ANET

Large public investors value Anthropic at $1.5 trillion, below the $2 trillion bankers floated, a sign that hype-driven AI prices are cracking. This does not stop AI spending. It moves money toward profitable suppliers such as Nvidia, Arista Networks and Applied Materials, away from loss-making AI model makers.

Perplexity's $750M revenue run-rate on agentic AI proves inference demand is exploding, but the real constraint isn't chips — it's the interconnect and memory bandwidth that agentic multi-step reasoning chains consume. While Nvidia gets the headline, large-cap US networking and HBM suppliers are the structural bottleneck play. Long high-bandwidth memory and data-center networking names.

US geopolitical pressure to build a US-aligned AI bloc forces dozens of nations to adopt American AI infrastructure. The obvious trade (long Nvidia, long cloud hyperscalers) is priced. The non-obvious trade is long mid-cap US networking and data center infrastructure suppliers who will fulfill the sovereign procurement wave over 2-4 years.

Kimi K3 joins a wave of capable Chinese open models compressing frontier AI pricing. Market sells US AI names on competition fear. The real beneficiary is inference infrastructure — cloud, networking, and AI platform companies that profit MORE when model proliferation drives usage volume up.

Enterprises are switching to cheaper Chinese AI models, spooking US AI software names. But cheaper inference lowers the cost of AI adoption, expanding total workloads and aggregate compute demand. The infrastructure layer — hyperscaler capex recipients, networking, and power — benefits from volume regardless of model nationality. Long US semiconductor and data center infrastructure into the demand expansion.

China closing the AI model gap triggers broad US tech selling (XLK -5.8% this week). But model competition doesn't reduce hyperscaler capex — it accelerates it. US AI networking and interconnect suppliers are being sold alongside model-layer names despite having no direct China model exposure.

Anthropic's pre-IPO positioning signals a wave of AI pure-plays going public, giving them war chests to challenge hyperscaler AI dominance. This forces defensive capex acceleration at Microsoft, Google, and Amazon. Long the infrastructure layer — networking and custom silicon — that benefits from all sides spending simultaneously.

Snowflake's earnings beat plus a $6B AWS expansion commitment confirms AI-driven data platform spend is accelerating, not plateauing. The 40% move is already priced in SNOW. The real trade is long the picks-and-shovels layer — storage semiconductors, network switching, and data pipeline middleware — that must scale to support this workload surge. Risk-on market (S&P +5.3% 1m, VIX 16.3) supports the setup.

Options market is pricing extreme uncertainty into NVDA earnings. Whether NVDA beats or misses, the elevated vol environment creates a long opportunity in AI infrastructure adjacencies — AMD and networking plays — that benefit from either NVDA's continued dominance (rising AI capex tide) or its stumble (rotation into alternatives). Long the non-NVDA AI infrastructure stack into and through earnings.

Trump-Xi summit surfaces AI chip guardrails as a diplomatic topic, confirming NVIDIA's China exposure is a known risk. The non-obvious trade: as China revenue uncertainty clouds NVIDIA's multiple, US-centric AI infrastructure plays — power producers, networking, and cooling — absorb redirected capex with less geopolitical overhang. Long US AI infrastructure enablers over NVIDIA directly.

Cisco's AI-driven earnings beat confirms the networking infrastructure layer is entering a multi-year capex upcycle. Hyperscaler GPU buildouts are now bottlenecked by bandwidth, not compute. Networking and optical connectivity suppliers — not yet re-rated — are the non-obvious beneficiaries. Long US-listed networking infrastructure and optical interconnect names.

OpenAI missed financial targets, pressuring the entire AI infrastructure complex. The hidden risk is not narrative but mechanical: data center capex was front-loaded on revenue projections that are now slipping. Long-duration AI infrastructure plays — GPU suppliers, networking, and cloud buildout names — face a valuation reset as the monetization gap widens.