PCB

somewhataccurate

The answer to that question is the same answer to the question "how tf did they get AI to generate a good PCB?" because there is not a lot of good data available on board layouts.

ImplementBrief44

If you’re hunting picks‑and‑shovels, a few names jump to mind: power side – Eaton, Siemens, Itron, Vicor and Vertiv; cooling – Motivair (now part of Schneider) and Limbach Holdings; PCB manufacturing – TTMI; solar for data‑center power – First Solar. Do a 10‑K look to see how much dependence on a single client they’re under.

Napster-mp3

Miss seeing all you guys at MTV Spring Break in PCB

meagermotto130

The next earnings call is where you separate the pump from the actual business, I always go straight to the revenue breakdown first not the headline number but where it came from. If they announced a new partnership and the revenue spike is just a one-time licensing fee instead of recurring sales, that tells me everything I need to know. The gross margin is my second stop, a lot of these small caps will boost revenue by selling stuff at cost which looks great on the top line but means they're basically just moving money around Cash flow from operations is the real tell though, if that stays negative while they're supposedly growing then they're still burning through cash to make it happen. I've been burned too many times by the "record revenue" press release only to find out they diluted shares by 30% to fund it One thing I picked up from keyboard building actually, you can have all the fancy switches and keycaps in the world but if the PCB is garbage the whole thing fails, same deal with these companies. The catalyst is just the keycaps, the financials are the PCB

Y0tsuya

>Batteries, suppliers, platform design, automation, scale, parts count, logistics and margins matter too. And those also cost more to manufacture here. I'm in a tech company making networking equipment. The cost to fab and assemble our prototype PCB in California is easily 4x the cost vs China. Labor cost accounts for a large portion of it. We managed to shave the difference down to 2.5x by doing the PCB fab in Taiwan. Cost difference like this percolate up the supply chain.

Fortnait739595958

Just saw this: https://www.reddit.com/r/BuyFromEU/s/BzFeTaYzUJ What company/ticker makes those PCB components?

bryandph

This is straight regarded. I say that as an early investor in Unusual Machines. UMAC is just getting to the point of making their own motors Look at UMACs retail storefront, Rotor Riot. How many of the components are from China that they are reselling. Almost all of them. Fat Shark, RadioMaster, TBS, etc are all Chinese brands. Even their own brand of brushless motors, “hypetrain”, is still made in China. A huge number of the NDAA parts are Chinese, counter to the claims on their site. Maybe final manufacturing is happening in Orlando, but the components are certainly not. Another “Made is the USA” grift. Look at the components actually produced in the USA on their site, like their flight controller. Look at the RotorRiot F7 https://rotorriot.com/products/rotor-riot-brave-f7-flight-controller-made-in-usa . Now compare it with the Chinese equivalent they based their design on down to the microcontroller: https://www.speedybee.com/speedybee-f7-v3-flight-controller/ The Chinese version is of higher build quality literally down to the PCB. I own both. The Rotor Riot motor controller doesn’t even come with a heatsink. It’s literally an enshittified version of the Chinese version that costs more. Grifters gonna grift. I may like UMAC and think it’s a good idea for the USA to manufacture drones at home, but what is happening now is stock market manipulation, not building American business.

Qweasdy

They’re also a PC component manufacturer, power supplies , RAM sticks (don’t get too excited they don’t make the dram chips themselves, just design and assembly of the PCB from components) and cooling components. The peripherals side of the business came later, the original business was making components. Though I don’t quite share OPs optimism on them, as mentioned they don’t make dram themselves so aren’t benefiting from the high dram prices but will still be suffering the effects of their products going up in price. They’ll sell less at the same (or lower) margin as they did before. I guess technically their power supplies and cooling components could find themselves more in demand but I doubt it’s all that big a money maker. It’s already a big market with lots of competition, with most PC components being relatively easy to manufacture from mass produced components, margins will be small.

Samjabr

IMO, they are going to get boned. IDK about earnings, because that s\*\*t is always a coin flip. But look at the record: Vision Pro, Project Titan, TITANIUM!!! = all garbage Everyone is fired up they are releasing a foldable phone - only 7 years after Samsung has had one. And you know that thing is going to be $3,000 plus - why else would they partner up with a BNPL company and not have any of the financial liability in lieu of a small portion of profits (they learned from their partnership with GS on that credit card) The BNPL company was probably salivating at the deal, hoping to work with the 800lb gorilla - so, I doubt they negotiated the greatest terms. Their biggest recent success has been the mac mini and mac studio. The funny thing - **it was a complete accident**. Apple uses unified memory - which was often viewed as suboptimal. You couldn't upgrade a MAC after purchasing it because the memory is soldered directly onto the PCB - contrast this to a PC that has memory slots. You can usually add memory, or even swap out for faster memory down the road. There is one benefit that has had huge unforeseen circumstance: Apple unified memory removes the strict memory capacity limits of traditional discrete GPUs - In other words the CPU and GPU can share the memory) Enter LLMs - People realized that instead of paying for Gemini/ChatGPT etc, they can just load their own open source models, such as BookFace's Llama or the Gynese stuff. The problem is it requires massive amounts of memory. And with PCs, because of the memory limitations, you could only load models with a certain amount of parameters - And that's where the Mac Studio and Mini win. So people started buying them up like crazy once they realized this - You can buy a mac studio and spec it up to 128GB of memory and load a 70 Billion parameter model- Get the 512GB config and you can run **400 Billion parameter models** \- That's essentially one step removed form frontier models. Toss in an OWC SSD for swapping out different LLMs and you are running your own agentic machine. I had to use an automated script searching the Mac Refurb store to get one. The Mac Studio M3 Ultra (which is almost 2 years old) with 512GB memory used to retail for around $10,000 Now you are lucky to find a used one on ebay for less than $40,000 - anything cheaper will likely be a scam [https://www.ebay.com/shop/mac-studio-m3-ultra-512-gb-ram?\_nkw=mac+studio+m3+ultra+512+gb+ram](https://www.ebay.com/shop/mac-studio-m3-ultra-512-gb-ram?_nkw=mac+studio+m3+ultra+512+gb+ram)

