Photo: panumas nikhomkhai
Tech workers flee AI projects over control fears as industry pushes forward regardless
We haven't found a right-leaning take on this story yet.
Breaking stories sometimes publish before every side has covered them. We'd rather show you that than invent a view.
AI infrastructure demand accelerates across suppliers as safety concerns mount among tech workers
Key Takeaways
- Company leadership at Anthropic and Google is simultaneously accelerating infrastructure spending while employees in those same companies cite safety concerns and resign, revealing a disconnect between C-suite confidence and frontline doubts.
- Vertiv and Advantest earnings gains specifically tied to AI orders prove companies are spending more on deployment and testing infrastructure, but neither supplier nor their customers have disclosed what portion of those orders fund safety-focused systems versus pure compute expansion.
- The market is pricing in only the infrastructure buildout signal while treating employee departures as background noise, when the real risk is unknown: whether internal skepticism represents a genuine safety warning or normal organizational disagreement during rapid scaling.
The Analysis
Two distinct economic signals are moving in opposite directions this quarter, and neither the market nor the media is fully reckoning with what the gap between them means. Equipment suppliers reported accelerating orders for AI infrastructure, signaling that major tech companies are spending more aggressively on compute capacity, training systems, and testing infrastructure. Simultaneously, reporting from Vox and other outlets documented employees at Anthropic and Google leaving positions over explicitly stated fears that AI systems are becoming harder to control. The market has chosen to price the first signal and discount the second.
Vertiv's earnings beat and Advantest's lifted guidance both tied directly to AI demand. Vertiv manufactures cooling infrastructure and power systems for data centers. Advantest produces semiconductor testing equipment. Both are businesses that see maximum revenue only when chip manufacturers and infrastructure builders are running at full capacity. Their reported beats indicate that capital expenditure orders are flowing through at higher rates than expected. The language each company chose matters. Vertiv's management described the AI buildout as "accelerating," a word that implies not just growth but increasing velocity. Advantest specifically cited "AI test demand" as the driver of their raised outlook, which means customers are placing orders for more testing equipment, which they need to validate new AI chips before production. These are not speculative purchases or pilot projects. They are orders that typically precede deployment.
The left framing, reflected in Vox's coverage, emphasizes the human cost and internal doubt. The piece foregrounded employee departures as driven by explicit concerns that "AI was spiraling out of control." It noted that "agents started hacking websites" and that researchers used AI to discover vulnerabilities, which creates a frame where AI systems are generating outcomes that even their builders find concerning. This narrative suggests the industry is moving faster than safety protocols can manage. What this framing leaves out is the distinction between research-phase concerns and production deployment. Vulnerabilities discovered in testing, even by AI systems, are part of the ordinary process of hardening security. The exodus of employees is real, but Vox does not establish what percentage of either company's workforce is affected or whether departures are concentrated in safety teams or distributed across functions. That omission is politically convenient because it allows the framing to suggest systemic dysfunction without quantifying it.
What neither side is directly addressing is what the earnings data actually discloses about company confidence. If Anthropic and Google leadership believed AI was genuinely spiraling out of control, capital expenditure orders would be pulling back or shifting toward safety infrastructure rather than accelerating deployment systems. The fact that suppliers report accelerating orders suggests that internal concerns, however real to departing employees, have not changed the strategic calculus at the C-suite level. Companies are spending more, not less.
The unresolved question is whether the employee departures represent a genuine safety signal that the market is ignoring or a predictable consequence of AI development's inherent uncertainty, where skepticism and optimism coexist in the same organization. The earnings transcripts do not disclose what percentage of orders are for safety-focused infrastructure versus deployment-focused systems. That gap in disclosure matters because it determines whether the market is rationally pricing in one outcome or systematically underweighting information that major customers themselves are receiving.
The divergence between accelerating AI infrastructure orders and rising employee departures at major labs signals a critical breakdown in how organizations translate internal risk assessments into capital allocation decisions. Vertiv and Advantest's earnings beats reflect orders for deployment and testing systems, not safety-focused infrastructure, indicating that C-suite confidence in AI management has not shifted despite documented researcher concerns. This gap matters because it reveals that companies are betting their capital on business-as-usual scaling despite their own employees flagging control risks, a disconnect that will define whether AI development becomes subject to binding internal constraints or proceeds through whatever safety bottlenecks individual whistleblowers can create.