I Sold My Warner Bros. Shares and Bought These 2 Growth Stocks Instead. The AI Supercycle Made It Obvious.

The article is highly constructive on Broadcom and AMD, arguing both are positioned for the next AI supercycle. Broadcom is highlighted for networking and custom AI chips, including a projected $100 billion in custom chip sales in fiscal 2027, while AMD is cited as a beneficiary of inference and agentic AI through its MI450 GPU, ROCm platform, and OpenAI/Meta partnerships. This is mostly bullish investment commentary rather than new company-specific news, so the likely market impact is moderate.

Analysis

This read-through is less about ‘AI winners’ and more about where margin shifts from silicon to system-level bottlenecks. If the market keeps rewarding the companies that remove friction in cluster scale-out, the next leg of spend likely accrues to networking, custom ASIC design, memory-heavy inference architectures, and rack integration—not necessarily to the most visible training-GPU suppliers. That implies AVGO and AMD are not just participating in AI capex; they are positioned to capture a larger share of the wallet as customers optimize for total cost per token, not peak FLOPS.

The second-order effect is that hyperscaler capex becomes more concentrated and harder to unwind once platform-specific hardware is embedded. Broadcom’s advantage compounds if custom silicon programs proliferate because switching costs rise after design-in, qualification, and software-stack tuning; the more custom the AI stack gets, the less fungible the supplier base becomes. AMD’s opportunity is more timing-sensitive: inference and agentic workflows tend to favor memory bandwidth, CPU orchestration, and full-rack integration, which could create a multi-quarter re-rating if MI450 adoption shows credible deployment traction.

The market may still be underpricing the supply-chain winners adjacent to the obvious names. GOOGL and META are likely to keep pulling through custom silicon and rack-level demand, but the more interesting trade is that their internal buildouts validate AVGO’s and AMD’s product roadmaps while pressuring smaller networking and legacy server vendors. Conversely, NVDA faces a relative valuation ceiling if investors start capitalizing lower gross-margin but more defensible infrastructure layers elsewhere in the stack.

Main risk: the narrative can outrun installation reality. The next 2-3 quarters matter for proof points on production ramps, software adoption, and customer concentration; any delay in MI450 availability, ROCm stickiness, or Broadcom design-in conversion would compress the premium quickly. A macro capex pause would hurt both names simultaneously, but AMD has the cleaner downside catalyst because expectations around agentic AI and full-rack systems are less proven than Broadcom’s more established networking franchise.

Key Decisions for Investors

• Long AVGO vs short a basket of legacy data-center infrastructure suppliers for 6-12 months; thesis is that cluster scale-out and custom ASIC design-in create higher-duration revenue than generic server/network exposure.
• Initiate a tactical long AMD position into evidence of MI450 ramp milestones over the next 1-2 quarters; use a defined stop if software adoption or customer deployment timelines slip, since the stock is more execution-sensitive than AVGO.
• Pair trade: long AVGO / short NVDA on any AI infrastructure rally over the next 3-6 months; the relative-reward is best if investors begin valuing networking and custom silicon as the higher-visibility monetization layer.
• Overweight GOOGL and META as demand validators rather than headline AI leaders; their capex intensity can remain supportive to AVGO/AMD for several quarters even if ad fundamentals are mixed.
• Avoid chasing near-term upside in AVGO if it gaps on AI enthusiasm; the better entry is on post-print consolidation, because the key risk is multiple compression if custom-chip revenue becomes fully crowded into consensus.