The Deep View: Conversations

What happens when AI moves from cloud-only to running everywhere, including on your laptop, your phone, and other devices around you?

In this episode of Deep View Conversations, senior reporter Sabrina Ortiz sits down with Olena Zhu, who leads AI for the client computing group at Intel, to explore one of the biggest shifts underway in AI: the move toward accessible, affordable, and privacy-first AI systems.

Zhu explains why the economics and infrastructure demands of cloud-only AI may not scale indefinitely, and why on-device AI could become a critical part of the industry's future. She also reflects on the evolution from traditional AI systems to LLMs and now to agentic AI, and why this wave feels fundamentally different from the hype cycles that came before it.

The conversation also dives into how AI is changing the way people work, learn, and experiment, including the surprising mindset Zhu believes helps people get the most value from AI tools today.

Topics covered include:
+ Why cloud-only AI has limits
+ The future of on-device and edge AI
+ AI affordability, energy use, and data sovereignty
+ How agentic AI changed Zhu’s workflow
+ Why experimentation matters more than expertise
+ Intel’s vision for privacy-first AI systems
+ The hidden infrastructure challenge behind AI growth
+ Why AI adoption may depend on trust and accessibility

If you're concerned about the affordability, accessibility, and privacy of AI, you don't want to miss this episode.

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology.

And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights to keep you ahead of the curve and help you put AI to work every day: subscribe.thedeepview.com

Creators and Guests

Host
Sabrina Ortiz
Senior Reporter at The Deep View

What is The Deep View: Conversations?

From frontier labs and enterprise platforms to emerging startups reshaping entire industries, The Deep View: Conversations podcast interviews the brightest minds and the most influential leaders in AI.

Jason Hiner (00:02.254)
In this episode, senior reporter Sabrina Ortiz talks to Dr. Elena Zhu, who leads AI strategy for the client computing group at Intel. Dr. Zhu has spent her career working on AI, and now she's focused on one the most important questions in the industry. How do we make AI accessible, affordable, and safe for everyone? She talks candidly about how the shift from backend AI tools to LLMs to aogenic systems has unfolded from her vantage point

as an AI practitioner and why the current moment feels different from all the hype cycles that came before. She makes the case that the cloud only model for AI has real limits from costs and energy consumption to data sovereignty and regulatory compliance. She explains how Intel's AI Super Builder platform is helping companies build privacy first AI experiences on device.

Dr. Zhu also talks about how Agenic AI has transformed her own workflow and why she thinks open-minded experimentation beats experience when it comes to learning AI. And she shares a tip for getting the most out of AI tools that might surprise you. Approach them the way a kid approaches a new toy. So here it is, our conversation with Dr. Elena Zhu of Intel.

Sabrina Ortiz (00:01.518)
Alright, perfect. We could start with you just telling us your name and then we'll hop into the first question.

Olena Zhu (00:07.015)
Okay, I'm Dr. Olina Jianfang Zhu. So Jianfang is my Chinese name. I go by Olina.

Sabrina (00:15.96)
Perfect, thank you so much. I'm about to hop in. If it sounds abrupt, that's because we're starting the podcast. But today on the show, we have Dr. Alina Zhu. And I would love to start by you telling us a bit about what you do, your background, and how you got into this world of AI to begin with.

Olena Zhu (00:35.691)
Awesome. So I'm leading the AI solution group and AI strategies for client computing group at Intel Corporation.

how I get into this role. So that probably need to started from like my childhood. So when I was a kid, I just loved solving puzzle problems. You know, I'm just, I was such a big fan about math and solving those complicated puzzles, know, geometrical problem and all that is the way of me killing time.

Yeah, and I dreamed about solving those problems in my dream. I remembered and I woke up. because I got the answer to the question I couldn't solve during the daytime and I just dreamed up and woke up. yeah, I got it. So it's really just fun to me. think that math is something I really love.

Sabrina (01:45.038)
How old would you say you were? Like how little did these dreams and this fascination of solving puzzles start?

Olena Zhu (01:51.359)
I think that was like elementary school.

