A Famed Billionaire Tech Investor Just Sold Nvidia Shares to Buy These 2 AI Chip Stocks

The article highlights that Nvidia is “cheap” versus a forward P/E of 17x (fiscal 2028 ending Jan 2028) and frames it as the dominant AI training platform with CUDA and end-to-end rack-scale AI infrastructure. Tiger Global’s Chase Coleman III cut his Nvidia stake by ~7% while adding Cerebras as a new position (12th-largest holding) and increasing Intel to a 15th-largest holding. Cerebras is pitched as an inference-speed leader (up to 6x vs LPUs) but more speculative, while Intel is described as no longer cheap and facing server CPU share pressure from AMD and competition from Arm.

Analysis

The real signal here is not a broad endorsement of AI semis; it is a rotation toward heterogeneous inference architectures. That favors the companies selling the full stack or the lowest-latency niche, while pressuring names whose valuation still assumes they own every dollar of AI compute. NVDA remains the cleanest structural winner, but the next leg is less about unit growth and more about whether mix shifts toward inference can preserve pricing power and gross margin; if hyperscalers keep internalizing more custom silicon, the multiple ceiling tightens over 6-18 months.

INTC looks like the classic flow trap: any AI-CPU uplift is likely cyclical and partially offset by continued share loss and foundry drag, so near-term enthusiasm can mask deteriorating competitive economics. AMD is the best relative beneficiary because it captures share without the same legacy drag, and AMZN benefits as the cloud layer that monetizes inference consumption regardless of which accelerator wins. CBRS is the most convex name, but also the most fragile: if deployments broaden, upside is outsized; if it remains a premium, single-customer story, the market will eventually discount it as a niche solution rather than a platform winner.

Contrarian take: the consensus may be overestimating how quickly inference will commoditize into a pure price war. If latency and throughput stay bottlenecked, NVDA’s software and networking moat can remain durable longer than bears expect, making the recent trimming by a top investor more about portfolio construction than a fundamental top. The key falsifier is a visible slowdown in hyperscaler capex or a sharp step-down in next-quarter data center demand; absent that, selloffs in the leaders are more likely entry points than trend breaks.

Key Decisions for Investors

• Long AMD / short INTC as a 1-3 month relative-value pair: AMD has cleaner share-gain torque, while INTC still carries foundry and legacy-PC overhang. Use strength in INTC to establish; thesis breaks if INTC data center share and gross margin both inflect materially higher next quarter.
• Maintain NVDA as a core long, but add only on 5-8% drawdowns or around any post-earnings pullback via call spreads; upside is driven by sustained AI capex, while downside is a capex pause or margin compression.
• Treat CBRS as a watch item rather than a full-size position unless liquidity and customer concentration are independently confirmed. If accessible, use a small speculative starter only for a 6-18 month convexity bet; exit if deployment news fails to broaden beyond one or two marquee accounts.
• Prefer AMZN as a secondary beneficiary of inference spend over pure-play chip beta; AWS monetizes the usage layer and should be less exposed to pricing compression. Falsify if cloud margins do not improve with inference traffic over the next 2-3 quarters.