TheRed2685

When people think of oil, I think they mostly only consider their cars' gas. Everyone tends to forget about plastic, how it is made from oil, and how it drives prettymuch everything that exists from simple storage, to single pack food stuffs, to medical equipment, etc etc. So would it hamper those stocks you mentioned? Lemmi think. NVIDIA: PCB boards are made primarily from plastic, made from oil. This would effect their bottom line across every device they manufacture. Eli Lily: every single use medical device they manufacture, they all are packaged in plastic, every dispenser pen and inhaler, made with plastic, which is made from oil. Berkshire: any company it is invested in that uses any form of plastic (or oil production, or any water investments) will take a hit, causing Berkshire to take a hit. I hope this helps provide clarity.

NegativeSemicolon

Stick to PCB’s and not licensing agreements 👋

Rbaseball123

I’m an expert in the PCB design field. What ever you think you are reading you are misinterpreting. Anyways have a great day!

SalemKinkTrials

PCE, PCB, CP3 don’t matter CALLS ALL DAY🚀

tke248

TTMI is a strong add. PCB/backplane/electronics infrastructure bucket. Grade: **B+ hidden shovel** — not sexy, but AI servers still need complex boards, backplanes, and high-speed electronics.

ReasonableBrother448

The cooling problem on off the shelf parts would be non trivial. Every smallest component in there could start to cook off as their only method for losing waste energy is conduction into the PCB substrate.

Effective-Leave-999

TTMI - They have had solid growth with past leadership and new leadership is already excelling in their current roles ( previous leaders retired not fired ). They are projected to have 20 % YOY growth. They work in space and defense with customers such as Raytheon, Lockheed Martin and many more. They are involved in not only military contracts ( examples include Javelin missiles , fighter jets, radar systems , and the golden dome ) but other projects such as the Artemis program alongside various other space programs . They are a leader in PCB fabs with a new plant that just opened in the USA. Most of the work is done in America coinciding with plants in China and Malaysia, 23 plants in total . The commercial side of the business is involved with everything from automotive to medical equipment and robotics. They are also an essential supplier for many AI businesses. The list goes on and the charts speak for themselves.

deltamoney

Gamers Nexus just had integrators on a whole expo. Literally showing their numbers. Are down 50% yoy. Corsair does not MAKE memory. They buy the inflated chips at the inflated price. Put their name on a nice looking PCB. GN literally had a whole expose on these companies being down mega hard.

37366034

It’s still a baby at $19Bil to be in the Vera Rubin ecosystems. Do you think we will need more or less PCB with the next NVIDIA roll out? If more, who are you buying? If less, don’t worry about this trade…

PsychologicalOne6234

TTMI is a clean thesis, ngl. Cheap-ish US PCB name tied to AI capex is exactly the kind of boomer stock that quietly 3x’s while everyone chases weekly NVDA lottos. Only thing I’d watch is how much of that incremental PCB spend actually flows to them vs the Taiwan guys, but if Blackwell / Rubin demand stays anywhere near projections this has serious sleeper potential.

bankshotting

Why are you talking like gold isn’t also a low impedance conductor for circuit boards and such—it’s very much so an industrial metal as well, if not as much as silver, speaking from someone who works in PCB design. I use just as much if not more gold at work than silver.

derprondo

I went to PCB for spring break in 2001, that might have legitimately been the peak just months before 9/11. I saw Johnny Knoxville jump off a hotel balcony onto a palm tree lol.

CaffeineAndCurves

Did my own checks. One correction worth flagging – the market cap figure in the OP is outdated. That $282M number reflects an older price point. The stock is currently trading around $8.10, which puts market cap closer to $440M. Still not huge for what the business is doing, but worth being accurate on. Also a few things have happened since this was posted that are worth knowing if you’re still researching: The company just announced (March 5) the completion of a processing technology upgrade at their San Rafael facility targeting printed circuit board recycling. They installed a secondary combustion chamber, surface cooler, and desulfurization tower. According to the release: PCB processing capacity is up 30%, gold and silver extraction efficiency improved 15-20%, and management expects gross margins to improve 8-12% on materials processed through the upgraded line. This matters because it extracts more value from the same input stream – better margin without requiring higher volume. Also worth adding: the November 2025 contract disclosures showed $7.7M in a single month – 634,000 kg of copper alloy ingots and 764,000 kg of aluminum. That gives a cleaner read on monthly revenue cadence than just the H2 aggregate that’s been circulating. H1 EPS was $0.0736 vs $0.0461 the prior year – not in the OP but worth having if you’re modeling anything. Not financial advice, still thinly traded and carries the risks OP outlined. Just making sure the numbers people are working with are current.