So I'm a nerd, you can see, starting from back then. So that is all the way through my education. So my education is on computational science. So it's all about use math to do the modeling and solve all kinds of physical problems and things like that. And at Intel, I started to do a lot of, you know,

algorithm for design, solving simulation problem, AI for design. And then in recent years, and I started to expand to see, hey, not only using AI to solve just electronic design problems and engineering problems, it's broadly, how do we use AI for every day and for everybody? so that's my current focus.

Sabrina (02:50.926)
How has it been for you observing firsthand a transition between AI first being something that was in the backend, something that again only maybe engineers or people working to optimize the business side were really focused on to it being this phenomenon it is today?

Olena Zhu (03:08.898)
I love your question. Love your question. Cause indeed it's just fascinating. You know, as an AI practitioner, it's, it's, have witnesses how this being transformed, you know, slowly and fast, right? You know, at the same time. So why I started, you know, like with my education and all that, it was not a big deal. It was like computer vision. was like machine learning.

and a lot of bubbles, it were bubbles and all that and people were like, you know, when I started and people even questioning, know, AI was a buzzword and whatsoever. And what we're doing the AI for design, there were some great success, you know, we built the tools together with different domain engineers and for the, you know, really good ones and the, you know, the domain really needed something to break through and they become very hungry and they do this.

with us and boom, know, we got tools up rounding 80 % of efficiency again, solved the problem they never could solve. But we also hit a lot of walls with other domains. some of the domains, were like, you know, just subconsciously they compete with AI. And, you know, like, you know, we build AI tools and telling them, hey, you know, in the beginning, will not be perfect and you have to be patient and then let's improve. And they were like,

I can do way better than this. I'm not going to use it. And all those things. So it was very, very interesting. then when LLM came about in 2023, 2024 and all that, boom, wow, really, I can ask it. It can answer questions. That already kind of like a big shift from industry and everybody, the way, the view, how they look at

Wow, AI can do so much more and so real and just touch each of us. And then this year, right, we started this, you know, the whole big open call agentic system and autonomous AI.

Olena Zhu (05:26.708)
push it to the next level. It's like, you know, don't need to really know coding. You really don't need to like, you know, do this question answer and trying to find things. Just give a task, leave it to AI, to the agent and let it figure it out for you. So that is another level of a game changer. And, you know, if you really interesting example, and I saw there's, you know, like a Reddit type of forum and people are talking.

about how they get so much comfort and a companion from AI. that's one aspect. And another aspect is how much it can empower people to do different things. One example, I have an eight-year-old daughter and sometimes I want to help her with understanding the math concept, but it's hard. But you know, every kid likes playing video games. So I was like,

I need to build something really quick and so that she can play it and understand it. But you know without AI and you think wow you know that will take a day or two. I don't probably don't have that time. Now like I can guide my daughter and she can do a video build a video game for herself like in 20 minutes and although you know it's it's not that polished and all that but

She loved it because she developed it. And then she now plays with it and try to understand the multiplication concept and the practices. That's how you use AI to empower things and do a lot of things we couldn't do before.

Sabrina (07:10.958)
I love that example because it really just makes it really clear how accessible AI tools are. Your eight-year-old daughter is building video games in 20 minutes. That's incredible. I don't even know if I could spell my full name back then. That's leaps and bounds what I was doing at eight years old. But with that in mind, there are so many advances in the AI space. You kind of just went through a really quick timeline where it was like, wow, we have a chatbot in 2023 that could actually

answer conversationally, that felt like the next cutting edge thing. So now where we have these autonomous agents that can do much, much more. You work on every day bringing AI to personal devices at Intel. What are you focused on? Where are you the most focused on right now bringing to consumers and their devices?

Olena Zhu (07:57.353)
Yeah, again, I love your question. You always ask all these beautiful questions. So I think in high level, our mission is really wanted to make AI democratized, more accessible, and also safer.

Sabrina (08:03.096)
Thank you.

Olena Zhu (08:15.37)
So why we are saying so is especially with this AI agent, it runs 724 hours. And if you just purely rely on cloud, no one can afford it. a lot of us actually would personally use it. And even within two hours, you almost can spend all your monthly subscription credits.

