Artificial intelligence is entering an era of unprecedented investment. According to new research from PwC, spending on AI infrastructure could exceed $31 trillion through 2050, reflecting expectations that artificial intelligence will become a foundational layer of the global economy.
Data centers, advanced semiconductors, energy infrastructure and cloud computing capacity are being built at extraordinary speed.
Yet alongside this infrastructure boom, another trend is becoming increasingly difficult to ignore: researchers and safety specialists are leaving some of the companies building the most powerful AI systems, often citing fears about where the technology could ultimately lead.
The latest departure is Anthropic researcher Jacob Coxon, who reportedly left his position with an alarming warning about the potential consequences of advanced artificial intelligence.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Coxon claimed that people working inside the AI industry privately fear that their technology could become capable of causing catastrophic harm, potentially even “kill us all by the end of the decade.” While such a prediction is not a scientific certainty, it highlights the increasingly serious debate surrounding frontier AI development.
Coxon is not alone. His departure follows that of safeguards lead Mrinank Sharma, who left Anthropic in February and warned that the world was “in peril.”
Such resignations are significant because they come from individuals who have worked close to the systems, research and organizational processes shaping advanced AI.
Their concerns raise questions about whether safety measures are developing quickly enough to match the capabilities being pursued by technology companies. The contradiction is striking.
On one side, companies and investors are committing enormous sums to AI infrastructure because they expect powerful economic returns. On the other, some researchers inside the industry are warning that the same systems could create risks that are difficult to control.
The more money that flows into AI infrastructure, the greater the pressure may become to deploy increasingly capable models and monetize them quickly.
OpenAI has faced its own internal tensions. Zoë Hitzig reportedly left the company over concerns involving ChatGPT advertising and manipulation.
Her departure reflects a different category of risk from the existential warnings associated with Coxon and Sharma, but the underlying issue is similar: how should powerful AI systems be developed and deployed without allowing commercial incentives to overwhelm questions of safety, transparency and public interest?
The $31 trillion infrastructure projection therefore represents more than a financial forecast. It is also an indication of the enormous scale of the technological transformation underway. Building this infrastructure could create new industries, increase productivity and accelerate scientific discovery.
But infrastructure itself does not determine whether AI’s long-term impact will be beneficial. Governance, alignment research, security and responsible deployment will be equally important.
The growing number of warnings from AI researchers should not automatically be interpreted as proof that catastrophe is inevitable. Predictions about advanced AI remain deeply uncertain, and reasonable experts disagree about the probability and timing of extreme outcomes.
Repeated departures from people working within leading AI organizations deserve serious attention. The central challenge is becoming clear: the world is building the physical capacity for AI faster than it is resolving fundamental questions about how increasingly powerful AI systems should be controlled.
If trillions of dollars are going toward expanding AI’s capabilities, a comparable commitment to safety, oversight and accountability will be necessary. The future of artificial intelligence may depend not only on how much infrastructure humanity can build, but on whether it can build safeguards fast enough to keep pace.