The chip exec who powered the AI boom says America’s $39 trillion debt is the same mistake that kills great companies

Cadence CEO Anirudh Devgan argued AI is being overhyped in the near term, saying software improvements could drive 10x gains in AI computation while warning that human behavior, not technology, is the recurring source of mistakes. He highlighted U.S. national debt as the real structural risk, citing $39 trillion in debt and more than $1 trillion in projected net interest in fiscal 2026, while noting Cadence posted over 14% revenue growth and roughly 45% non-GAAP operating margins in FY2025. He also pointed to a $3 billion Hexagon deal and sees major upside in autonomous transport, defense, robotics, and drug discovery.

Analysis

The market implication is not that AI demand is slowing; it is that capex migration is likely to become more selective. If software-driven efficiency gains continue to compress compute cost curves, the benefit accrues disproportionately to the layer that monetizes faster design cycles and higher utilization, while the most capital-intensive infrastructure names face a greater risk of multiple compression as investors question peak spend intensity. That is structurally supportive for EDA, chip-design enablement, and workflow software versus undifferentiated compute or power-exposed infrastructure.

The second-order winner is anyone selling picks-and-shovels into complexity reduction: faster iteration means more chip tape-outs, more heterogeneous systems, and a higher value per engineer. That should reinforce pricing power for the small set of mission-critical tool vendors, while downstream semiconductor customers may see better product cycles but also tighter ROI scrutiny on every incremental dollar of AI capex. The real loser is not AI itself but the “AI-washing” cohort—companies whose equity story relies on optics of transformation without demonstrable unit economics.

On fiscal policy, the key market takeaway is that debt is becoming a crowding-out mechanism rather than a macro abstraction. Rising interest expense is increasingly a claim on future discretionary spend, which is bad for long-duration public R&D and eventually supportive for private-sector substitution in defense, healthcare, and infrastructure productivity. That creates a medium-term tailwind for firms that can monetize automating what government underfunds: autonomous systems, defense tech, and drug discovery tools.

Contrarian risk: consensus may be underestimating how fast the software stack itself can reduce power and compute intensity, which argues against blanket shorting of AI buildout names. The better expression is a relative-value rotation, not a macro bet against the entire theme. The catalyst window is 3-12 months, as earnings guidance and capex plans separate real adoption from narrative-driven spend.

Key Decisions for Investors

• Long CDNS vs. short a basket of AI infra/power beneficiaries (e.g., SMCI/ETN/CEG depending on mandate) on a 6-12 month horizon: thesis is that software efficiency gains shift value from hardware intensity to design automation; target 15-20% relative outperformance, stop if AI capex guide remains upwardly revised for two consecutive quarters.
• Buy CDNS outright on any 5-8% pullback; the setup is a quality compounder with margin resilience and direct leverage to rising chip complexity. Risk/reward favors owning versus chasing higher-beta AI names because its cash flows are less dependent on peak data-center sentiment.
• Initiate a long AXON / short defense-prime pair over 6-9 months: autonomous and software-defined defense should capture budget reallocation better than legacy platform contractors if procurement starts prioritizing cost-per-effect. Target modest multiple expansion in the long leg and de-rating in the short leg if budget pressure accelerates.
• Consider a small long IBB or XBI call spread funded by short-duration AI hardware beta: if public-sector R&D remains constrained, private drug discovery tools become a relatively more important source of innovation. This is a 12-18 month catalyst with asymmetric upside if compute efficiency lowers discovery cost faster than expected.
• Avoid or reduce longs in names whose bull case is predicated on endless utility/grid capex without clear software leverage; if the market starts to price efficiency gains, these become the most vulnerable to estimate cuts and multiple compression over the next 1-2 quarters.