CadenceLIVE Silicon Valley 2025: Industry Insights and the Rise of Agentic AI
I was yesterday years old when I learned the full meaning of Agentic AI. Meanwhile, my "smartphone" doesn't seem to recognize it as a word just yet. I was using the email app to take notes but even the Pages app suggested agent/agents instead.
At least, Google has my back for writing this stuff. While I do not see Google at conferences all that often, their Cloud division was represented at CadenceLIVE along with other booths from IBM, Intel Foundry and, of course, the largest footprint for Cadence themselves.
Figure 1. The Sprawling Santa Clara Convention Center hosted CadenceLIVE again this year. Check-in was easy and the day had my mind boggled by all of the cutting edge technology presented. Image Credit: Author
Checking on my collection of new pens, it's safe to say that there were also booths representing ARM, AWS, GUC, UMC and Silicon Creations to name a few. The common thread was an interest in chip design including substrates. Right from the start, it was a different affair from last year's version.
Keynote Number One: Jensen Huang and Dr. Anirudh Devgan Discuss AI
The hall for the keynote address was the same with one main distinction. Rather than rows and rows of seating, we had tables for ten like at a typical wedding reception. We were treated to a continental breakfast while the CEOs of Cadence and NVIDIA had an unrehearsed chat. The new layout left people standing in the back of the venue so it may not be like that next time.
"AI is not going to take our jobs, but someone who uses AI probably will"
Last year, included an introduction to Palladium while the two executives were doing this same congenial exchange. This year showcased a product that lived in that virtual world for most of the time since that introduction. NVIDIA's Blackwell "AI Superchip" goes beyond the typical GPU. For one thing, it's on a 4nm process node.
Figure 2. The top sentence fragment is a quote from Lip Bu-Tan who is heading up Intel these days. The second line just proves my point that Agentic AI is a new breed of information processing. Image Credit: Author
Blackwell promises to make those Large Language Models more efficient so that they can keep up with the rapidly evolving lingo around this technology. Their founder and CEO, Jensen Huang mentioned a 20X improvement in power efficiency. I believe that he mentioned that figure would translate to a 20X gain in profit in an "AI factory". One moment of candor had him lightly admonishing us that AI is not going to take our jobs but someone who uses AI probably will.
Figure 3. The "tin man" and his dog looked bigger in the video. Meeting it in person did not disappoint but it would need a booster seat to drive a car. The dog had learned some pretty cool tricks though. Image Credit: Author
If my notes are correct, there is only one company in the world capable of producing the Blackwell GPU with its 208,000,000,000 transistors. They are, if you didn't already guess, TSMC who also sponsored the lunch where all of the booths were set up. Jensen also surprised us by announcing NVIDIA's plan to acquire 10 Millenium Supercomputer systems from Cadence that had not been public knowledge up to that moment.
I mentioned the Google booth. I wanted to chat them up but there was a long line of people. The line was for branded water bottles and I already have both the plastic bottle and the thermos from when I worked there. Their Cloud building is walking distance from my home office so I would go there for a hot breakfast and then jump on the shuttle to the main campus.
I knew that they were a chip company because of all of the footprints I was given for fan-out/stack-up studies by their G-Chips team. They announced the existence of that team back in 2019, a couple years after I had moved on to design LiDAR sensors. Ironically, my new manager was the same one who authored the main logic board for their Pixel laptop. We got on really well so I followed him to yet another job in the aerospace industry but I digress.
Figure 4. The CEO of Cadence gives us his thoughts on the five levels of AI prior to meeting with Jensen Huang of NVIDIA. I sat in the last row and it was still larger than life. Image Credit: Author
I approached most of the other booths to get an idea of their wares. Autodesk and SKILLCAD represented layout tools where most of the thrust was on simulation and pre-silicon verification. This really dovetails with the presentations I attended between the first keynote and lunch.
Breakout Sessions - The More You Know
One of those seminars was about datasheets. More precisely access to those fonts of information. There are chip companies that hold their data close to the vest. I recall that Qualcomm was one of those. If you have billions in funding and are starting a phone company, sure, you can get their attention.
For the engineer on the street, or on the DigiKey website, such information is not available. I can't say that I know about blockchain but that was some of the technology that was mentioned as a way of getting tech giants to share their data.
After lunch, I moved from the PCB track to the AI design talk, hoping to find a warmer room. The Q&A at the end was quite lively but I was starting to slow down by then. Retirement will do that to ya! I spent the following 45 minutes walking around nearby Levi's Stadium before returning for the end-of-day keynote.
Figure 5. Using the five levels of ADAS (Advanced Driver Assistance Systems) as a metaphor for different levels of artificial intelligence, it was a nice slide deck and presentation to share. The hexagons represent different "AI Agents" involved with various disciplines. Image Credit: Author
The final keynote was preceded by a "warm-up band", a solo act where the meaning of Agentic AI was covered in detail. Even though it was only 90 minutes past lunch, there were snacks available on the large round tables. Someone came by and offered drinks so I ordered a Coke. The nice man returned soon after with a red can and a glass of ice. That and a giant chocolate chip cookie finished me off.
An Epiphany: There Might Be Something To This AI Stuff
By the end of the day, I began to see the light of something I'd been thus far avoiding. The practical implications here include being able to produce two articles per day rather than two per month. Then I could take the rest of the year off and do it all again.
Keynote Number Two: Intel Weighs In On The Company's Current Focus
Finally, the second "fireside chat" came about. The bosses of Cadence and Intel sat in comfy chairs and discussed the current state of affairs. Lip Bu-Tan's goals include flattening the org chart at Intel so that he gets information that is closer to the ground level while strengthening their ability to adapt to today's challenges.
Figure 6. Another slide from the Cadence graphics team relating the progression of AI tools. I spent three hours on this article. I bet a more locked in writer could produce something similar in a fraction of the time by leveraging AI tools. They could even feed all of the writing I've done over the past seven years into the machine and have it written in my own style. Image Credit: Author
The Intel Chief admitted that the higher up the organization you go, the more rosey an outlook you're likely to present to those above you. That was a moment of candor among several others that rang true with this old designer.
By the end of the day, I felt like I was going from a typical workday at Google. My belly was full and my mind was swimming in new ideas. Maybe there's something to this "AI fad" after all. Outside of Cadence, I had Google, Velodyne, SA Photonics, Microsoft and GoPro using machine learning in some form for over a decade. They can't all be wrong!