OpenAI Is Building Its Own AI Chip With Broadcom. Should Nvidia Investors Be Worried?

OpenAI and Broadcom launched Jalapeño, OpenAI's first custom AI chip, highlighting a growing custom-silicon challenge to Nvidia. Broadcom's AI semiconductor revenue surged 143% year over year to $10.8 billion in its fiscal Q2, with more than $30 billion of AI orders booked and management reaffirming AI chip revenue above $100 billion in fiscal 2027. The piece argues Nvidia still dominates on scale, software, and flexibility, but custom chips may gradually pressure pricing power and margins.

Analysis

The immediate winner is Broadcom, but not because it steals share from Nvidia overnight; the more important effect is that custom silicon legitimizes a second sourcing layer for hyperscalers and model builders. That tends to reallocate wallet share from merchant GPUs toward co-designed accelerators and networking, while increasing the value of whoever controls the integration layer and the scarce manufacturing/packaging slots. In other words, the supply chain beneficiaries are likely to be the vendors closest to design wins and capacity reservations, not just the chip name on the slide deck.

For Nvidia, the deeper issue is not unit loss from one customer, but margin normalization as inference workloads become more price-sensitive. Inference is the first place custom chips can economically displace general-purpose GPUs because the workload is repetitive, measurable, and highly power-constrained; training remains much stickier to CUDA and Nvidia’s broad platform. That means the erosion should show up first in mix and pricing, then in gross margin, with the market likely to misread the lag as “no impact” until later quarters when renewal cycles hit.

The contrarian view is that the market may be overestimating how fast custom silicon can eat the pie. AI capex is still expanding so rapidly that near-term substitution is mostly additive, and every large model builder pursuing its own ASIC still needs Nvidia for training, experimentation, and spillover demand. The real risk to Nvidia over the next 6-18 months is not a collapse in growth, but sentiment compression if investors conclude peak-margin assumptions are fading even while revenue keeps compounding.

For Meta and Alphabet, custom silicon is an indirect positive because it lowers internal compute costs and gives them more leverage in supplier negotiations; that can translate into better AI unit economics before it shows up in headline margins. Netflix is largely irrelevant here except as a sentiment tell: investors are still treating AI infrastructure as a crowded trade, so any disappointment in Broadcom order conversion or Nvidia guidance could trigger a fast de-rating across the group.

Key Decisions for Investors

• Maintain a relative-value short NVDA / long AVGO pair for 3-6 months: Broadcom has clearer near-term order visibility from custom silicon while Nvidia is more exposed to future pricing pressure; target 8-12% spread capture, stop if Nvidia re-accelerates data center growth above consensus by >10%.
• Use a staggered entry to buy AVGO on any 5-8% pullback over the next 4-8 weeks: the setup is favorable as the market can tolerate a premium multiple when order backlog is converting into multi-year revenue, but position size should reflect execution risk around customer concentration.
• Sell covered calls on NVDA 2-3 quarters out against core holdings: upside is likely capped by valuation compression even if revenue stays strong, while the premium helps monetize the expected sentiment volatility from custom-chip headlines.
• Add GOOGL and META on dips as a secondary beneficiary basket over 3-12 months: both stand to gain from lower internal inference costs and negotiating leverage versus external GPU suppliers; preferred as a lower-beta way to express the custom-silicon trend.
• Avoid chasing headline momentum in the AI infrastructure complex until the next earnings cycle confirms whether custom-chip adoption is broadening or just being pre-bought into current guidance; the key risk to the trade is a re-acceleration in Nvidia’s backlog that invalidates the scarcity narrative.