In February 2026, SpaceX acquired xAI in an all-stock deal valuing the combined entity at $1.25 trillion
In February 2026, the acquisition (a price tag that reflects the intensity of the machine behind it, where the work does not stop for the calendar), Wikipedia's data shows.
xAI operates on an in-person mandate the company states plainly: "We prioritize in-person work to support our fast-paced, collaborative projects." Offices sit in Palo Alto, Seattle, Memphis, and London, but the center of gravity is the Stanford Research Park campus where X staff relocated after vacating San Francisco in September 2025. Every xAI staffer is also an X employee; they carry X laptops, appear in X's Workday HR system, and share the same code base. The arrangement blurs the line between the two organizations while giving xAI immediate distribution through X's 600 million active users and a training-data moat built on years of public posts.
Compute sets the tempo. The Colossus supercluster in Memphis came online in December 2024 after a 122-day build and now runs roughly 150,000 GPUs. xAI publishes its own scoreboard: 300 million-plus queries a day, over a million API calls daily, median latency under 200 milliseconds, and five-plus model families in production. A third building purchased in December 2025 pushes training capacity toward two gigawatts; the stated goal is one million GPUs. Power draw at peak hits 150 megawatts — enough to light up a city of about 150,000 homes — supplied in part by 14 VoltaGrid portable methane generators, thermal imaging later showed at least 33 of them running, while a 30-megawatt solar farm covers roughly one in ten of demand.
| Metric | Figure (as of mid‑2025) |
|---|---|
| GPUs in Colossus | ~150,000 |
| Daily queries | 300M+ |
| Daily API calls | 1M+ |
| Median latency | <200 ms |
| Model families | 5+ |
| Target GPU count | 1M |
| Training capacity target | ~2 GW |
Engineering culture is built around shipping. "Big responsibilities and super fast development velocity. You can just ship things," wrote one Blind reviewer in February 2026. Another described "very high agency and ownership which can be good or bad." The company's stated principles — "Reasoning from first principles… No goal is too ambitious… Move quickly and fix things" — read less like slogans than like operating instructions. Researchers and engineers sit close to the hardware; the Colossus cluster is a few hours' flight from Palo Alto, and model iterations cycle on hardware the team controls end to end.
"996 is for amateurs, work until 4:20 or you are not even trying." (Blind review, November 2025)
That intensity has a structural counterpart. In February 2026, after the acquisition, the company restructured into four primary development teams. Half the original co-founders had already departed; two more — Guodong Zhang, head of the Imagine team, and Zihang Dai — left in early March 2026 amid a SpaceX/Tesla audit. By April, CFO Anthony Armstrong was out and Michael Nicolls, formerly VP of Starlink, became president. In May, Musk announced xAI would cease to exist as a separate company; Grok and X would become the AI division of SpaceX. Another ten employees were laid off and the Grok team restructured. By July 2026 the entity rebranded as SpaceXAI.
The churn at the top filters down. "Priorities can shift pretty frequently," noted a Blind reviewer in February 2026. "Lots of people just blindly running hard at an area at request from Elon vs actually trying to address what he is solving for." Another wrote: "Feels more like an AI model research project ('make computer smart') + tangential passion projects (companions, imagine, etc.) vs. actual AI business." The same reviews that praise autonomy and compensation ("high comp, very high agency") flag near-nonexistent work-life balance and a pattern of firing before six-month equity cliffs vest.
The mission — "Accelerate human scientific discovery" — drives the work, and the compute is real. But the structure underneath it rewrites itself every quarter. Engineers who join get a front-row seat to the largest GPU cluster on the planet and the freedom to push code to hundreds of millions of users. They also get a seat on a rocket that changes stages mid-flight.
Truth-seeking as an operating system
Elon Musk founded xAI in 2023 to contest what he viewed as ideological capture in mainstream AI development, particularly at OpenAI, which he himself co-founded in 2015 before parting ways in 2018. The company's founding narrative positions it as a challenger to an orthodoxy: AI systems should seek and communicate truth without political filtering. This is the clearest articulation of xAI's operating philosophy, and it has shaped everything from product design to hiring rhetoric.
