing the majority of co-founders within three years is a structural signal about something more fundamental.
Grok launched in November 2023, initially available only to X Premium subscribers, positioned explicitly as a more irreverent, "anti-woke" alternative to ChatGPT and Claude. The framing was part product, part political statement — and it priced xAI into a specific corner of the market that made it hard to compete purely on capability.
The benchmark controversy that Babuschkin responded to in February 2025 illustrated the credibility problem Grok has carried since Grok 2. When xAI published Grok 3's results, the claims were strong enough that an OpenAI researcher publicly questioned the methodology. The actual performance gap — while not as dramatic as critics suggested — reflected a real dynamic: Grok 3 was genuinely competitive in some domains but consistently trailed Claude 3.5 Sonnet and GPT-4o in the tasks developers and enterprises actually care about, particularly coding, multi-step reasoning, and instruction following.
Grok 4, launched in July 2025 at $300 per month for the premium "Heavy" tier, represented xAI's most aggressive push yet on the capability frontier. The price point — higher than any comparable tier from Anthropic, OpenAI, or Google — positioned Grok 4 as a power-user tool rather than a mass market product. Independent evaluations of Grok 4 showed genuine improvements in reasoning tasks, but the model still underperformed Claude 3.7 and GPT-5 series models on the coding benchmarks that had become the primary competitive battleground for enterprise customers by late 2025.
The $300/month price tag without market-leading benchmark performance created an awkward position: xAI was charging premium prices for a product that independent users consistently ranked below the alternatives it was priced against. That disconnect between pricing ambition and benchmark reality contributed to the perception that Grok's development trajectory was being driven by revenue targets rather than product quality.
The safety dimension added another layer of reputational drag. In early 2026, TechCrunch asked directly whether safety was "dead" at xAI, a question prompted by a series of incidents in which Grok generated content that would have been blocked by any other major lab's guardrails — including racist outputs, political bias that alternated between directions apparently based on model updates, and briefly, in February 2025, what appeared to be content filtering specifically applied to unflattering references to Trump and Musk. That particular incident generated a separate TechCrunch investigation that Grok 3 had "briefly censored unflattering mentions of Trump and Musk," a story that damaged Grok's credibility with the developer audience xAI most needed to win.
In March 2026, TechCrunch reported that xAI had hired two senior executives directly from Cursor — the AI-powered coding tool built by Anysphere that has become one of the most widely used developer productivity tools since its 2023 launch — to lead a revamped push on xAI's own AI coding product.
The move is notable for several reasons.
First, it is an acknowledgment of failure. The headline of the TechCrunch piece quoted an internal assessment that xAI's coding product had not been "built right the first time." That kind of admission, delivered via a leadership change and talent acquisition from a direct competitor, is unusual for any company and especially unusual for one led by someone who treats public admissions of weakness as strategic liability.
Second, the specific choice of Cursor as a talent source is revealing. Cursor is not the largest AI company in the world. It is not a household name outside developer circles. What it is, is the best-regarded AI coding tool on the market — a product that developers have chosen over GitHub Copilot and over Grok's own coding features based purely on quality and workflow integration. By recruiting from Cursor, xAI is implicitly conceding that Cursor's engineers have figured out something about AI-assisted coding that xAI has not.
Third, the pattern repeats. This is not the first time xAI has "started over" on its coding initiative. The TechCrunch headline specifically noted it was starting over "again, again" — signaling that a previous restart had already occurred without producing a competitive product. The Cursor talent hire represents at least the second significant reset of xAI's coding product efforts, each reset consuming engineering time, management attention, and credibility.
The irony is that Grok theoretically has advantages Cursor does not: a massive compute infrastructure, a platform integration with X's social graph and Tesla's vehicle ecosystem, and essentially unlimited capital. That the company with those structural advantages is recruiting away from a smaller, leaner competitor suggests the problem is not resources but execution.
The working conditions at xAI have been described in reporting from multiple outlets as extreme even by Silicon Valley's notoriously high standards. Reports of 100-hour work weeks, constant reprioritization of engineering roadmaps based on Musk's shifting interests, and an absence of the institutional processes that most mature engineering organizations use to maintain continuity across leadership transitions have all emerged from people who worked there.
The dynamic is not surprising to anyone who has followed Musk's management approach across his other companies. At Tesla during the Model 3 production ramp, at SpaceX during Dragon capsule development, and at X during the 2022 post-acquisition restructuring, Musk has consistently operated on a principle of maximum urgency applied simultaneously to everything — which creates an environment where nothing gets the sustained, focused attention required to build production-quality AI systems.
Building frontier AI models is different from manufacturing cars or launching rockets in one critical way: the output is almost entirely a function of the people doing the work and their ability to collaborate on complex, multi-month research programs. You cannot accelerate language model training by adding more shifts. You cannot fix alignment problems by staying in the office longer. The research that produces genuine capability improvements requires time, iteration, and institutional knowledge that lives in the heads of the researchers — which means when researchers leave, they take the most important assets with them.