Last night I was talking to a few smaller business companies and they were like, we just cannot afford. So we started to set up our own on-prem workstations and all that. So we see how powerful it is, but we need to make it sustainable economically. And that's one point. And another point is the energy limitation.

infrastructure limitations. So now everything goes to cloud and all centralized. And even the AI providers, like Anthropec and everybody, like so constrained on scaling. So that's why they put a lot of like, you know, different types of usage controls, burst usage controls and all that, because it just cannot scale. And that lead to the environmental penalties to us. You know, like...

the water, how much we use to cool and all that. It's really a globalized sustainability issues. And the last point is the safety, security. And what OpenClaw just got invented and a lot of people rushed into it and played with it and use it. And some of them, like their API key were stolen and credit card information.

and the whole email box was deleted and all that. So all the security aspects is getting improved. again, if we everything we do and rely on cloud and.

Olena Zhu (10:26.026)
That's not going to work, right? We have our personal stuff, health care record, and our kids' pictures, videos, and all sorts of things. And a company has their secret IPs and their financial data, customer information. Are we going to all just give it to the cloud? A lot of these disruptors in the market, like the AI

startups, they have a very hard time to get into enterprise space to service the enterprise needs because the regulations from global wide, know, different geo government and vertical industries, healthcare or legal, they all have pre strict regulations, you know, you know how you handle all these local files and the data's and they, you know, it's not, it's not like you can do everything in the club. So

With all that said, we are focusing now is really trying to build a hybrid.

solution that connecting local and cloud so that we can just keep user data, especially the critical, confidential, and personal data local. But we still utilize the cloud AI advantages. In simple words, we use cloud AI to decompose the very complex test into small steps, and we distribute those small tasks across

local and cloud and use local and to handle the local data without local sending the confidential data to the cloud. and from the device perspective you know like when OpenClaw started like people say you know like we need we need this and that device to run but

Olena Zhu (12:27.562)
In reality, everything still goes to cloud because the local device is only a gateway solution. That's why you saw on the market, like people say, hey, I can do a small device even like less than $100. I can do this. I don't need a fancy devices. So what we are trying to do is to do a real local first AI device. That's the goal. we want people to keep their stuff local. Go ahead.

Sabrina (12:40.718)
Thank

Sabrina (12:55.104)
No, I was just saying, I like that you're touching upon this whole entire hybrid model concept because I feel like, again, when AI first became insanely popular, the focus shortly after became this idea of local running of everything or on edge running of every model that you could, which also led to building smaller models and then people feel more constrained about what the AI could do. But as these smaller models become more powerful,

the idea of it running locally becomes more feasible, but also to your point, there is the need for the hybrid still. So in your opinion, I'm curious, are you finding that there's also a demand from customers for it to be hybrid? Or because I think for a while there, what we were all hearing was that everything needed to run local to be safe, to be privacy first. But this hybrid model again kind of challenges what was said in between there.

Olena Zhu (13:50.475)
Yeah, I think everything is journey, right? It's like, you know, like either one or zero cloud or local, right? It's a journey. and this, you know, a lot of, you know, very complex task and all that has to be done with local and the frontier models. are updating every day and all that. So we have to, we have to take advantage of that. And also on the other side, like you said, local is getting more and more capable and

model are being more and more capable and also the local devices, much higher T-tops and better memory and memory bandwidth. So all that is evolving. So maybe in the next few years, we don't know when, maybe all of a sudden, a pure local solution can be really great. You buy a financial advisor, AI in a box. Boom.

does your tax and consult with you about your investment strategy whatsoever. It's good enough. You don't really need to tap to cloud, right? That could happen, but it's a journey. So we start with now like

cloud dominant and then we start to see, okay, people realize local and much, you know, give you more economic and security and all these benefits. So we see people are coming from this way. We need to bridge the gap and build the bridge to help this transition happen.

Sabrina (15:27.256)
So I like too that we're talking right now about like the backend, right? Cause on the user, if done right on a user's end, if I, you know, buy a PC and I wouldn't really necessarily be cognizant of whether it's running locally or in the cloud, it should be somewhat of a seamless handoff. Yes.

Olena Zhu (15:42.229)
See you next time.

Yes, that's one of the key components we are building is seamless experience between local and cloud and the algorithm automatically figures out it is suitable for locals, it's suitable for cloud, it is like how complex this problem is and how sensitive the information is and how time sensitive it is and all that. So a lot of the criteria based on those criteria and automatically route to local or cloud

combined.

