AMD is teaming up with Cerebras on a new server designed to slash response times, taking direct aim at Nvidia and promising some of the fastest AI systems on the market. Cerebras CEO Andrew Feldman explains how the partnership works, why speed is becoming the next battleground in AI infrastructure, and what it means for the rapidly evolving AI chip market. He joins Ed Ludlow on “Bloomberg Tech.”
——–
Like this video? Subscribe to Bloomberg Technology on YouTube:
Watch the latest full episodes of “Bloomberg Technology” with Caroline Hyde and Ed Ludlow here:
Get the latest in tech from Silicon Valley and around the world here:
Connect with us on…
X:
Facebook:
Instagram:
Follow Ed Ludlow on X here:
Follow Caroline Hyde on X here:
Listen to the daily Bloomberg Technology podcast here:
More from Bloomberg Business
Connect with us on…
X:
Facebook:
Instagram:
LinkedIn:
TikTok:
@mMorPG13
July 24, 2026 at 4:18 pm
This guy just a pipedream with no actual product.
@briancase6180
July 24, 2026 at 4:50 pm
“they have an enormous backlog.” Yet, somehow, AMD will be able to send Helios systems to your Cerebras cloud rather than the customers who created that “enormous backlog.” Hmmm, maybe their backlog isn’t as big as we thought…. And, “we don’t depend on HBM,” instead, we depend on the literal MOST EXPENSIVE MEMORY IMAGINABLE, on-chip SRAM. And, we depend on whole-wafer integration, also the most expensive fabrication imaginable. Look, they can run small models and small parts of large models very fast. But you have to ask “at what cost?” And the cost is VERY HIGH. Prohibitively high. If you ask me, Cerebras is paying AMD full price for the Helios systems they’ll be getting. So, yes, they *are* using HBM because Helios uses GOBS of HBM, way more than Nvidia. Which makes Helios one of the most expensive systems you can use. Etc. You get the idea. Just be informed, that’s all I ask.
@dwiss2556
July 24, 2026 at 6:29 pm
For the non-techies: Inference is the actual result of the AI thinking. The output. The part that every company currently points to in their Terms of Services as this:
‘We cannot guarantee that answers, services, or other output from this device will be accurate, reliable, appropriate, or complete. It is your obligation to verify the accuracy and appropriateness of any answers or other output and to use your own human judgment, particularly when it involves matters having a legal or material impact on you, such as, for example, health, medical, legal, education, safety, insurance, financial, housing, or employment matters.’
@Very_generic_username
July 24, 2026 at 7:34 pm
How big can their chip scale tho. If the full model + context can’t fit within a single chip you’re just interface limited.