/Tech37d ago

Cathie Wood cites OpenAI CFO Sarah Friar to argue agentic AI workloads will shift demand toward CPUs rather than GPUs

GPU focus previously stemmed from parallel processing in model training.

27113154516.8K

#1257

Original post

Rohan Paul@rohanpaul_ai#1257inTech

Agentic AI may be forcing the old computing stack with lot more focus on CPU back into the center of the story.

Here, Ark Invest CEO and CIO Cathie Wood quoting OpenAI's CFO Sarah Friar who has said - "people are chasing GPUs. They're going to be really shocked at how agentic AI activates CPUs"

The market has spent years treating GPUs as the scarce ingredient, because training large models made parallel math look like destiny.

But agentic AI changes the bottleneck.

An agent does not simply ask one giant model for one answer; it plans, calls tools, checks memory, retrieves files, writes code, queries databases, and loops until the task is done.

That means inference is not just matrix multiplication.

It is orchestration, data movement, networking, storage, scheduling, and a lot of general-purpose work that CPUs still handle better than accelerators.

----

From "Bloomberg Podcasts" YT channel (link in comment)

4:38 AM · May 23, 2026 · 14.3K Views

Sentiment

Many users welcomed Cathie Wood's claim that agentic AI will activate CPUs over GPUs as good news for Intel and AMD, while others dismissed her statement or the value of most AI products.

Pos

60.0%

Neg

40.0%

5 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

YOUTUBEVia

#1257

Posts from X

Most Activity

VIEWS2.4KBOOKMARKS2LIKES3

Rohan Paul@rohanpaul_ai

https://www.youtube.com/watch?v=_NRg47lxqR0

Rohan Paul@rohanpaul_ai

Agentic AI may be forcing the old computing stack with lot more focus on CPU back into the center of the story.

Here, Ark Invest CEO and CIO Cathie Wood quoting OpenAI's CFO Sarah Friar who has said - "people are chasing GPUs. They're going to be really shocked at how agentic AI activates CPUs"

The market has spent years treating GPUs as the scarce ingredient, because training large models made parallel math look like destiny.

But agentic AI changes the bottleneck.

An agent does not simply ask one giant model for one answer; it plans, calls tools, checks memory, retrieves files, writes code, queries databases, and loops until the task is done.

That means inference is not just matrix multiplication.

It is orchestration, data movement, networking, storage, scheduling, and a lot of general-purpose work that CPUs still handle better than accelerators.

----

From "Bloomberg Podcasts" YT channel (link in comment)

37d2.4K32

RETWEETS9

Rohan Paul@rohanpaul_ai

Agentic AI may be forcing the old computing stack with lot more focus on CPU back into the center of the story.

Here, Ark Invest CEO and CIO Cathie Wood quoting OpenAI's CFO Sarah Friar who has said - "people are chasing GPUs. They're going to be really shocked at how agentic AI activates CPUs"

The market has spent years treating GPUs as the scarce ingredient, because training large models made parallel math look like destiny.

But agentic AI changes the bottleneck.

An agent does not simply ask one giant model for one answer; it plans, calls tools, checks memory, retrieves files, writes code, queries databases, and loops until the task is done.

That means inference is not just matrix multiplication.

It is orchestration, data movement, networking, storage, scheduling, and a lot of general-purpose work that CPUs still handle better than accelerators.

----

From "Bloomberg Podcasts" YT channel (link in comment)

37d14.3K10943

REPLIES1

Dan McAteer@daniel_mac8

@rohanpaul_ai Great news for Intel and AMD.

37d2551

Rohan Paul@rohanpaul_ai

@daniel_mac8 exactly

37d1551

Minh Nhat Nguyen@menhguin

@rohanpaul_ai >cathie wood on bloomberg oh, oh dear

Rohan Paul@rohanpaul_ai

Agentic AI may be forcing the old computing stack with lot more focus on CPU back into the center of the story.

Here, Ark Invest CEO and CIO Cathie Wood quoting OpenAI's CFO Sarah Friar who has said - "people are chasing GPUs. They're going to be really shocked at how agentic AI activates CPUs"

The market has spent years treating GPUs as the scarce ingredient, because training large models made parallel math look like destiny.

But agentic AI changes the bottleneck.

An agent does not simply ask one giant model for one answer; it plans, calls tools, checks memory, retrieves files, writes code, queries databases, and loops until the task is done.

That means inference is not just matrix multiplication.

It is orchestration, data movement, networking, storage, scheduling, and a lot of general-purpose work that CPUs still handle better than accelerators.

----

From "Bloomberg Podcasts" YT channel (link in comment)

37d14210

High Jack@jackadoresai

@rohanpaul_ai Oh gosh, we spent years optimizing training throughput and now the bottleneck's just scheduling? Inference stack's getting weird.

37d791

Raven Protocol 🐦‍⬛@raven_protocol

@rohanpaul_ai Raven supports heterogeneous hardware: GPUs, CPUs, ASICs, edge nodes.

Agentic AI just proved why that matters.

37d691

arbid intelligence@arbidthink

@rohanpaul_ai Agentic AI does not reduce the system to one giant inference call.

It restores the older stack: CPUs, memory, storage, scheduling, and orchestration. Once cognition becomes multi-step, the scarce layer is no longer only acceleration.

It is coordinated execution.

37d461

Michał Piszczek@cdiamond

@rohanpaul_ai orchestration is real, but token generation still owns the invoice. CPU just holds the queue

37d86

Chen Avnery@MindTheGapMTG

@rohanpaul_ai Most agent compute is waiting on file I/O, API calls, and browser automation. Not matrix multiplication. We run 12 production agents daily - none of them need a GPU. The bottleneck is latency and orchestration, not FLOPS.

37d47

Blake Heron@BlakeHer_on

@rohanpaul_ai this matches what you see building agents. most of the loop is orchestration, retrieval, and tool calls, not matrix math. the gpu sits idle between steps.

37d44

David J.@lordofblocks

@rohanpaul_ai OpenAI’s CFO flagging this publicly means the market repricing is coming.

The question is which CPU and general compute plays benefit most when the rotation happens.

37d40

未知@luyun0120

@rohanpaul_ai AI产品遍地都是，能打的确实没几个。

37d30

Home@homeMetaX

@rohanpaul_ai The core shift with agentic AI is that inference increasingly resembles distributed systems coordination rather than isolated model execution, which naturally elevates the importance of CPUs, memory hierarchies, storage, and networking infrastructure.

37d29

tawer@tawer1O

@rohanpaul_ai Load shifts to cpus past ten tool calls

37d26

...::a.... Murray baggins@SiSiso077

@rohanpaul_ai Napster lol

37d23

David T Kramaley@simplydt

@rohanpaul_ai interesting shift-what's your biggest orchestration pain point now?

37d16

Bnaf.OG | 🟧@bnafOg

@rohanpaul_ai This is the underpriced agent-infra point: GPUs make the model smart, but CPUs coordinate the mess around it — tool calls, sandboxes, tests, logs, queues, and state. Agent speed is orchestration throughput, not just tokens/sec.

37d12

AI Mastery Guide@aiseomastery

@rohanpaul_ai The market spent years learning one playbook and agentic AI just changed what the scarce resource actually is

37d7

Claire Dubois@Claire_Dub18

@rohanpaul_ai TSMC: pricing power first. If margins hold as capex normalizes, the theme survives. Read more here: https://orion.aiwork.app/theme/ec467cbe-0125-48af-8855-68fa4e39b4c0?hl=en&utm_source=X&utm_medium=social

37d1