Nvidia's Biggest Risk Isn't Custom AI Chips From Broadcom or AMD -- It's Something That's Hidden in Plain Sight

PwC estimates autonomous AI systems could add $15.7T to global economic value by 2030, but the article argues Nvidia’s biggest competitive risk isn’t AMD or Broadcom. It highlights that Nvidia’s hyperscaler customers (key by net sales) are developing cheaper in-house AI chips that could reduce Nvidia’s GPU pricing power, accelerate supply dynamics, and delay data-center upgrade cycles. Despite Nvidia’s continued hardware leadership across multiple GPU generations, the internal-competition risk is framed as the core threat to its AI data-center dominance.

Analysis

The market is likely overfocusing on direct chip-to-chip rivalry and underpricing buyer vertical integration. The bigger mechanism is bargaining power: when hyperscalers internalize more inference/training silicon, they don’t need to win the performance race to pressure merchant suppliers — they only need enough deployment scale to cap pricing, slow replacement cycles, and force more customized, lower-margin SKUs. That is a multiple problem for NVDA before it becomes a unit-volume problem.

AMD and AVGO are less threatened than headline comparisons suggest, but for different reasons. AMD can still serve as a secondary source of GPU capacity when customers want supply optionality, while AVGO benefits if custom ASIC spend rises; both are beneficiaries of wallet reallocation away from the pure-GPU model. The catch is that this is a share shift within AI capex, not a demand supercycle expansion, so the upside is more about relative resilience than a sustained re-acceleration in the entire semiconductor group.

Catalyst-wise, the next 1-3 quarters matter more than the next few days: watch hyperscaler capex commentary, gross margin trajectory, and any indication that AI buildouts are moving from scarcity to optimization. The contrarian risk is that NVDA’s supply advantage has been the real source of its pricing power, and once customers can self-supply enough silicon, the market may start discounting a slower refresh cadence earlier than consensus expects. Falsifier: if NVDA continues to post stable-to-rising margins alongside unchanged backlog commentary, this thesis is premature.

Key Decisions for Investors

• Maintain a tactical bearish bias on NVDA only on strength: use post-rally windows to initiate a 3-6 month relative short versus AVGO, targeting 8-12% underperformance if customer self-supply commentary gains traction; stop if NVDA margins and backlog re-accelerate.
• Long AVGO on a 6-12 month horizon as the cleaner beneficiary of custom-ASIC spending and networking attach; risk/reward is favorable if hyperscalers keep reallocating capex from merchant GPUs to bespoke silicon.
• Treat AMD as a selective hedge, not a core short: it can capture share from buyers seeking second-source optionality, but the trade works only if the market starts valuing supply diversification over raw peak performance.
• Set an alert around the next two earnings cycles for any NVDA guidance tied to margin compression, longer refresh intervals, or softer backlog conversion; those are the fastest falsifiers and the most actionable catalyst points.