Okay... just some more weekend shower thoughts about $XFAB.
I still feel like it could be the next $TSEM, just early stage at a $1.4B MC?
They kinda leapfrogged current gens (which $TSEM are getting volume from) to compete for H2 2027 CPO scale up inflection point ($ASX docs cite Xfab (aka. photonixFAB) as focusing on CPO)
By building out some black magic MTP (transfer printing) architecture for lasers w/ other stuff like TFLN.
Basically next-gen integration IP, they're still behind on yields, sure.
But $NVDA evaluating it for transceivers/switches to see if it can volume ramp. That $NOK sets the specifications/assembly for. (nvidia invested in nokia for this these switches/networking too btw).
And if their MTP supply chain works... (eg. with Smartphotonics providing lasers, EU players doing assembly).
It basically volume ramps with $NVDA just like why Nvidia signed long term agreements with $TSEM?
Downside risk?
Already below replacement book value, can always go lower yeah, but typically to a certain point.
Maybe more CHIPS act subsidies next few months from chips act 2. If it doesn't go well there's SiC (152% Y/Y Growth, 195% Y/Y SiC wafer shipment growth)/GaN power semi upside.
Europeans /LLMs will say "oh evaluations doesn't mean it's a future contract!".
This is kinda different since the European Union is behind this effort and $XFAB for soverign photonic supply chains.
Not your typical company + hyperscaler evaluation, since $NVDA wants to be nice to Europe's regulators. They'd prob be pissed if nvidia just stayed in US/Taiwan/China.
So if they can make this MTP black magic work with mass production, feels almost for sure nvidia/nokia volume ramp on some tiny $1.4B silicon photonics foundry or at least throw them a bone with smaller contracts.
In terms of timelines, maybe just a months early since it volume ramps H2 2027/H1 2028 (which happens to be in line with CPO scale up timelines)...
Or just unknown because they named their project something stupid like photonixfab?
Like XFAB Photonics would have been better? so institutions/screeners can connect the dots when looking at CPO silicon photonic foundry players?
Automotive should also coming out of a slump medium term, sped up by self-driving (TSM Chairmain comments yesterday said ai automotive was TSM's growth vector alongside robotics). So their core business also should pick up speed too medium term.
Obviously markets/europeans want a "Nvidia signs $2B+ contract, XFab volume ramping 2027!"
But by then it will be a $9B+ company and you miss out on all the upside. And especially since everyone analyst/institution is blind to volume expectations for these....
Normally don't invest in companies in evaluation stages, but this just seems very de-risked by EU sovereignty + Gov backing, and you have Nvidia + Nokia there for volumes if they can make the IP work.
I think markets are probably missing something here... there's almost 0 value being assigned to being CPO exposure in Europe as their long term upside.
I mean I can do it for you if you’d like but all you have to do is paste your tweet into cc and tell it “how can I do this using Ultracode? I am trying to learn how to implement dynamic workflows into my routine, so please be as explanatory as possible. If dynamic workflows are missing a feature to be able to do something mentioned, please be transparent and honest about it”
@ClaudeDevs Agents mode has a lot of glitches still:
Worktrees show as part of the overall repo instead of a worktree
The whole page randomly starts to flicker with every key hit on the keyboard.
There is one little/medium problem lately. FSD slams the brakes for school buses randomly for no reason.
You can see it going up to 9mph and then slam the breaks down to 5mph as the school bus comes.
This happens all the time.
It happens even more if they have a left/right turn signal on, but it even happens without a blinker on at all.
It obviously does well to stop when the bus has the stop sign open and lights on, but it is way too sensitive. It slams breaks for school buses
@ashen_one Haha i got a similar schizo Claude on Claude Design.
My hypothesis is they prob tried to trim the system prompt with less instructions that explain what system reminders are etc, maybe hoping it would be accustomed to them by now from training, but it's not.
Hey be careful the command to give to Claude can be silently broken sometimes, I suggest using .zip, or, make sure to read the thinking blocks and command outputs. Because after you paste the command, it uses its fetch command and sometimes it does not retrieve all of the data from it.
@mattpocockuk This stems from them intertwining “permission to run commands” and “go ahead and implement/write code”.
There should be an “auto ‘plan’ mode”, where Claude can run commands but understand we are planning/brainstorming.
@trq212@bcherny