Jim Cramer's latest take on tech investing signals a structural shift in how we should think about the technology stack. According to the CNBC host, semiconductors have replaced software as the market's center of gravity, driven by the infrastructure demands of generative AI.

Nvidia's latest earnings—$1.87 adjusted EPS on $81.62B revenue—underscore the point. The iShares Semiconductor ETF is up roughly 72% year-to-date, while the software-focused ETF has dropped about 12%. For data and ML engineers, this mirrors what we've seen in production: compute is the new bottleneck, not application logic.

Cramer argues that the old SaaS model—recurring revenue, high margins, predictable unit economics—is losing its edge. AI allows teams to build custom solutions that replace expensive enterprise licenses with open-source models and fine-tuned pipelines. The cost of inference is dropping faster than the cost of Salesforce seats.

From an architectural perspective, this makes sense. Traditional software abstracted complexity away from hardware. AI inverts that: the model is the product, and the chip determines latency, throughput, and total cost of ownership. Companies like AMD, Arm, Broadcom, and Intel are now foundational to the data pipeline, not just peripheral suppliers.

Cramer warns that legacy vendors like Salesforce and Adobe aren't disappearing, but their pricing power is eroding. When a team can spin up a custom chatbot using Anthropic's API on Nvidia hardware for a fraction of the cost of a CRM license, the unit economics shift. The hardware layer becomes the profit center.

For engineers, the takeaway is practical: optimize for compute efficiency, not just code elegance. The stack is being rewritten from the silicon up.

Source: CNBC