The safety concerns that surfaced in early 2026 are partly a symptom of this dynamic. When senior researchers with safety expertise and institutional authority depart, the systems they built degrade over time as the people who understood the design choices are no longer present to maintain them. The Grok outputs that generated negative coverage in 2025 and 2026 are consistent with what happens when a model's safety fine-tuning was implemented by people who have since left the company.
The Colossus data center in Memphis, Tennessee is one of the most significant pieces of AI infrastructure built in the United States. xAI began construction in 2024 and by late 2025 had acquired a one-million-square-foot property in Memphis to support the facility's expansion. The data center draws an estimated 300 megawatts of power — a figure derived from xAI's own announcement that a proposed 30-megawatt solar farm adjacent to the facility would supply roughly 10% of its energy needs.
The scale of Colossus is genuinely impressive. At 300 megawatts, it is among the largest purpose-built AI training facilities in the world. xAI has also secured permits for 15 natural gas generators on the Memphis campus to supplement grid power, a decision that drew environmental scrutiny given Memphis's air quality challenges.
The disconnect is between what Colossus represents as an infrastructure investment and what it has produced as model output. Competitors have achieved state-of-the-art results with training runs that consumed a fraction of the compute xAI has available. The gap suggests that raw infrastructure, while necessary, is not the binding constraint on AI progress — which has consistently been the people who know how to use it.
xAI has announced plans to expand Colossus further and to build an adjacent solar farm to reduce operational carbon footprint. The environmental debate around AI data centers has intensified as facilities like Colossus have become symbols of the sector's energy appetite, and xAI has faced specific criticism for the Memphis facility's impact on local air quality.
The facility's existence does create one genuine advantage: for any serious AI research organization, access to compute at that scale is a prerequisite for competing on frontier models. The question is whether xAI can staff the teams capable of using it effectively.
xAI has raised approximately $30 billion in external capital across multiple rounds:
The January 2026 Series E gave xAI one of the highest valuations of any private AI company globally. Nvidia's participation was particularly notable — it represents a bet by the world's dominant AI chip manufacturer that xAI will be a significant customer for future GPU generations, a signal as much about hardware sales strategy as AI model quality.
The total capital raised puts xAI in the company of OpenAI and Anthropic as the most expensively funded AI labs in existence. The difference is that OpenAI and Anthropic have produced models that consistently top independent evaluations in the tasks enterprise customers pay for. xAI has produced Grok — a model that is genuinely capable, genuinely improving, and genuinely behind.
Valuation in the AI sector has increasingly decoupled from current revenue and attached instead to the perceived potential of infrastructure assets, model capabilities, and distribution advantages. xAI has strong arguments on infrastructure (Colossus) and distribution (X's 500+ million users, Tesla vehicle integration). The model quality argument is harder to make, which is partly why the Cursor executive hires and the admission of a coding product restart landed with such impact in the press.
The $30 billion in capital raised has produced a company with world-class compute, an enormous potential distribution channel, and a model that users consistently rank below its two main competitors. Investors are betting the talent crisis is temporary and the infrastructure advantage is durable. That bet may be correct. But the departure of the majority of the founding technical team is the kind of signal that makes institutional investors quietly revise their probability estimates.
In April 2025, xAI and X formally merged into a combined entity. The merger gave X shareholders equity in xAI and gave xAI shareholders ownership of X's social platform, advertising business, and data assets. Anthony Armstrong, a former Morgan Stanley banker who had previously advised Musk during the 2022 Twitter acquisition, was hired as CFO of the combined entity in October 2025.
The strategic logic of the merger was clear: X's user base and data firehose are genuinely valuable assets for training and distributing conversational AI. A company building large language models that also owns the platform where hundreds of millions of people communicate has structural advantages in data collection and model feedback that pure-play AI labs cannot easily replicate.
The merger has not, however, solved xAI's fundamental challenge, which is that data and distribution advantages compound slowly while capability gaps are visible immediately. Enterprise customers evaluating Grok against Claude or GPT-4o do not give credit for X's user data — they run the models on their tasks and pick the better performer. Until Grok closes the capability gap, the merger's value remains theoretical.
The X integration has also created complications that a pure AI lab would not face. Grok's outputs are visible to X's entire user base and subject to political pressures that don't affect models deployed primarily through enterprise APIs. The incidents in which Grok appeared to filter content selectively based on political considerations — whether real or perceived — damaged the model's reputation in the developer community precisely because Grok is deployed in a political environment that Claude and GPT-4o are not.
The March 2026 revelation that xAI was restarting its AI coding product — again — deserves more attention than it has received as a case study in organizational failure.