GrogRhodes

Also repairing is a thing. I’ve been repairing some old EVGA 3090s that went Red Light of doom that’s I’ve been snagging for cheap over the years. Most of the failures in these things are caught in the PCB. It’s time consuming but expect to see this become more common practice moving forward especially considering how efficient you can become once you figure / dial in the process

W4spkeeper

honestly with Corsair you just pay for the brand and special aesthetics/ecosystem shell. Their PSU's -> no manufacturing they slap their name on (made by CWT, Andyson or Seasonic) they may produce the PCB/heatsink for their RAM but don't make NAND chips. I think the only thing they do design and make the majority component layout for is their peripherals

hidetoshiko

Actually that's not really true. It is *incredibly* hard to make these things, especially the HDDs, just that the ROI was never really there to keep extending it, unless you're the incumbent with all the requisite organizational and tribal knowledge to build it. And no, the hyper scalers will never waste their money designing their own HDDs. There's a reason the supply side shrunk from dozens of players to just 3 today. For SSDs the barrier of entry is relatively lower and I have heard of in-house designs, i.e., they buy the NAND chips and populate a PCB of their own design, but still everything would be ultimately gated by NAMD wafer fab capacity, and greenfield fabs take anywhere from 5-10 years to build.

futurefinancebro69

Gold literally just hit $5k again. The trade already worked. If that's what I think is happening, then explain why gold bounced back after the "crash" while the biggest companies in the world (NVDA, AAPL, MSFT) are still building out AI infrastructure that requires gold in every connector, PCB, and thermal interface. Physical demand from tech didn't disappear. Data centers didn't stop being built. The 13% drop was macro noise, not fundamental deterioration. That's exactly what the Monte Carlo analysis showed: after a 2.6th percentile move, mean reversion was more probable than continuation. The stats were right, the fundamentals were right, and the trade worked. If you think structural tech demand for gold is irrelevant, I'd love to hear your thesis on why industrial consumption doesn't matter for commodity pricing. (Ur lost af fr fr)

Paul_Robert_

The thing is, you need a specialized facility to make the DRAM chips that are used in RAM. The actual PCB and assembly can be done basically anywhere, but the photolithography step is the limiting factor as you only have a handful of places that can do it. Combine that with a huge spike in demand induced by tech companies wanting more AI shit + other companies stockpiling in advance in case RAM gets even more expensive, and you have MU at 400. Honestly, it seems so obvious in hindsight, I wish I had stayed in MU 😭

Illustrious-Rush8797

You can probably restart making competitive automobiles like 20 years after your own industry is decimated. Do you want to wait 20 years and live with a destroyed manufacturing sector that might not even make it back after that time? I can tell you from a US perspective for example that many manufacturing industries moved to China exactly on this premise. And it's almost impossible (not entirely impossible) to reverse because they become entrenched with supply chain and synergistic effects in their new locations and that impacts far beyond just the industry in question. For example China now leads in drone technology because all the little thousands of industries that support drones (PCB manufacturing, etc...) all went to China. If the US wants to lead in that it has to build not just drones but all the little industries that feed it. Long story short you don't want a national industry to go through this because there's a good chance it's never going to come back. Even European countries have placed 100% tariffs on Chinese cars and they're not reducing them.

-colin-

In terms of quality, I feel like Gemini is way above ChatGPT, especially for technical questions. I asked Gemini to analyze a PCB for me and identify where the boost converter is, and it correctly identified components and the purpose of each one. ChatGPT in contrast told me that the voltage I was looking for isn't possible because "the wires are too thin". I am now seriously considering moving my pro subscription to Gemini.

zeromussc

Mr. Altman at OpenAI bought production capacity by committing to wafers. The wafer production that was bought by openAI is for the big thin sheets of semiconductor that is then cut up and turned into RAM. The intermediary material in these wafers can be used to make either DRAM or VRAM. The actual chip itself, and the PCB that they get attached to, matters of course. But the actual lithographed silicon that stores the bits and information used as part of a GPU or as system memory for a CPU to access, as far as I understand it, is the same. Open AI didn't go to SK hynix and say "give me your DRAM", they went to the RAM lithographers and said "we want your wafers", and then bought that production capacity. This means that when an nvidia board partner wants 16GB VRAM to go alongside the GPU they want to put on a board, they need to ask for RAM wafers that haven't been committed to openAI to be used in the production of the GPUs themselves. It's fucked all the way down.

