Elon Musk has launched another sweeping overhaul of xAI’s engineering ranks as the company is folded deeper into SpaceX ahead of a record-setting IPO, which underscores intense pressure to close the gap with OpenAI and Google while stabilizing a company hit by departures, layoffs, and mounting questions over execution.

According to an internal memo reviewed by Business Insider, SpaceX executive Michael Nicolls, the senior vice president of Starlink, has now assumed the role of xAI president as the AI company is integrated more closely into SpaceX’s structure.

In the memo, Nicolls told staff that xAI is “clearly behind” competitors and that the company is moving urgently to close the gap. On the compute side, he reportedly described xAI’s training performance as “embarrassingly low,” adding that management intends to improve it significantly within the next two months.

For a company central to a potential $1.75 trillion to $2 trillion SpaceX IPO narrative, that internal assessment is a revealing signal of how much execution risk still sits beneath the valuation story.

SpaceX, which earlier this year absorbed xAI in a transaction valuing the combined group at about $1.25 trillion, has already filed confidentially for an initial public offering, according to reports.

That means xAI is no longer simply a startup fighting for relevance in the generative AI race; it has become a key growth pillar inside the investment case for what could be the largest listing ever brought to market.

The urgency behind the overhaul reflects a widening competitive gulf. OpenAI, Anthropic, and Google have moved aggressively in model performance, enterprise adoption, coding tools, multimodal systems, and developer ecosystems. By contrast, xAI’s flagship Grok platform remains under pressure to demonstrate that it can compete not only in consumer-facing chatbot products but also in enterprise-grade coding, reasoning, voice, and multimodal workloads.

This helps explain the scale of the internal reset. The company has reassigned leadership across nearly every core layer of model development.

Devendra Chaplot will now lead pre-training, the foundational phase where models absorb broad statistical patterns from massive datasets. Aman Madaan takes charge of model factory and tooling, overseeing the infrastructure and pipelines that determine how quickly models can be iterated. Aditya Gupta now leads post-training and reinforcement learning, the crucial final stage where models are aligned with human preferences and refined for deployment.

On the product side, former Cursor engineers Andrew Milich and Jason Ginsburg are now leading Grok Main, Grok Voice, and Grok Imagine, the company’s multimedia generation suite.

Once again, SpaceX talent was imported. Daniel Dueri, a senior SpaceX software engineering leader, is now overseeing compute infrastructure, while Matt Monson, Starlink’s software director, has taken over data operations at xAI.

This mirrors a familiar Musk management pattern. At Tesla and SpaceX, Musk has often responded to execution bottlenecks with rapid centralization, flattening reporting structures, and bringing in trusted lieutenants from adjacent companies. Here, he appears to be applying the same playbook to AI, treating xAI less as a standalone lab and more as an engineering division inside a much larger industrial and infrastructure machine.

The reorganization also comes against a backdrop of significant instability. Since January, eight founding engineers have exited, including senior figures such as Ross Nordeen, Guodong Zhang, Manuel Kroiss, and Toby Pohlen. xAI has also reportedly shed dozens of employees since February, including cuts to Grok Imagine, Macrohard, and, more recently, parts of its recruiting function.

That level of churn is unusual for a company approaching a public-market debut through its parent. Executive departures at this pace inevitably raise questions about culture, strategic clarity, and whether the company’s internal trajectory matches the valuation expectations being built into the SpaceX offering.

Musk himself has acknowledged the scale of the rebuild. In March, he wrote on X that “xAI was not built right first time around, so is being rebuilt from the foundations up.”

He later added that “many talented people over the past few years were declined an offer or even an interview @xAI,” signaling that the company is now revisiting earlier candidates as it attempts to replenish its technical bench. Those remarks are notable because they suggest that the restructuring is not merely cosmetic.

This is a foundational rebuild of architecture, talent, and product direction at a time when the economics of AI are increasingly defined by scale. Model quality today depends as much on compute utilization and training efficiency as on research talent. If, as Nicolls wrote, the compute performance is “embarrassingly low,” then the issue directly affects development speed, inference costs, and eventually profit margins.

But it comes with more central bearing for IPO investors. A SpaceX listing will likely be marketed not only on launch services and Starlink’s subscription cash flows, but also on AI-driven future growth. Yet xAI remains a capital-intensive business with heavy burn rates and a still-unproven commercial moat.

That tension makes the overhaul strategically important as Musk is effectively trying to ensure that by the time SpaceX’s prospectus is fully public, xAI looks less like a company in crisis and more like a scalable strategic engine capable of supporting a trillion-dollar valuation.

How quickly that transformation happens is expected to determine how much of the SpaceX IPO premium markets are willing to attribute to AI rather than rockets and satellites.

