OpenAI Unveils First Custom AI Chip With Broadcom | Bloomberg Tech 6/24/2026

Bloomberg highlights OpenAI's first custom AI chip, Jalapeno, developed with Broadcom, underscoring continued investment in AI infrastructure and chip design. The segment also notes SK Hynix’s planned $29 billion US listing and Cerebras CEO Andrew Feldman commenting on the company’s first quarterly earnings since going public. Overall tone is factual and mildly positive for AI/semiconductor and IPO-related names.

Analysis

AVGO is the obvious near-term beneficiary, but the more important read-through is that custom silicon is moving from strategic optionality to an execution mandate for frontier model owners. If OpenAI is willing to co-develop an in-house chip path, hyperscale buyers will use that as leverage in pricing negotiations with merchant AI accelerators, which can compress attach rates and gross margins across the broader AI supply chain over the next 2-4 quarters.

The second-order winner is the ecosystem that can monetize design wins without depending on one architecture cycle. Broadcom has the best shot because it is not just selling a chip; it is selling integration, packaging, and a serviceable supply chain. The losers are more likely to be vendors exposed to a single-node, single-customer narrative: any AI compute name priced for perpetual GPU scarcity is vulnerable if even a modest share of training spend shifts to custom silicon over the next 12-24 months.

The SK Hynix listing matters because it reinforces that memory capacity remains the hidden bottleneck in AI infrastructure, and capital markets are willing to fund it at scale. That usually translates into a capex reflex that benefits tool vendors and advanced packaging suppliers first, but it also raises the probability of a mid-cycle supply response that eventually normalizes pricing. In other words, the bullish part is near-term earnings leverage; the bearish part is that the market may be underestimating how quickly memory and interconnect supply can catch up once financing is abundant.

Contrarian setup: the consensus may be overpricing the durability of merchant AI accelerator margins while underpricing the winner-take-more economics for infrastructure enablers with custom-design capability. The best risk/reward is likely not a blanket long semis basket, but a relative-value long in companies that profit from AI adoption regardless of chip architecture versus shorts in names most exposed to a single dominant GPU spend path.

Key Decisions for Investors

• Long AVGO on a 3-6 month horizon; add on any post-news pullback, with a favorable risk/reward if the market continues to re-rate custom AI design capability as a recurring revenue stream rather than a one-off win.
• Pair trade: long AVGO / short a high-multiple AI accelerator name most dependent on merchant GPU share over the next 6-12 months; thesis is margin compression from custom silicon adoption.
• Stay long the AI infrastructure complex on any weakness in memory/tooling names tied to capex expansion, but prefer suppliers with diversified customer bases; this captures the spending cycle while reducing single-customer risk.
• Avoid chasing the broad AI basket at current levels; use call spreads instead of outright longs to express upside in AVGO, since the catalyst is real but the valuation support is likely to be uneven over the next 1-2 quarters.
• Set a 6-12 month watchpoint for evidence of custom-chip volume ramp; if design wins broaden beyond one flagship model, rotate exposure away from pure-play accelerator beta and toward integration, packaging, and memory bottleneck beneficiaries.