TomatoSpecialist6879

I'm curious as to what's the bear thesis for gold aside from broke retards thinking it's too expensive and should fall, the price increment came from actual physical demand for gold because it's needed for PCB traces

Flipslips

SpaceX has the largest PCB factory in the world, that isn’t nothing

DopeBook

Well. All these chips (as in every chip on a PCB) that the AI monsters want have to be produced with various machines and consumable materials. One of the most integral components is the photo mask. It is the template for the design and simply put, it is used to “print” the silicon. They wear out, and they are especially vulnerable to dirt and wear in the EUV process that makes the highest of the high end chips. Companies like Hoya, Asahi Glass, Photronics (PLAB) make the masks.

ProbsNotManBearPig

There is not a single chip on the planet with any physical component that is 3nm. Those node names are pure marketing. Theres not anything smaller than ~20nm on the most advanced chips today. Source - me. I work in the semiconductor industry and look at scanning electron microscope images of these every day. Tesla designed the SoC here, which is not the PCB. It’s a monolithic chip combining multiple sub systems into one chip. They did do some actual chip integration design here. Just not what you think of as core IP of a cpu or gpu. Their SoC design is not advanced at all though and ultimately Elmo is trying to milk it on the marketing front. So you’re right anyways.

ProbsNotManBearPig

I mean, they’re making their own SoC, which requires at least micron level design. An SoC is not a PCB. It’s a fully integrated chip printed as a single monolith on a wafer that connects multiple components in a single package. The square with “Tesla” written on it is their SoC, not the full PCB. The sub components of the SoC are designed by various groups they license from, but integrating them into an SoC is Tesla’s design, right?

Suitable_Air_2686

Not to insult an PCB designers. But “chip” design and PCB design which Tesla does are too VERY different things. Real AI chip makers like Broadcom AMD, NVDIA, etc are doing transistor level 3nm or 5nm design. What Tesla does is grab a bunch of parts and put them together on a circuit board. This isn’t even like Apple which at least makes their own CPU units the A and the M series. You’re not “designing” chips unless you’re doing nm level designing.

29da65cff1fa

is this true? the actual memory chips are from samsung or hynix, but who puts them on the PCB, tests them, and packages? i always thought corsair did the last few steps themselves...

jensenskawk

LMAO There are AI CAD, AI PCB design startups already. Even Autodesk is all in on generative design. Meta's SAM3, Hunyuan3D, DeptAnything etc etc etc is augmenting the space. You have no idea what you are talking about.

ElectricRing

I have been doing hardware development for nearly 30 years, now I am doing ultra high performance test and measurement for audio applications. I do analog circuit design, my colleagues directly do PCB design, write firmware, digital hardware, software, as well as doing ERP and manufacturing the product. We also have customer support. It’s an electronics manufacturing business at the end of the day and we sell our customers solutions. When the people who do the actual work, not management and CEOs are saying AI is very helpful and improves productivity substantially, you may have something. Right now AI is a fun toy that costs a lot of money and is marginally useful for most business tasks.

Colest

Sales tax is a transparent % added to a purchase anyone can easily calculate beforehand, have flexibility by being exempt from certain categories of goods on a consumer level as needed, and business are exempt from them entirely. Tariffs hit early in the supply chain and (largely) do not change based on the finished product being made with the good/material. Raw steel and aluminum is taxed at 50% if it's used in the manifold of a bugatti, a rivet on a child's backpack, the PCB in a ventilator , or an iphone chassis. Likewise, the US Customs and Duties code along with the executive orders that modified them are incredibly vague because at the end of the day it's up the interpretation of a CBP chucklefuck stamping the paperwork at port of entry to determine what exactly consistutes the value of raw foreign material for an import and how far up the supply chain before it hits US shores that applies. With all this context of the nebulousness of rates, businesses in the middle of the chain don't want to sacrifice margins in the instance that their lawyers misinterpreted a vaguely worded statute from the White House so they take a very inflated estimate and price their margins accordingly and their customers don't know any better and assume the pricing is accurate and use their own inflated estimates to pass the costs on down. Every link on down the chain does the same thing which leads to higher prices than buying goods with a hypothetically equivalent sales tax. For the record, I am pro-tariffs for the purpose of allowing domestic goods to compete with cheap foreign labor; however, they have to be predictably steady and incremental in their changes over a long period of time (a decade or more).

GeoLogic23

Can you point me towards the state legislators pushing in the opposite direction? [THC Drinks Face New Restrictions as Florida Senate Advances Hemp Bill | FloridaStateCannabis.org](https://floridastatecannabis.org/news-18mar2025) >Restrict THC-infused beverages to 5mg per serving and limit their sale to **alcohol-licensed retailers**—shutting out cafes, wellness shops, and specialty hemp retailers **not licensed under alcohol statutes**. >Ban festival, street, and unpermitted sales, and **require event organizers to register vendors**, verify source compliance, and maintain testing records. I am not seeing anything that would restrict hemp beverages from being sold alongside alcohol at events. [Florida may limit THC in hemp drinks, require liquor licenses for sale](https://www.tallahassee.com/story/news/local/state/2025/04/01/florida-may-limit-thc-in-hemp-drinks-require-liquor-licenses-for-sale/82749055007/) >"....would require THC-infused beverages to be distributed by alcohol wholesalers and sold by retailers and **establishments that have liquor licenses.** Sherrie said his products undergo testing and are distributed by **Anheuser-Busch wholesalers**. He welcomed the proposed regulations. One of the proposals (PCB HAT 25-01) would require THC-infused beverages to be **sold by establishments with liquor licenses**. So pretty much as long as you sell alcohol you would be able to sell THC. Is there anything I am missing?