Sabrina (16:15.822)
I'm sure as you're also thinking about building these devices that are like AI first, you're also considering though the experiences that users will notice when they're opening the device and they see an AI feature or an AI enabled application. And I think something that comes to mind, for example, is a super builder. So I'm curious how you think of those experiences. We're slightly different, right? These are more front-facing and need to be more noticeably useful to the user.

Olena Zhu (16:45.61)
Yeah, so SuperBuilder, we started with SuperBuilder like two and a half years ago. And when we started, it was like a purely local solution. And then we use it as a reference design. we build like we're baking the performance local models, small models optimized with the four Intel platforms. And also bake all kinds of solutions like REC.

and documentation scoring and all those agent workflows and components into it and service for our customers in two ways. One, like you said, it can be a standalone application and user by AIPC download they can use it and it's purely local.

Sabrina (17:36.994)
Yeah, how about you remind our readers or our listeners about what Super Builder is in case they're unfamiliar.

Olena Zhu (17:44.395)
Yeah, yeah. Like I said, you can go to AIBuilder.intel.com and we also have a GitHub to it. So that is where you can find Intel AI Super Builder. And it is, like I said, is a reference design platform for our customers, partners, and even users to utilize the AI solutions on Intel AI PC. So it started with local only in 2015.

It was local only. We select and optimize a set of small models and optimize them, make them run best on Intel platform. And then we bake all kinds of components, including reg, including agent workflows, like scoring documentations and things like that, and all into it. And to service all kinds of customers.

Some of them, you know, like they can directly download and use it. It's a chat bot and all your documentation. You can talk to your local documentation and all that without, you know, uploading to cloud and whatsoever. Or, you know, for our, you know, customers, some of them use partial of the solution because we give them APIs or use the entire solutions and for their own purpose. We serve two categories of

One is the traditional customers including OEMs, including traditional desktop application vendors, they use it to build their own agents. For example, Samsung launched the AI book.

bookstore agents, like in the beginning of the year for Kyobo Bookstore. Many examples like that, they use this as a building block, reference design to build for other enterprise customers with their devices. And some others, like, you know, they use it to build developer kit, you know, to encourage more developers to develop on their platform. And then the second category is more like the AI solution pro.

Olena Zhu (20:03.98)
providers, all kinds of solution providers, from model vendors and to application vendors. So I'll give you one example. We already productized one with a China company called ChatPPT, not GPT. So basically, it's an AI-powered PowerPoint generation company. Everything is on the cloud.

50 mini users and all that. It's a very promising startup company. their concern was like they have to...

again, service, all kinds of customers, including government, including legal and finance and all these verticals. Like, hey, people have concerns. They need a privacy-enhanced version. How do we utilize AIPC? And then we developed the MCP, the Model Context Protocol, that type of framework and APIs to give that to them. So what they can do is they have a small local model. And AIPC, when you download, automatically downloads there.

and they also have their full blown cloud services. And then when you want to make a PowerPoint, what happens is it will scan your local files, use a local model, and then just destue it, right? And just abstract the high level outlines and the summaries and remove financial information, all those things. And then send the outlines to the cloud. And the cloud, to generate the PowerPoint template, do the picture rendering,

you know, color theme, beautified and all that, and send back to the local and local can insert back those, you know, local content to it. So they launched it in last November with, you know, with AIPC users. And so the telemetry data showed the initial telemetry data showed like the user sessions is increased by 33 % because the user trusted it's their device, you know, it's not like you go.

Sabrina (22:12.142)
Mm.

Olena Zhu (22:14.828)
that website and do stuff you know that's their device and then also they have estimation the projected estimation for token saving is 50 %

Sabrina (22:26.264)
I like that you just mentioned this tool that as much as I've been covering the space for now about four years and you know, write about this every day, I'd never heard about this tool and that just kind of shows you how many AI solutions, applications and tools are out there. I'm super curious on your personal everyday workflow. Are there any AI tools that you gravitate towards or that have really upgraded how you work?

