SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Qualcomm Moderated Thread - please read rules before posting
QCOM 181.03-3.5%Oct 28 9:30 AM EDT
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: blinkered who wrote (196275)10/28/2025 5:35:59 PM
From: Jim Mullens1 Recommendation

Recommended By
John Hayman

  Read Replies (1) of 196287
 
Blinkered, re: how to adjust for 'output'? i.e. TOPS or TB

I believe this is what you’re looking for

15\4) 8:34 AM PDT

Corrected Comparison: Cost per Token Throughput

Vendor

 Platform

 Racks Needed for 2B tokens/sec

 Total Cap Ex

 3-Year TCO

 Cost per 1M Tokens
QCOM

 AI250

 1,000

 $332M

 $500M

 $0.0038
AMD

 MI300X

 ~1,667

 $505M

 $1.03B

 $0.0054
Nvidia

 Rubin

 ~1,200

 $889M

 $1.52B

 $0.0063


>>>>>>>>>>>>>>>>>>>>>>>>>>

10) 7:47 AM PDT — Cost per 1M Tokens: Qualcomm vs. Nvidia vs. AMD (Efficiency Comparison)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Here’s the updated table showing Qualcomm’s AI200 and AI250 rack solutions compared to Nvidia and AMD, with relative efficiency gains (%) based on cost per 1M tokens inferred.



?? Cost per 1M Tokens — Comparative Table with Efficiency Delta

Vendor

 Platform

 Cost per 1M Tokens

 Efficiency vs. QCOM AI250
Qualcomm

 AI250

 $0.0038

 — (baseline)
Qualcomm

 AI200

 $0.0056

 32% less efficient
Nvidia

 Rubin CPX (NVL144)

 $0.0065 (avg)

 42% less efficient
Nvidia

 GH200 Grace Hopper

 $0.0080 (avg)

 52% less efficient
AMD

 MI300X

 $0.0060 (avg)

 37% less efficient
AMD

 MI300A

 $0.0070 (avg)

 46% less efficient


?? Efficiency Notes

  • AI250 racks deliver up to 52% lower cost per token compared to Nvidia GH200 and 46% vs. AMD MI300A.
  • LPDDR5X memory, near-memory compute, and Qualcomm’s low-power SoC architecture are key drivers.
  • Nvidia and AMD platforms are optimized for training + inference, which inflates power and cost for pure inference workloads.
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  

HomeMailHotSubjectMarksPeopleMarksKeepersSettings
Terms Of UseContact UsCopyright/IP PolicyPrivacy PolicyAbout UsFAQAdvertise on SI
© 2025 Knight Sac Media.  Data provided by Twelve Data, Alpha Vantage, and CityFALCON News