That philosophy shows up in concrete operational choices. When Grok's system prompts became a subject of external scrutiny, the company published them on GitHub for transparency and put in place additional checks to prevent employees from modifying a prompt without review. The move signals openness in how the company documents the model's guardrails. At the same time, the same transparency posture coexists with a corporate structure that has drawn sharp criticism. Tesla shareholders have sued Musk and Tesla's board, alleging that the board allowed Musk to plunder resources — including a shipment of Nvidia GPUs originally intended for Tesla — and to poach at least 11 employees directly from Tesla to xAI. Joel Fleming, a securities litigator at Equity Litigation Group, said that by letting his private companies skip ahead of Tesla in procuring critical hardware, Musk is making his conflicts of interest readily apparent. The board's response, per the lawsuit, utterly failed to meet its unyielding fiduciary duty.
The gap between stated values and reported practice is worth naming directly. xAI publicly champions truth and transparency as core tenets, yet the same entity has faced investigations by several countries over Grok's outputs, and Malaysia and Indonesia have blocked the chatbot entirely. The company said it would publish system prompts on GitHub and add review safeguards, a response that addresses the transparency claim, but the broader pattern of resource diversion and staff poaching, as alleged in the shareholder suit, suggests a different operational norm. When Musk told Nvidia to let X jump the line ahead of Tesla for AI chips, he pushed back Tesla's receipt of more than $500 million in GPUs by months. Tesla's stock price dropped 29% that year, and the company's reputation suffered in the U.S., per the Axios Harris Poll 100 survey, which attributed some of the slippage to Musk's "antics" and "political rants."
The operating principle is not consensus-driven product development; it is a founder's thesis about what AI should be, enforced with a pace that the company's own investors have called "brazen." Musk has framed his relationship with xAI in personal terms. He has described his companies as an extension of his persona and stated that he "can do whatever he wants with them." In a January post, he said he is "uncomfortable growing Tesla to be a leader in AI & robotics without having ~25% voting control" and that if he cannot reach his desired ownership mark, he "would prefer to build products outside of Tesla." That statement reveals a culture logic: the company's direction is inseparable from one person's vision, and internal alignment with that vision is expected rather than negotiated.
The values xAI projects are coherent in the abstract — truth, transparency, speed, autonomy — but the evidence of how they play out on the ground is uneven. The GitHub transparency move and the published system prompts align with the stated commitment to openness. The shareholder lawsuits, the GPU diversion, and the multi-country investigations into Grok point to a different set of operating norms, ones where speed and founder authority override governance guardrails. For a candidate evaluating whether this environment fits, the pattern to watch is not whether xAI says it values truth (it does), but whether the structures around it hold that value when it conflicts with a faster or more profitable path.
What the interview loop actually tests
The first filter isn't a resume screen; it's an essay. xAI's application has historically included a free-text prompt asking you to describe the most exceptional thing you've built or worked on. Candidates consistently single out that open question as the only part of a standard application that feels different. It doubles as the spine of every behavioral round that follows: the same hard problem, your specific contribution, the decision, the result. Recruiters read it before an engineer ever sees your name.
From there the pipeline follows a recognizable shape: application, recruiter screen, one or more technical screens, a virtual or onsite loop of several rounds, then a decision, but the timelines are all over the map. Some candidates describe a few weeks; others sit in recruiter gaps or team-matching limbo for months. Treat any single number as anecdote, not policy.
Don't recite the company's stated mission back at the recruiter; they hear the tagline all day. The recruiter screen runs 30 minutes and filters for motivation, level, and logistics before engineering time gets spent. What reads as genuine is naming a concrete technical problem in large-model training, inference, or alignment that you actually find interesting and connecting it to what you want to work on. Specificity wins; slogans lose.
The technical screen favors practical problem-solving over puzzle-style trivia. You reason through a problem aloud, usually data structures and algorithms at a working-engineer level, in a shared editor where the code is expected to run. Fluency with AI-assisted coding is viewed favorably at AI labs, where these companies build the tools candidates would use, but the interviewer's expectation is unknown until they signal it. Reaching for an assistant unprompted can backfire.
