nVidia CES 2025 Keynote & AI Supercomputer at your House (Bonus: Agentic AI is here)
UnWired Blog - nVidia and CES Keynote Takeaways
Is everyone an AI robot developer, from home? Seems on the horizon.
The push for an optimized development environment that can create full-stack AI applications is here. (launching in May 2025, and announced by nVidia during the CES keynote).
I was lucky enough to attend the nVidia keynote at CES '25 on Monday, Jan 6th, 2025. I was lucky because despite being held in arena that likely held 10-14K ... the line to attend this event (that was not let in, may have been more people outside). The only our team made it in was a true line soldier saved us a spot and waited hours (Thanks Lionel!! Warrior!) or there is no way we would have made it. An electric atmosphere, which is hard to imagine for a GPU company? CEO Jensen Huang is an incredible orator and story teller. He was on his game, and the entire show (nearly two hours) was a one man show. There were many highlights, but one or two ideas intrigued me more than others. The data center buildouts, iterations of Blackwell, and speeds and feeds for GPU, this seems to be intuitively obvious and needed to simply continue the AI build out and momentum.
Jensen wows the crowed for 2 hours at CES ‘25 in Las Vegas. (Mandalay Bay Convention Center)
The first was the desktop AI platform. The nVidia "Projects Digits" which sports a Blackwell GPU in a Desktop form factor is at first glance curious? Not requiring a cloud interface, it is effectively a full-stack AI supercomputer in a box. This might enable inexpensive, home AI development - or just an incredibly powerful gaming environment? If the former, I think WAY back about one of my early jobs as systems and application programmer for Sony. One of my early development efforts was to offload our teams CICS/VSAM development from the mainframe environment (IBM 308x) to PC Lan Networks running OS/2. The purpose was to save money (compute time in this era was expensive - see table) and to enable aggressive coding (faster) to save time. This was allowed because writing code on Mainframes was delicate. A crash or STOP RUN, or poorly crafted scripts could actually take a region of the system or system down. Restarting these beasts was a slow and painful process. I use that analogy with the cloud based AI programming of today (albeit not as sensitive to errors as it once was). However; The token cash register is painful during the development process, especially during training, and removing these barriers for the DIY home guy, the 1 man band startup, early stage companies, or even big companies R&D, could save significant time and money while experimenting and building the next generation of AI services.
The second intriguing product was nVidia's AI engineering of the Physical world.. The World Foundation Models (WFM) to have AI train AI models that understand human and real world physics. At least they did not call it "Universal" :>)
Hypothesis:
An AI desktop can enable a new AI development paradigm for AI modelers and developers; perhaps even regular, old users? Maybe anyone can develop AI solutions? This combined with the World Foundation Model for real world physics. The possibilities to create intelligent, physically capable home equipment and robots seems within reach? We may not need Tesla or Boston Scientific. On the other hand, it is absolutely frightening what kind of weapons they may be able to create.
Implications: and possible advantages
More aggressive programming, training, and testing. Using AI to do it.
Cheaper (avoid the token charges in the cloud). Once ready, deploy to the cloud. (or not)
Faster (not competing for resources) development and less dependence upon tight resources.
Historic Comparison of Compute Costs (for fun)
General “cloud” comparisons from the 1980’s thru Today
`
nVidia Announcements - Key Highlights
Cosmos Platform: AI ecosystem for physical AI, enabling developers to train robots and AVs using world foundation models (WFMs) that simulate real-world physics.🤯
GB10 & DIGITS: AI compute system designed for desktops, providing AI stack for enterprises and developers. Available in May. Starting around $3K.
GeForce RTX GPUs: Next-gen AI gaming processors to enhance realism and performance. Laptops with these GPUs will launch in Spring 2025 at $1,300–$2900 depending upon model.
nVidia's AI Arc.
Jensen explains the AI Arc and where we are now.
Agentic AI is upon us.
KEY TRENDS
AI Technology Arc: The evolution of AI follows a clear path, starting with Perception AI and progressing through Generative AI, Agentic AI, and finally Physical AI. Each stage represents an increasing level of autonomy, reasoning, and real-world interaction:
1) Perception AI (Recognition & Understanding) focuses on analyzing and interpreting sensory data (e.g., images, speech, text). Foundational for computer vision, speech recognition, and NLP.📌 Examples: Facial recognition, speech-to-text, medical image analysis.
2) Generative AI (Content Creation & Synthesis) builds on Perception AI to create new content—text, images, audio, video, and even code—by learning from vast datasets. Models generate outputs that mimic human creativity and reasoning.📌 Examples: ChatGPT, Midjourney, AI-generated music, and synthetic speech.
3) Agentic AI (Autonomous Decision-Making & Action) introduces goal-driven autonomy, where AI actively makes decisions and actions based on objectives, rather than only responding to inputs. AAI systems, self-optimize, plan, and execute tasks without constant human guidance.📌 Examples: AAI agents alone trade stocks, optimize logistics, conduct scientific research, or function as personal EAs.
4) Physical AI integrates AI with robotics and hardware, enabling intelligent machines to interact with the physical world and manipulate objects, navigate spaces, and perform tasks in real-world environments. 📌 Examples: Autonomous drones, self-driving cars, humanoid robots (e.g., Tesla Optimus, Boston Dynamics), AI-driven surgical robots.
There are several related Agentic AI announcements of note that have followed since CES.
OpenAI's "Operator" release. AI Agents are here!
First, to test "Operator" you do have to upgrade to use Pro (which is an additional $180 on top current Plus subscription), but you you really don't have to pay for Operator. (the upgrade includes a bunch of other stuff too, including "unlimited" inference. - we will see about that.)
However, at this point, there is no charge for specific agents? Worried about what the OpenAI business model is here?
There are only two logical options.
a. Charge for the agents (i.e. agent "app" store)
OR
b. OpenAI Sells YOUR data
Given how they have started this .... Looking like option B.
The use cases they are promoting are mostly 1lame. Meal plans, dinner reservations, shopping .... creativity using these agents is severely lacking. My favorite meme is one day all old computers will be down in the basement
Stock Trading
Bitcoin Mining
😋
The theory that I do appreciate regarding this Agentic journey is that this (Operator) AI Agent is the first tool that enables an agent to do things just like the humans do (same input and output for AI as for humans). Typically humans operated computers via (mouse, keyboard, browser, etc.) and computer software (Programmed BOTS, APIs). Operator works like humans do (in the browser). The follow on will be more Agents that expand on this theme - imitating humans in the real world in multiple vectors. NOTE of improvement: Operator needs to run in our personal browser in the next version though. Information here: Computer-Using Agent | OpenAI
There are open-source versions of Operator like capabilities (run via your browser) that won't be selling your data. Highly recommended. 😋
DeepSeek is making noise, a Chinese based open source competitor with OpenAI and Operator. DeepSeek Much less expensive AI alternative.
Happy hunting.
*****UnWired Founder and Protagonist James "Jim" Pyers ****
Show Credits:
ME : Check me out here at Wiresworld or on linkedin - James (Jim) Pyers | LinkedIn.
We are value for value! Please Support UnWired @ www.wiresworld.net
Please follow me and what I am working on at wiresworld.net/mybiz.
For show notes and AI musings Check out my blog at wiresworld.net/blog
Keep the show thriving by donating at wiresworld.net/donate
Show Music: Thank you Kevin Keith, “Like a Holiday”. Kevin Keith - "Like a Holiday" (youtube.com). Visit Kevin at www.kevinkeith.com