Competitive_Willow_8

Adding that fabs simply produce the wafers and dies themselves rather than the end product. The die cut from the water needs further manufacturing to say place the die on the PCB, integrate the controller, add higher speed memory as a cache, etc. This is necessary to differentiate the product for use in data centers vs pc or mobile devices with differing requirements. It’s not that TSMC produces the whole product from start to finish, it’s just that they manufacture the most important piece. And, Nvidia, and other chip companies still need to get the backend right either internally or through contract manufacturing. Even if the manufacturing is outsourced, they still need QA facilities for testing the quality or the different shipments they receive. Tldr; Nvidia and AMD and other fabless chip companies are not solely r&d and sales centers. They also have their own teams to manufacture the end products and ensure their quality

ak1111111

All the electronics distributors are all global businesses. All of them help designers in the US, but helping deliver chips to other countries for production. They all have sites in Asia and Europe. Mouser stores products in duty free zone in Texas and ship globally. I'd think digikey is the same case I would think this tariff hurts PCB assembly house and manufacturers in the US. It's also really difficult to execute. I get your point, but I just think there could be more delicate approaches to this rather than just say 100% on everything. On the other hands, I'm pretty sure all of major chip design company already have some type investment in the US and they can easily announce some type future investment and doesn't change any behaviors.

flamingtoastjpn

That’s a machine to make PCB’s, chips are a lot (truly, and almost incomprehensible amount) more complex than a PCB. One high-NA EUV machine costs almost $400 million

ak1111111

There almost no impact of this tariff. Almost no one imports semiconductor or chips by itself. At the best, some manufacturers imports assembled PCB. It's also really difficult to assess how much semiconductor are used in each assembled parts and goods. I don't see any practical way to implement this.

MayIServeYouWell

This will destroy US PCB manufacturing and assembly.  The only reason people import semiconductors is to assemble boards.  It takes years and millions of dollars to switch fabs to make a chip, and that’s only if an equivalent process exists in the US. Nobody will do that, ever.  Instead, the companies that buy chips will move their operations overseas. 

PM_ME_UR_DRAG_CURVE

Anyone remember the Eagle->KiCAD transition everyone in PCB industry did after Autodesk fucked up the license? Now imagine it in EDA space courtesy of China and TACOman.

Responsible_gambler

Still in the beginner phase, although need to get back to personal projects (PCB design, embedded programming stuff) while I have the summer off of school

Alive_Worth_2032

> For basically every purported use case of humanoid robots you need to ask, "is the humanoid form actually an advantage here, or is it just that only humans are still capable of this task?" Exactly. The humanoid robotics fan crowd would have solved manual device soldering and assembly of the past, by replacing human assemblers and solderers with humanoid robotic ones. Rather than change the whole supply chain to suit pick and place machinery and wave form soldering. Where no human barely/if at all touch the PCB until fully populated with components.

generally_unsuitable

I don't even understand your question. I have a small electronics business in the US that requires me to import PCBs and other components from China. I order almost everything from China that I can because the price differential is enormous. For short quantities, a PCB that might cost $10 in the US will cost me 50 cents or less in China, shipping included.. Same with things like switches and audio jacks. A 3.5mm jack that costs $1.15 here costs 4 or 5 cents in China. For me, even a 145% tariff still doesn't dissuade me from ordering from China.

PM_ME_UR_DRAG_CURVE

I hope this kicks off a shift to open source the same way Autodesk triggered the shift from Eagle to KiCAD for PCB designs by buying Eagle and fucking with the license.

henryx7

I dont know why it's so hard for people to understand this. "Cheap shit from China" is the complaint about the cheapest stuff coming from China. The fact of the matter is they make products up and down the value chain, likely in the same factories. They make the cheap microwave, they make the expensive microwave. They even make the plastic that becomes part of the exterior of the microwave. The reason China is a manufacturing powerhouse isn't because of cheap labor and 1.4B people. It's because they have the entire supply chain in one place. Your PCB needs buttons? How far do you need it to click, how much force do you want for it to click? What kind of package do you need it in, pins, surface mount? It's not just a bunch of people sitting at a long table with a screwdriver in their hand.