Olena Zhu (22:51.774)
Yeah, so now I build my own agents to do all kinds of tasks for me. For example, we build this next version, next generation of super builder. So we'll have a major announcement coming soon. But we have to do a lot of testing. Previously, it's more like human testing, click buttons.

giving task and see how the output and all that. was like, hey, it's the perfect agent work. And so I started to use AI to build an agent for this. in the end, a lot of iterations, right? But in the end, you will see, OK, you don't touch your computer. And one agent will go to this platform you build and start to click buttons and go through every steps.

and do the test for you.

Sabrina (23:55.703)
It is so cool seeing now with AI applications, seeing basically your computer know how to use itself. I'm sure that it's, is it fascinating to you too or less so because you've been seeing probably things of this sort happening for a bit longer than the rest of us.

Olena Zhu (24:03.847)
Yeah.

Olena Zhu (24:13.702)
It's still pretty fascinating, right? Because in the past, even like maybe past six months, you still need to give more attention to it. You need to guide more. But now it's like, if there's a problem.

And this agent has more autonomous, you know, to go to everywhere and look for solutions. And it's more like before you may have like a high school student as your intern, right? That's your AI. And you just need to step by step. What do you need to do? Bah, bah, bah, bah. And it got stuck and come back to you. What do I do? Right now you probably have a PhD level student as your intern. If they get stuck, they do research and they,

They either solve it themselves or give you a proposal, discuss with you. So it's a delta job.

Sabrina (25:16.694)
Yeah, it's awesome for me too whenever I'm testing these tools again, just seeing it click around and do things that before, even if it would have taken me an additional couple of minutes out of my day, it's nice to be able to offload that. With that too though, I think there's still some hesitation from users on what to use exactly or if they should be using what tools. I'm curious for your point of view.

What's something that everybody should keep in mind when interacting with these AI applications and newer cutting edge technology that could help supercharge their workflow or take it to the next level? Maybe it's like a frame of thinking or maybe it's a way of interacting with them that can maximize how they use it.

Olena Zhu (26:06.982)
Yeah, I think probably two points. Number one is, know, just you think about like who goes the fastest with AI learning. Kids, Like the college students, know, the high school basically people have less knowledge about the history.

and they even don't even know about coding much maybe. So they just jump in and then just, you know, use all this. So I feel like, you know, use AI in a more playful way, right? You know, like all of a sudden you get a new toy and you go experiment it and you know, just take.

of all this baggage like you know, you know, for example, I'm you know, I'm I have many years of experience of this and that, you know, but just give it a try, give it a try and with very open mind, open minded, you know, yeah, it's when you start experiment,

And you realize, hey, wow, I can actually make it to be a really useful tool. But starting with more open mind, mindset. So I'm also a drug professor at Purdue. So I went back to Purdue last week and gave talk and connecting with students and professors. And I realized the students actually goes faster with AI.

and more than professors that's what I mean and so I was like hey you know professors try it and you'll be surprised so that's that's what I mean and then once you get onto it and you go really fast with it and you can just become extremely helpful and you can run a hundred miles a day and and there was something really interesting I saw on the website it were like wow like before I log off my computer I was thinking

Olena Zhu (28:26.476)
What kind of task I should give to this agent overnight? What kind of magic miracles it will create? you know, like all of a sudden you, you, got, you got so many, you know, you know, like AI helpers for you and they maximize your productivity is your imaginations, your credit creativities without your attention. So that's, that's, that's one thing. And another thing is, um,

Be careful at the same time, right? So the security is a big deal. It's really a big deal. That's so, you know, with AI you can, you will get out of it.

based on what you put it in. for all the confidential information, your personal stuff, be really careful about it. So those type of things, because we also saw a lot of lessons from the stories we have seen. That's why experiment with AIPC and with SuperBuilder. And like I said, we are rolling out more

versions and all that and try to use it in a safe way.

Sabrina Ortiz (29:52.738)
I like it. Keep the whimsy and experimenting with AI while also keeping safety and privacy first. Thank you so much, Dr. Alina, for all of your time today and all of your insight. This has been such a great conversation. I really enjoyed getting the chance to chat with you.

Olena Zhu (30:09.086)
Thank you so much. I really appreciate the opportunity. It's fun. Yeah. Thank you.