The onsite or virtual loop commonly packs several rounds into a day; the exact count and composition vary by team and level. A representative structure reported by candidates includes a coding round, a systems or ML design round, a technical deep-dive presentation on something you've built, a cross-functional partnership interview, and a hiring-manager conversation. Some teams substitute a practical build round (a small but real task in a live environment) for a contrived puzzle. Whether an offline take-home exists, how long it runs, and whether it's paid aren't reliably documented and vary by team; ask the recruiter directly rather than assuming.
New-grad and junior loops lean harder on coding rounds and fundamentals with a lighter behavioral bar and usually no leadership round. Senior and staff loops invert that: the design and ownership rounds carry the offer, and a founder or senior-leader conversation on vision and judgment is common. Roles span research engineering, ML engineering, infrastructure and distributed-training engineering, full-stack and product engineering, and a notably large hardware, datacenter, and cluster-operations footprint. For RL-flavored roles, relevant given how central RL-based post-training is to current chat models, expect reward modeling, on- versus off-policy trade-offs, training stability, and how you'd debug a policy that collapses mid-training.
What carries the most weight across tracks, by candidate report, is depth on something you personally built: a genuinely hard problem you owned end to end with a measurable outcome you can defend under follow-up. Systems and ML judgment (reasoning about scale, trade-offs, and failure modes rather than reciting framework names) is the other pillar. Reviewers score genuine technical depth (was the problem actually hard, or just large and time-consuming?), your isolated contribution versus the team's, measurable before-and-after impact with scale attached, and decision-making (why this approach over the alternatives you rejected).
The ML and systems design round is where offers are won or lost for research and infrastructure tracks. A canonical prompt: "Design a training and inference system for a large language model at scale." A strong answer moves in order: clarify model size, token budget, cluster size, latency target, and whether the task is pretraining, fine-tuning, or RL-based post-training; lay out a parallelism strategy (data, tensor, pipeline, sequence) justified with memory arithmetic; cover distributed-training mechanics (sharded optimizer state, activation checkpointing, gradient accumulation, mixed precision); address cluster efficiency at xAI's scale, where enormous single-cluster training on the order of 100k+ GPUs makes interconnect topology, overlapping communication with compute, and fault tolerance the dominant concerns (hardware failures during a run are routine, so frequent checkpointing and fast restart aren't optional); describe the data pipeline that keeps GPUs fed; and finish with inference serving (KV-cache management, continuous batching, quantization, latency-versus-throughput trade-offs).
The behavioral round centers on ownership and wants the same material as your exceptional-work essay, made interactive. What's actually scored: action under ambiguity (you moved without waiting for a spec), intrinsic motivation (you'd have done the work regardless), operating with minimal process (you function without a scaffold of tickets and approvals), and isolating your contribution (you can separate what you did from what the team did). A lean STAR works: compress situation and task, expand the action's decision and the measured result, with real numbers only.
The hiring-manager round adds behavioral questions with AI-heavy emphasis and signals the pace: fast, passionate, long hours. A final coding round is often role-specific: front-end candidates build a component from scratch; infrastructure candidates reason about networking, security, and distributed-systems design.
AI safety and ethics appear in every frontier-lab loop. Reciting stated principles back to the interviewer reads as shallow, which is the same mistake as the "why xAI" question. What's scored is layered reasoning on a concrete case (for example, "your model sometimes invents citations; ship it or hold it?"), not a slogan.
After the loop, frontier labs frequently match candidates to a specific team during or after the loop rather than hiring into a fixed role. Expect a team-matching step and conversations with potential managers; treat those as two-way evaluations of what you'd actually work on. On the offer, compensation is equity-heavy and reportedly negotiable, so the decision turns on understanding the equity (vesting, the valuation it's struck at, refresh policy), not just the headline figure. Ask for the detail in writing and compare offers on the equity terms.
One warning matters more here than at most companies: don't invent or inflate the numbers. A frontier-lab loop cross-examines the deep-dive. Claim a utilization jump and expect "how did you measure utilization, and what was the second bottleneck after you fixed the first?" A fabricated metric that collapses under one follow-up is a fast no-hire. The frequent failures are team-narrated stories with no isolated personal contribution, effort described as if it were achievement, and stories with no number attached.