Cruezin

There are exactly 2, yes, TWO manufacturers of cameras- you know the security camera that every military, government, building, etc have everywhere- in the USA. One of them is Arecont, I forget the name of the other one. They cannot meet the demand, but you can imagine that with security requirements at secure government facilities alone, there is no way those two companies can handle the demand for everything else (it's why those cameras, which are top notch btw, are so expensive). A TV? HA! Even Samsung, LG, and Sony don't make their own panels anymore, they got pushed out of the "lower end" market by the likes of BOE, etc- and they are ALL made in China. Heck even several of the higher end model panels are also built in China. And the chips for those TV's? You guessed it, Samsung buys the vast majority of those 3rd party. It's just cheaper, even considering they have captive manufacturing of MCU's etc. And where are they manufactured? Taiwan, or... China. The PCB, and all the connectors? China. The plastic and metal frames? Yup, China. Most of them are assembled in Malaysia, but assembly is kinda all over the place. This is where things start getting confusing with tariffs. It's a global market, and we've basically bowed out of it. Stupid- the world economy is symbiotic, not one side wins one side loses. /Endrant

Googgodno

Imagine, A small PCB in a random electronic device you have at home or office. Imagine a random part in your car that needs replacement. Imagine a drug that you need, whose precursor chemical comes from China. There are a lot of chinese widgets that go into final products made by rest of the world. CATL batteries for model 3 is from China.

gibson486

Integrated circuits? You mean if I make my own PCB? Or are they just saying ICs?

Johnbolia

I have heard this 'fact' repeated many times. It is a pundit fact and a massive oversimplification. On final assembly items where margins are tight, this 'fact' is probably true. iPhones, sneakers, etc. However, the Chinese have monopolies (or near) in many industrial components. I know LiPo batteries and PCB fabrication first hand. Even if the prices went up by x5, they will still own the market for several years, the alternatives to China are just far too expensive.

rasselbido

Apple will have the choice between sacrificing its US market, and buying and selling parts including its valuable chips in a tariff-free world OR Stay in the US market, lose your main fab service provider, lose your main PCB service provider, lose your main display provider, lose your main battery provider, and lose a promise of market growth from third world countries getting richer and buying iPhones

lpan000

Most of chips are not sold by US.. by US corporations not manufactured in US. A lot of chip assembly and PCB manufacturings is more China dependent. And a lot of electronic components are manufactured in China.

retroclimber

Lumber and steel being more expensive reduces housing inventory by making it more expensive to build, raising rent and home prices since there is less supply. I work for an IoT device company and we will have to raise our prices by 54% because of the increasing tariffs on China. The world’s PCB, metal, and plastic are made in China and that is not changing anytime soon due to the price of labor over, the availability of raw materials, and the presence of the world class factories. It would take 50 years to reconfigure the USA to be like China. We also simply don’t have the population. Did you know the economy is connected to raw materials? Or do you think supply side jesus just creates everything out of thin air for your own personal consumption? Personally, I’d rather our country work on the office side of things making 10x the income than doing repetitive manual labor for $4 an hour.

Johnbolia

I think a problem here is that you, and many others who discuss this, think of an automotive assembly line when they think of manufacturing. An assembly line which makes one product for one company with minor variations. If you take PCB fab and assembly. A typical factory will have hundreds or thousands of companies as customers. The machines will make an entire years worth of parts in hours or a few days. The process is automated for a customer, but the labor is in the setup cost. This will not be automated any time soon. This will be a similar story for machining, plastic injection molding, tool and die shops. I don't know the economics for manufacturing individual electronic components. But in order to use these, they normally need to be incorporated into a PCB. Mechanical assembly can be automated, but this is super expensive and is only suited for products with assembly runs in the millions, this is a very small minority of products.

RGrad4104

I just bought a toaster oven...to solder PCB boards because the assembled ones ordered from china just became to expensive without de minimus. Please let us stop winning already... American manufacturing sucks with every jackass wanting a six figure salary. The same 20$ circuit boards from china cost 400$ to produce in America... a few factories can't bridge that friggin financial divide...

the-retrolizard

I've sort of assumed the big unemployment numbers we'll get might lead to aggressive rate cuts but maybe not. I also think on-shoring is optimistic. Some might take place, but the cost and time horizon to build factories and train and hire labor, especially skilled labor, probably won't be an even 1:1. Plenty of companies will be happy to just raise prices to continue hitting their profit margins. Either way shit gets more expensive by the time it hits the shelves, unless they nuke minimum wage and there are literally no alternatives for employment. Trades have been dying for a long time, and I'm not sure how many people will be itching for their kids to go work in the lithium battery or PCB factories. A big part of the allure of the trades is owning your own business down the road. That doesn't translate to large-scale manufacturing. I also suspect the deficit is an economic boogeyman, given that he cranked it up last time. Would love to be wrong about that one tho.

Alex_Trenholm

5. ARRY. Are you going to keep that bet for government incentives and renewable energy adoption ? PCB looks good though.

WatLightyear

The 5090 is ~30% better because it has ~30% more of everything on the PCB and also consumes ~30% more power. It’s not a serious upgrade at all, unless you love fake frame generation instead of raw rasterisation.

VNG_Wkey

Ya it is much less likely that Open AI becomes a direct competitor than it is for Nvidia and Open AI to partner. Nvidia benefits from the advancement of AI because they use it for things like modeling new PCB's and chips. With an AI doing that work they can try every way imaginable and find the most effective way to build on the new architecture. Open AI benefits because these new chips from Nvidia will give them vastly more compute power.

mangodrives

you wouldn't know them by name, but 1. PCB (going into MLB) 2. post-process packaging company 3. dry vacuum pump (this one is really fcked) 4. I don't even know what they do now. Gave up on this one.