OpenAI has introduced a new $100-per-month ChatGPT Pro subscription tier, sharply escalating the competition in the fast-growing AI coding assistant market as it moves to capture professional developers and power users who have outgrown standard usage limits.

The new plan, announced Wednesday, is designed primarily around heavier use of Codex, OpenAI’s software engineering assistant, and comes as the company faces mounting competition from Anthropic’s increasingly popular Claude Code.

According to the company, the new tier offers five times more Codex usage than the $20-per-month Plus plan, targeting developers engaged in what it described as “longer, high-effort Codex sessions.”

“The Plus plan will continue to be the best offer at $20 for steady, day-to-day usage of Codex, and the new $100 Pro tier offers a more accessible upgrade path for heavier daily use,” the company wrote in a post on X.

This creates a new midpoint in OpenAI’s consumer pricing stack. Until now, the jump from the $20 Plus plan to the existing $200 Pro tier was substantial, leaving advanced users with limited upgrade flexibility. The new $100 option effectively plugs that pricing gap and broadens the monetization ladder for developer-centric workloads.

That matters because coding assistants have become one of the most commercially significant segments in generative AI. The coding assistant market is no longer an experimental category. It is rapidly becoming one of the most fiercely contested battlegrounds in enterprise and consumer AI.

Codex can automate code generation, debugging, test execution, feature building, and bug fixes, dramatically compressing software development cycles. Since its rollout, adoption has accelerated at a pace that suggests AI-assisted coding is moving into mainstream developer workflows.

OpenAI chief executive Sam Altman said this week that Codex has reached three million weekly users, underscoring the scale of demand.

The growth trajectory has been particularly indicative of its accelerating adoption. Codex’s annualized revenue run rate surpassed $2.5 billion in February, representing growth of more than 100% since the start of 2026, according to a CNBC report. That kind of expansion signals that coding agents are quickly becoming a major revenue engine for AI firms.

This is precisely where the competitive pressure from Anthropic becomes most visible. Anthropic’s Claude Code has emerged as one of the strongest rivals in the AI development tools space, particularly among engineers who value longer context windows, repository-scale reasoning, and sustained coding sessions. Its subscription structure already includes $100 and $200 premium tiers, branded as Max 5x and Max 20x, which offer elevated usage limits for heavy development workloads.

OpenAI’s move mirrors that framework almost directly. The AI coding market is beginning to resemble the early cloud-computing era, where vendors compete not only on model quality but on compute quotas, workflow integration, and pricing flexibility.

In effect, usage limits are becoming a commercial weapon. The introduction of the $100 tier suggests OpenAI is responding to a growing class of users whose consumption patterns fall between casual usage and enterprise-grade heavy deployment. The $20 Plus tier remains positioned for routine coding assistance and everyday prompts. The $100 plan is aimed at serious developers running extended debugging sessions, codebase-wide refactors, and parallel task flows. The $200 plan remains the premium option for the heaviest users.

This tiering strategy is also indicative of the economics of inference costs. Coding assistants tend to be computationally expensive because they require sustained reasoning over long contexts, multiple files, testing environments, and iterative revisions.

Unlike simple chatbot interactions, software engineering tasks often involve multi-step execution loops that can run for minutes or longer. OpenAI itself has highlighted that Codex tasks can take anywhere from one minute to 30 minutes, depending on complexity.

That makes granular pricing almost inevitable. The company has also been expanding Codex beyond subscriptions. Last week, OpenAI introduced pay-as-you-go Codex-only seats for ChatGPT Business and Enterprise customers, moving toward token-based pricing for teams and organizations.

This signals a broader shift in business model. Rather than relying solely on fixed subscription fees, OpenAI is increasingly aligning pricing with actual compute usage, similar to cloud infrastructure providers such as AWS and Microsoft Azure.

The logic is to monetize AI as infrastructure, which will likely intensify the competition with Anthropic.

Claude Code has built strong traction among developers who value deep code comprehension and long-form reasoning. OpenAI, meanwhile, is leveraging ChatGPT’s much larger installed user base and ecosystem familiarity.

The result is an emerging two-horse race in AI-assisted software engineering. What began as a feature inside chatbots is evolving into a distinct software layer for coding productivity.

However, AI coding tools are moving beyond autocomplete and into agentic software engineering, where systems can independently interpret technical documents, write code, run tests and suggest pull requests. This shifts AI from a passive assistant into an active participant in the development lifecycle.

For OpenAI, the new $100 tier is a direct response to rising demand, rising compute intensity, and rising pressure from Anthropic. In short, the battle for the future of AI-assisted software development is increasingly being fought through quotas, workflow depth, and developer retention rather than headline model launches alone.