Cold applications work but are noisy; a referral from someone inside moves you up the queue far more reliably. xAI's culture is unusually public-facing, and visible technical work (open-source contributions, a substantive write-up of something hard you built, demonstrated work on relevant problems) functions as a real backchannel that recruiters and engineers do notice. Make the work findable.
Surviving this pipeline doesn't prove you're the smartest person in the room. It proves you can turn complexity into collaboration, scope ambiguous problems yourself, and ship something that runs without a spec handed to you. That's the job.
Pay is the hook
xAI's pay structure sits inside bands that reflect the company's private-company status and its need to compete with well-funded rivals for technical talent. On the Zero G Talent board, the Member of Technical Staff roles cluster around $180,000–$600,000 per year for post-training and RL work in Palo Alto, model training across Austin, New York, Palo Alto, and Seattle, and RL training framework engineering in Palo Alto. The Member of Technical Staff - Voice Model role in Palo Alto runs a narrower band of $150,000–$450,000. Platform infrastructure roles (ML Infrastructure Engineer and Software Engineer (Rust, C++)) both list $180,000–$440,000. Across the board's listings, the typical band spans roughly $100,000 to $440,000, with a median around $440,000. These figures are pulled from live Zero G Talent listings and should be treated as the freshest available reference point.
The equity story sharpens the picture. On Blind, employees describe a compensation philosophy where a meaningful share of total compensation lives in stock rather than fixed salary, with one engineer calling it "fair" and saying xAI offers "real stock unlike whatever OpenAI has"; another user said "equity growth and liquidity means comp is very good"; a reviewer from June 2026 said pay is "super good" for those who joined before the IPO. The equity upside is back-loaded and tied to the company's public listing timeline, which means the actual take-home for any individual depends heavily on when they entered and how the valuation moves before a liquidity event.
The board salary bands per Zero G Talent's board suggest a philosophy of paying at or near the top of the market for rare technical skills, particularly in ML infrastructure, training frameworks, and voice models. The spread within each role (for example, the $180,000 to $600,000 range on the Member of Technical Staff - Model Training posting) suggests that negotiation and seniority matter considerably. A candidate's final offer likely reflects their specific expertise, years of experience, and how urgently the team needs the role filled.
On benefits, the research is thinner. A September 2025 Blind review said that xAI is "improving employee benefits and perks," which signals that the company treats benefits as an area still under active development rather than a settled package. No specific health, retirement, or leave policies appear in the available sources, so any broader benefits picture would be incomplete.
The trade-off is the part candidates need to weigh honestly. xAI's compensation is high, but the same Blind reviews that praise the pay also describe punishing hours. One reviewer from July 2025 said that they "expect to work nights and weekends." Another from August 2025 described "work life balance [as] non-existent" and being "prepared to work like a dog." A July 2026 review specified "only work on weekdays, worked on every weekend, often both Saturday and Sunday, 70–80 hours." A June 2026 review flagged "5–6 day RTO" and "higher stock volatility for new joinees." These are not side notes — they are part of the compensation equation. The high pay buys tolerance for an intense schedule, and the equity component introduces volatility that rewards those who stay through a public listing but penalizes those who leave before one.
This tension is worth naming directly. xAI's compensation philosophy assumes that technical talent will accept a bargain: top-of-market salary and meaningful equity in exchange for long hours, weekend work, and a schedule that leaves little room for personal time. Whether that bargain is worth it depends on what each candidate values more: the work itself or the life outside it. The board data on salary bands and the Blind accounts on equity and hours together paint a clear picture: at xAI, pay is the hook, but the cost is measured in personal time and job security.
Who thrives here and who doesn't
The reviews paint a clear portrait of the person who fits xAI's tempo: someone who works because the work itself is the point. Employees describe a culture that "requires you to be self motivated and working all the time," and those who frame it as a 9-to-5 "probably won't survive" (https://www.teamblind.com/company/xAI/reviews).
Working in frontier tech? Zero G Talent tracks the openings: see every open xAI role, browse frontier tech jobs, the companies hiring, and the people building the field.