No2reddituser

>car assembly plants. They don’t manufacture, they assemble. Semantics. Assembling piece parts into an entire product is manufacturing. >Yes, i buy a phone every 4 years, i am okay with that. Ok. most Americans are not. They got used to their cheap Chinese electronics. Tripling the price on these will cause a revolt, which maybe is a good thing. And don't say, "well, we will just make them in the U.S." You don't set up a large electronics manufacturing line up overnight. And there are only a few PCB vendors left in the U.S. Most high-volume PCB fab is done in China.

polaarbear

This is such an ignorant view it isn't even funny. My wife works for a small business. They import small PCBs to assemble an electronics product by hand in the United States. It's not made here. Can't be, won't be any time soon, a custom processor mounted to a custom board. They have 5 employees. They aren't drop-shipping anything, they engineer and design a new product every two years and sell it on American shelves. They buy that PCB from China for like...50 bucks. Then they sell it in a product here for hundreds.  80% of the money from that item is handled and stays in America. But now? That price is gonna go up by 30 bucks to cover the tariffs. And you know what happens when you raise your price by 30 bucks? You sell fewer of them. It goes out of some people's price range. But you owe that 30 dollar tariff to the government for every purchase.  That's not extra profit for you.  Your margin is the same as it was yesterday. But you just sold fewer items due to the high price, so you lost money.

the91fwy

Yes but NVIDIA fabricates all of their chips via TSMC in Taiwan and then ships the chips to a factory in China for final assembly onto the PCB. The CHIPS act is what helped us secure the TSMC fab in Arizona (and Samsung's new Taylor, TX fab). Fabrication is half the battle; and we've only made a small amount of progress on that, but there's very few PCB assembly factories in the USA so even when we make chips here (next iPhone SE is having the first TSMC made in USA processors) you still have to send those fabricated chips to another country for PCB assembly. NVIDIA could move more PCB assembly to Taiwan, India or Vietnam potentially provided there's enough factory resources to dedicate to NVIDIA as a client. But at the current moment there's not too many great options for manufacture outside of China and a repeal of the CHIPS act will just hamper that further instead of advance American manufacturing.

justacrossword

> A regular Nvidia H100 GPU, cost approximately $3,320 to manufacture from TSMC. However it is being sold at $25,000 to $30,000 each! This is stupid math. You are supposing the cost of a single chip. Then you pretend thinks like HBM, PCB, NVSwitch, FPGA, 900 GB/s GPU-switch lanes, and every other ducking thing is free. Then you assume a list price that you find on the internet for a stay card.  There is a lot more that goes into an H100 system than a chip.  NVIDIA’s margins are great, but they are nowhere close to your magic math.  Also, a lot of that CapEx is building and upgrading data centers.  Intel just changed their AI strategy again. They have been changing their accelerated computing strategy since 2012 with nothing sticking in the market. But I’m sure they got it right this time!

Bozhark

Quantum data processing. Massive leap in storage capacity  Massive leap in processing power and speed (not from quantum)  New manufacturing techniques for PCB designs  Drone technology of the wars start 

ReasonableAthlete636

PCB Board

iamwhiskerbiscuit

Did you even listen to the last Powell talk? He was very clear that the economy was strong and wanted to cut rates proactively because inflation has come down close to their desired levels. Reiterated multiple times that the economy was strong, and their primary motivation for lowering rates was reducing mortgage payments for homeowners, and secondly wanted to cut rates to strengthen job growth. The 50 PCB cut was unexpected, but like 100 politicians signed a letter asking the Fed asking for a 50 PCB rate... So it wasn't a surprise that they gave the people who elected them what they wanted.

iamwhiskerbiscuit

The stock is up over 50% this year. The market looks like it's forming a top and the 50 PCB rate cut that was expected to blow us past ATH didn't. The market could fall quite a bit from here and institutions aren't interested in averaging up on their positions right now. This is an overall market trend. Asset managers are more concerned about the near term future of the S&P than Costco.

NextTrillion

Is your kids phone the iPhone 7 Plus by any chance? Tbf, that phone was garbage for a variety of issues. People were opening them up and jumping the PCB with incredibly thin copper wires. Much thinner than a human hair. Side note: I tried to type “iphone 7 plus,” but my iphone autocorrect was insistent on capitalizing it (as “iPhone 7 Plus”). Seems a little bizarre to me.

myironlung6

calls on mercury and PCB toxicity

ImpossibleChemistry2

Copper as an engineering metal is incredible. It is *the* preferred material for electrical conductors in PCB's and electronics everywhere; no figures but the amount of copper we as a planet utilise each year must be incredible. The fact that some medals were 'bad quality' would not mean copper is now less useful, or supply is more limited. In fact, it likely only reflects badly on the medal supplier who used the wrong type of bronze or wrong protective finish.

Botlikereflexes

Nope. People got hyped up over motherboards, PCB’s like these were GPU’s. Shit is not cutting edge and that profitable.

BaronDavis12

For those trying to catch a value play, TTMI (TTM Technologies) seems to fit the bill.  First Quarter 2024 Highlights Net sales were $570.1 million GAAP net income of $10.5 million, or $0.10 per diluted share Non-GAAP net income was $32.8 million, or $0.31 per diluted share Cash flow from operations was $43.9 million Repurchased 600,000 shares of common stock for $9.3 million at an average price of $15.56 per share Market cap of 2.28B TTM Technologies, Inc., together with its subsidiaries, manufactures and sells mission systems, radio frequency (RF) components and RF microwave/microelectronic assemblies, and printed circuit boards (PCB) worldwide.