AI coding tools became one of the most commercially significant product categories in the AI sector between 2023 and 2026. GitHub Copilot crossed one million paid subscribers. Cursor grew from a small startup to a product with millions of active daily users and a reputation for being the best-integrated AI coding experience available. The market was large, growing fast, and technically accessible — building an AI coding tool requires capabilities that a frontier model lab should have by definition.
xAI had Grok, a frontier model. It had Colossus, compute infrastructure that dwarfs what Anysphere used to build Cursor. It had the distribution potential of X. It could not build a competitive coding product.
The first restart presumably produced something. The second restart, prompted by the Cursor executive hires in March 2026, has not yet produced public results. The pattern — attempt, fail, restart, bring in outside talent, attempt again — is consistent with an organization whose internal execution problems are structural rather than incidental.
The Cursor hires may help. Bringing in people who have already solved the problem once, in a leaner and less resourced environment, provides the institutional knowledge that internal restarts cannot generate. But those hires are also stepping into an organization with unresolved management dynamics, depleted technical co-founder depth, and a reputation with developers that the coding tool failures have already damaged.
xAI's talent crisis has implications well beyond the company itself.
The most immediate effect is on Cursor. Losing senior executives to a well-funded competitor — even a competitor that has publicly struggled — is disruptive. Cursor's strength has always been product quality driven by a small, focused team. The departure of experienced leaders creates execution risk at exactly the moment when larger competitors are trying hardest to close the gap.
More broadly, xAI's co-founder attrition is reshaping how researchers evaluate opportunities at well-funded AI labs. The pattern that has emerged — join a heavily funded lab with impressive infrastructure, encounter management dynamics that make sustained research difficult, and eventually leave to join a competitor or start something new — is now a known risk factor that top researchers actively consider. The AI talent market, already extremely tight, is becoming more discerning about the difference between capital raised and working conditions offered.
The companies that have successfully retained their founding research teams — Anthropic being the clearest example, where Dario Amodei, Daniela Amodei, and most of the original OpenAI researcher exodus have stayed together — have consistently outperformed on model quality. The correlation between team stability and model performance is not coincidental. It reflects the fact that frontier AI development is a cumulative, collaborative, institutionally-dependent process in ways that other engineering disciplines are not.
For investors writing checks into the AI sector at the valuations that prevailed in early 2026, the lesson from xAI is that compute and capital are necessary but not sufficient. The $30 billion raised by xAI represents a real and durable infrastructure bet. Whether that bet pays returns depends almost entirely on whether Musk can attract and retain the research talent that left — or whether the Cursor hires and whatever recruiting follows can seed a new generation of technical leadership capable of actually closing the gap with Anthropic and OpenAI.
That is a talent problem. And in early 2026, talent problems are the hardest problems in AI to solve with money alone.
How much has xAI raised in total? As of January 2026, xAI has raised approximately $30 billion across multiple rounds, including a $6 billion Series B in 2024, $10 billion in debt and equity in July 2025, and a $20 billion Series E in January 2026. Nvidia participated in the Series E round.
What is the Colossus data center? Colossus is xAI's AI training facility in Memphis, Tennessee. It draws an estimated 300 megawatts of power and is one of the largest dedicated AI compute facilities in the United States. xAI acquired a one-million-square-foot property in Memphis in March 2025 to support the facility's expansion.
Who are the Cursor executives xAI hired? TechCrunch reported in March 2026 that two senior executives from Cursor joined xAI to lead a restructured AI coding initiative. Their specific names had not been publicly disclosed at time of reporting.
When did xAI merge with X? xAI and X formally merged in April 2025, creating a combined entity with shared ownership. Anthony Armstrong, a former Morgan Stanley banker, was appointed CFO of the combined entity in October 2025.
What is Grok 4? Grok 4 is xAI's fourth-generation flagship AI model, launched in July 2025. It is available for $300 per month at the premium "Heavy" tier. Independent evaluations found Grok 4 to be a significant improvement over Grok 3 but still trailing Claude 3.7 and GPT-5 series models on coding benchmarks.
Why did xAI co-founders leave? Public reporting and insider accounts consistently point to a demanding work environment, frequent reprioritization of engineering priorities, and the difficulty of doing sustained research in an organization whose product direction shifts frequently based on Musk's priorities. No co-founders have made detailed public statements about their reasons for leaving.
Sources: TechCrunch xAI coverage, including reporting by Tim Fernholz, Maxwell Zeff, Amanda Silberling, Ram Iyer, and Tim De Chant. Funding data from TechCrunch's reporting on individual rounds. Colossus power estimates derived from xAI's solar farm announcement (November 2025).
xAI Grok 4.20 Beta 2 lands with a 4-agent collaboration system, native video generation, 31% fewer hallucinations, and brand safety scoring — the clearest preview yet of Grok 5's architecture.
Sunday Robotics raises $165M Series B at $1.15B valuation, planning to ship consumer humanoid robots for under $25,000 by Thanksgiving 2026 — the first unicorn focused exclusively on household robotics.
A new Morgan Stanley report says 10x compute scaling will produce AI progress that 'shocks' investors in the first half of 2026, while warning of a 9-18 GW power shortfall and massive workforce displacement.