Are you insane? It was more than a cultural jolt-it was a direction embedded in the engineering DNA of the company. Speaking at an all-hands meeting after another record-breaking earnings quarter, Nvidia Chief Executive Jensen Huang batted away fears that automation would erase jobs, insisting instead that every job which can be automated should be automated. “I promise you, you will have work to do,” he told employees, underscoring that AI fluency is now table stakes for relevance inside the $4 trillion chipmaker.
Huang looked to proof of concept from Nvidia’s own engineering teams. Software developers across the company are leveraging AI coding assistants like Cursor, a context‑aware tool that can query entire codebases, refactor complex modules, and summarize large files. Those capabilities enable engineers to accelerate development cycles while maintaining architectural consistency, critical in GPU‑accelerated environments where software has to fully exploit the parallel processing capabilities. He encourages them to “use it until it does” when AI tools fall short and encourages hands-on involvement in improving the models themselves.
Nvidia’s attitude reflects a wider change in Silicon Valley: Microsoft has made AI usage “no longer optional,” baking GitHub Copilot into engineering workflows. Meta will start tying performance reviews to AI usage, while at Google it has made its Gemini AI requisite for coding and engineers from Amazon have petitioned for integration of Cursor. The tools are redefining software engineering by removing boilerplate and accelerating refactoring and allowing AI-assisted code reviews to go directly into enterprise CI/CD pipelines.
Yet despite the internal push, a set of external headwinds is facing Huang. Michael Burry, the investor famous for “The Big Short,” has likened Nvidia’s role in the AI boom to Cisco’s in the late‑1990s telecom build‑out a cycle that ended in oversupply and collapsed valuations. Patterns of capital expenditure and GPU depreciation schedules are at the heart of Burry’s critique. He says that hyperscaler forecasts of nearly $3 trillion in AI infrastructure spending over three years are based on too-optimistic assumptions about data center power capacity and how long the GPUs will last. Nvidia rebuts that its customers write off GPUs over four to six years, with many citing continued high utilization rates for older models, such as the A100. Strategic investments represent only a small fraction of revenues.
The debate also surrounds the economic dynamics of the AI hardware demand cycles. At an enterprise scale, GPU adoption means not only capital investment in compute but also parallel upgrades in networking, storage, and cooling systems. Each layer has its own refresh rate and ROI calculation, and sustained demand depends on the continued expansion of AI workloads. If deployment stalls, the supply chain-from chip manufacturing through the integration of data centers-faces the same overcapacity risk that haunted the fiber-optic boom.
Huang’s insistence on embedding AI into every workflow is also a hedge against that risk: by making AI indispensable inside Nvidia, he aims at normalizing the technology as a core productivity layer rather than some discretionary tool. In engineering terms, this translates into shifting from pilot projects to production systems where AI agents are woven into version control, automated testing, and deployment pipelines. Tools like Cursor and Copilot are not just coding aids but gateways to GPU-intensive workloads, driving demand through Nvidia’s hardware ecosystem.
The productivity gains Huang contemplates are in line with what was emphasized by industry discussions, like those by Andreessen Horowitz: the biggest benefits surged to experienced engineers who could critically evaluate AI output. These “superhuman” teams bring AI into complex problem-solving with context-aware assistants that compress development timelines without sacrificing quality. For Nvidia, cultivating such expertise internally may prove to be as strategically important as winning external GPU orders.
In practice, the challenge might be less about technical capability than cultural adoption. As Huang acknowledged, some managers remain reluctant to overhaul workflows-a resistance which, if left unchecked, could slow the company’s own transformation. For the leader selling the future of AI across the globe, retraining the habits of 36,000 employees may prove one of the most complex engineering problems on the list.
