Slight delay in this week’s episode due to work stuff, but better late than never!
NB - I’m currently also working on a chunky future episode regarding glass substrates and co-packaged optics, and how Intel is storming ahead on these advanced technologies. Both of these are incredibly interesting moonshot projects that complement one another nicely, and could take Intel Foundry to an entirely new level around the 2030 timeframe… it’s something that I have never really gotten my teeth into until now, so I’m very excited about this one. Stay tuned for that in the near future!
Anyway, we have got a lot of interesting info over the last week which I’m going to digest newsletter-style. Let’s go through one topic at a time.
Lip Bu on “The Angle” Podcast
This was a great interview where Lip Bu divulged some very interesting tidbits. I strongly suggest giving this one a listen, but for those who don’t have the time, to summarise; Lip Bu is here for the long haul, and he is incredibly excited about Intel’s prospects.
He confirms that “he is quietly building the GPU” - likely a reference to Jaguar Shores. He also confirms that the new partnership with Foxconn is going to be instrumental for getting Intel CPUs & GPUs into a proper, integrated “rack-scale” solution for AI data centres.
He wants agentic AI to be a big revenue driver, and to this end he is looking to hire “one or two” top-notch CPU designers to steer Intel into a leadership position here. I’m super excited to see who he brings onto the team. The immediate names that come to mind for this are Jim Keller (now CEO of Tenstorrent), and Gerard Williams III, who recently left Qualcomm and started Nuvacore (which is specifically focused on designing the world’s best clean-sheet agentic CPUs). Bringing either of these CPU Titans on would likely involve very large purchases of Tenstorrent or Nuvacore, so perhaps Lip will go for someone a bit more under the radar with a smaller price tag. Personally, I think buying the entire Nuvacore team would be worth the investment if agentic AI is here to stay.
Lip also divulged that he believes the primary future growth of Intel will be in “custom purpose built silicon & connectivity”. I’ve been banging on about this for eternity, but this essentially means Intel becoming a custom chip designer that integrates both design and foundry services to rival companies like Broadcom & MediaTek. “Your margin is my opportunity”.
There was also an interesting comment that Lip Bu made about Arm which I couldn’t fully decipher, but it was potentially a hint that Arm may use Intel Foundry for capacity (either wafers or packaging) for their future chips. He has hinted at this before when he stated that the SoftBank $2Bn share purchase was tied in to “securing future capacity at Intel Foundry”, which would fit with Arm using IFS at some point.
Finally, what came as music to my ears was that Lip Bu plans to steer the ship at Intel for the next 10 years. This is so incredibly bullish. Intel is clearly shaping up to be the pinnacle of his career, and ultimately I think it will be 10x greater than his turn-around of Cadence.
Foundry Updates
Now there’s been some important foundry news lately that we should wrap up. Firstly, 18A-P is in risk-production now and is on track to enter HVM in Q4 2026 (& 18A-PT base die scheduled for 2029). Importantly, this means we could see potential customers such as Apple putting Intel-fabbed products on the market as soon as 2027. All the latest metrics and IP for 18A-P are looking great, and it’s shaping up to really be what the vanilla 18A was supposed to be.
14A development is strong, with a D0 of 0.5 - this is fantastic considering they are only at the 0.5 PDK stage. For reference, 18A was reported to have reached a D0 of 0.4 in September 2024, with the 1.0 PDK already being released in July 2024. The fact that 14A has a similar D0 at an earlier design stage is brilliant news, especially considering it's an even more advanced process with PowerVia 2.0. The 14A 0.9PDK is expected around October, so I imagine big customer rumours for 14A will come in Q1 ‘27 after they have had a chance to tape out and fully evaluate the test chips/final refinements for the 1.0 PDK. I am seeing new rumours from Bloomberg of Apple potentially using Intel 14A-E for 2028/2029 iPhone chips alongside TSMC in a dual sourcing strategy.
Lastly, Intel is gearing up to spend more on capex, and it is reported that the Intel board have confirmed to Lip Bu they are happy to green-light heavier spending here. One concern I have is that Intel was not expecting the agentic CPU surge to run concurrently with increased external customer wafer demand, with Fab capacity only planned to cater to the latter. I calculate that with all current shells tooled and Ohio Fab 27 Mod 1/Mod 2 constructed, it is going to be EXTREMELY tight, if not impossible to meet both the internal agentic CPU demand and all the new external demand.
I think Intel is going to have to announce a new “Fab 37” for Ohio, or an additional new module at the Arizona site (“Fab 72”) or the Irish site (“Fab 44”). If they don’t, they are putting a hard limit on their future market cap, which I am now revising to around $1.5-2Tn by end 2032 (assuming AI demand continues as it is at present & no major geopolitical/macro setbacks). This is a maximum ceiling based on internal/external wafer revenue & product sales that I believe they cannot physically exceed without more fabs.
We also got some more news at VLSI on other bits such as progress on their GaN-on-Si Power Semi efforts;
“GaN + Si integration for power management: Intel demonstrated 300mm monolithic integration of gallium nitride power devices with silicon logic, including a ~1,000 gate digital control block, enabling efficient, large scale digital control alongside high performance power devices in a single process and reducing system complexity”
Now this is important as terrestrial AI data centres move towards higher voltage infrastructure, as well as potential future applications for space-based data centre satellites due to the inherent radiation tolerance of gallium nitride.
Miscellaneous
Ohio construction job hirings at Bechtel continue. They are clearly planning for the steel roof truss construction to start ASAP. Back in January when I first reported on the new Bechtel hirings, I predicted that roof truss construction would start around July/August; I’m hoping that I’m not wrong and that we see some evidence of this in the coming months on Lukateake’s YouTube channel.
In terms of SpaceX/Terafab, Elon has been courting ASML to get allocations. I’m interested to hear more about how this partnership turns out. Clearly the market are pricing in success of SpaceX 14A-E based AI chips with a valuation of $2.5Tn. If SpaceX really is this successful with AI data centres, certainly Intel’s licensing revenue would be huge; likely in the order of billions of dollars of extremely high margin revenue per year. I’m also interested to see how the partnership may work out in terms of Intel chips potentially being produced in Terafab - seeing as they would share the same process node - which may allow Intel to gain additional capacity if SpaceX have the tools but don’t maximise their utilisation.
That’s it for this week. Stay tuned for the next episode coming out soon which will explore Intel’s efforts in glass substrates and co-packaged optics!
For paid Intellionaire supporters below, I will draw some comparisons between TSMC and Intel Foundry in terms of valuation - I believe that with a market cap of $600Bn, almost all of Intel’s current valuation is coming from products/agentic CPU demand, with no Foundry success being priced in.