Wall Street Lunch: OpenAI's JalapeñO Tops Nvidia's Blackwell In Some Inference Tests

OpenAI’s in-house inference chip “Jalapeño,” built with Broadcom, reportedly delivered better performance-per-watt than Nvidia’s Blackwell across multiple SemiAnalysis scenarios, a positive development for AI hardware competition. Offsetting this, Dick’s Sporting Goods shares dropped after Q2 results missed and management issued a cautious full-year outlook, attributing pressure to higher promotional activity in athletic footwear and apparel.

Analysis

The AI-chip readthrough is less about one benchmark and more about hyperscalers proving they will arbitrage away merchant-GPU dependency wherever the economics justify it. That is structurally constructive for AVGO: every incremental custom silicon win creates follow-on demand for high-speed interconnect, packaging, and system-level integration, which tends to be stickier than a single chip design cycle. For NVDA, the near-term risk is multiple compression rather than an earnings shock; the market may start discounting a larger share of AI workloads migrating to bespoke inference silicon even if training still runs on GPUs.

The key second-order effect is competitive bifurcation inside AI hardware: inference-heavy workloads should become more fragmented, while training remains concentrated. That means NVDA’s upside is increasingly tied to software/platform lock-in and networking attach rates, not just raw accelerator share. The falsifier is simple: if large-scale deployment economics still favor NVDA on total cost of ownership after software and integration, this narrative fades quickly; watch 1-2 quarter hyperscaler capex commentary and any delay in custom-chip tapeouts.

DKS looks like a margin-reset, not just a one-off miss. Elevated promotions in footwear/apparel usually spill into the broader sporting-goods channel, pressuring both gross margin and inventory turns across DKS/FL/ASO and, upstream, forcing brands like NKE to defend shelf space with markdown support. The risk is that this becomes a multi-quarter earnings revision cycle if promotional intensity remains high into the holiday build, but the contrarian case is that the stock may already be pricing an extended drawdown if management can normalize inventory faster than feared.

Key Decisions for Investors

• Long AVGO vs short NVDA into the next hyperscaler capex update: use a 1-3 month horizon to express the view that custom silicon and networking content expand faster than merchant GPU share; stop if NVDA re-accelerates networking attach or if AVGO guidance does not show design-win conversion.
• Buy AVGO call spreads on pullbacks rather than outright equity: the upside is tied to multiple expansion from AI-customization optionality, while downside is cushioned if the market treats this as narrative rather than revenue impact over the next quarter.
• Short DKS on rallies or buy put spreads for a 1-2 month catalyst window: the trade works if promotional pressure forces downward EPS revisions and margin guidance resets again; cover if gross margin stabilizes or inventory days improve meaningfully.
• Pair trade DKS short vs long NKE only if channel checks confirm markdown pressure spreading to branded footwear; this captures the second-order winner if retailers eat the promo burden while suppliers preserve mix.
• Set a watch item on quarterly hyperscaler capex and inference deployment commentary: if custom ASIC adoption accelerates without delaying AI spend, rotate further from NVDA into AVGO and ancillary networking names.