Video summary

Qualcomm Says Data Center Chips to Produce ‘Billions’ in ’27

Main summary

Key takeaways

News and Commentary

Qualcomm, Micron, and the AI chip + memory supply question

Qualcomm, Micron, and the broader data-center chip complex are moving under a shared theme: surging AI demand, but (so far) insufficient supply expansion to bring prices down.

Chip rebate/newsflow & Qualcomm weakness

  • Qualcomm shares fell by more than ~5%.
  • The company said its data-center chips could generate “billions” in revenue by fiscal year 2027, reflecting confidence in the AI infrastructure opportunity.
  • Despite the upbeat outlook, the stock reaction suggested market skepticism about how quickly new chip entrants can win meaningful share.

Meta adoption of Qualcomm chips (timing + market crowding)

  • Qualcomm disclosed that Meta Platforms agreed to use Qualcomm’s new data-center processor:
    • Dragonfly C1000
    • and future generations
  • The processor is expected to be available in 2028.
  • While this is a positive customer win, the commentary emphasized that the market remains wary because:
    • the data-center chip space is crowded, and
    • major players and customers are increasingly building or announcing alternatives (including OpenAI pursuing its own chips).

Memory “oligopoly” dynamics and why prices may stay high

Bloomberg Intelligence’s Mandeep Singh (global head of technology research) framed memory constraints similarly to oil supply dynamics:

If capacity doesn’t expand, higher memory prices eventually cause demand destruction.

He highlighted a key determinant:

  • Whether Micron and peers increase CapEx to expand capacity.

Micron in focus after the selloff; demand vs. supply

  • After a selloff in the memory group, Micron was singled out as a bellwether stock.
  • The discussion pointed to:
    • ongoing memory shortages, and
    • the view that the industry has not yet increased capacity enough, despite strong demand.

Shift from training to inference boosts memory intensity

An additional analyst noted that AI workload mix changes memory economics:

  • During training, memory needs were linked to assembling large GPU clusters.
  • During inference, memory demands multiply, potentially requiring much more memory per GPU—described in the discussion as a step-up up to “10x.”

This helps explain broader strength in:

  • memory, and
  • related infrastructure (including storage).

Why the recent “correction” may persist

The commentary suggested memory/storage may not revert to typical commodity-style pricing soon, because:

  • buyers face budget and ROI pressure (often described as “token budgets”),
  • CFOs worry about adoption costs if scaling doesn’t deliver the usual price declines.

Qualcomm’s outlook tempered by incumbents’ lead

The takeaway on Qualcomm’s Meta win was cautious:

  • Even with Meta, established suppliers—especially Nvidia and Google TPUs—still have major advantages.
  • The view is that other players may capture only a small share until multiple chip generations prove out at scale—i.e., ramping to incumbents’ levels isn’t easy.

Presenters / contributors

  • Amy (referenced)
  • Mandeep Singh (Bloomberg Intelligence, global head of technology research)
  • Ken (Janus Henderson—identified as Mike Anthopoulos in the discussion)
  • Cameron Christ (Bloomberg journalist/author referenced)
  • Mike Anthopoulos (Janus Henderson—spoken as “Mike Anthopoulos” in the discussion)
  • Cameron Christ and Mike Anthopoulos (as cited commentators)
  • Jon / Mike / Ken (Janus Henderson) (spoken as “Mike Anthopoulos” and “Ken”)

Original video