ThirdLast

To be fair I believe a piece of software decides the most appropriate routes for traces on the PCB but that's probably as much of the process that isn't engineered by a human.

augustusSW

No real military (only US in the world currently, maybe China and maybe is doing massive lifting since they haven’t done a single thing except take potshots at the Philippines) would have trouble against drones. I don’t care if they are FPV or the shaheds. The only reason Ukrainian civilians are dying is because the world didn’t give them enough resources to stop it. You are watching what is happening because we chose to let it play out that way not because we can’t stop it. Who do you think sells the passive and active PCB components to both Iran and ruZZia? 70+% I’ll let you guess.

NickSicilianu

Yea, I believe the acquisition happened like 6 years ago. I have been programming and designing PCB around Atmel MCU for 15 years now, it’s my favorite brand. I do love their ATXMega series, it provides DMA for asynchronous IO and their complexity is far less than a ATSam ARM versions. I am a AVR junky. Recently I had to switch to EsspressIf because sadly Atmel have nothing comparable with WiFi and Ethernet. I was using ATSam4E16xx for a main controller board that was powered via PoE, but they have no WiFi options. Also lwip library overhead started give problems with missed frames on busy network where port isolation wasn’t available. I believe the CPU can’t jump fast enough to the ISR and free the DMA channels for the EMAC peripheral. We had similar problems with a STM, and the ESP32 seems to handle the lwip and embedTLS very well. I bought the ATSamA5D3, as it can run Linux and should handle TCP stock and TLS, but haven’t had time to try it. Downside is this is significantly more complex hardware design then drop an ESP32 module on a PCB, and way more expensive BOM. I really wish Atmel do make an ESP32 equivalent option. https://preview.redd.it/n81eb5euatyc1.png?width=3024&format=png&auto=webp&s=1585bff43ed44cf07dc902d74477086bc9179d59

NickSicilianu

55 shares. I mean, worse case can sell off and be a tax write off, and move the money elsewhere and return to this later, if the CEO pulls his head out of his ass and start to take advantage of the massive capital they sit on to expand. I would dive in microcontrollers and expand to sick military contracts if I was in any position to make propositions to Intel. They had a few mobile or robotic oriented processors optimized for Linux, cherrylake if I don’t mistake, I was looking into designing a PCB for a client around that board. And right after the Chinese came out with a latte panda single board computer using one of those processors and run windows 10, aiming at robotic and kiosk applications. But idk, it kinda died, and price wise there are more efficient ARM processors when it comes to embedded systems applications. Intel definitely need to do more if they want to grow. https://preview.redd.it/v40kfdccqgyc1.jpeg?width=1284&format=pjpg&auto=webp&s=627533a328ba409011042b2cd184aa2e5f2d50e6

NickSicilianu

I hear you, I am in the same boat. Electrician for 10 years, roofer for 2, and painter for 1. Machine operation for Pick and place machines at a PCB local factory here in FL and never earned a decent amount in my life. Zero retirement savings, no healthcare insurance, and all my saving which I invested into my startup, that I lost my investors due to my dad being diagnosed with cancer and me not able to progress the project. So I understand, yet, I wouldn’t give up my dignity and honor for a lousy million dollar, my dignity and honors is all I got left at this point.

Future_Gain_7549

I work in tech, hardware security. It's a con. The PCB looks like an Amazon Echo. What a joke.

Botlikereflexes

INTEL is getting into the game. Pissed off seeing PCB manufacturers run up.

Araniet

My other play (29.4) is LSCC. They do chips for PCB's used in Data Centers.

Lobbel1992

Anyone here who invests in ELTK? The stock dropped after a capital raise. Nice balance sheet + outlook. They make PCB for defense in Israel and companies in the world.

ResponsibleJudge3172

There are thousands of suppliers other than ASML for TSMC. Not to mention Nvidia themselves rely on other companies to make their packaging (like Intel but also TSMC), memory, PCB, etc If Nvidia is selling shovels in a gold rush, TSMC molds steel into their current shovels and ASML sells TSMC smelting furnaces.

BoyKai

What I’m saying is the market doesn’t understand the domain. This has nothing to do with emotion btw. I work in tech, ML/AI for almost a decade - so I understand the domain marketplace better than most on Wall Street. It’s not that AMD hardware is defective. It’s that they are locked out of the domain due to adoption and user lock in of higher level services in the stack required for development of ML/AI. Basically from the PCB (hardware) up to Python(software layer a human interfaces with) Nvidia controls the keys to the kingdom. Go check out CUDA, the middleware. PyTorch, TensorFlow, Kerras - all uses CUDA.

BoyKai

What I’m saying is the market doesn’t understand the domain. This has nothing to do with emotion btw. I work in tech, ML/AI for almost a decade - so I understand the domain marketplace better than most on Wall Street. It’s not that AMD hardware is defective. It’s that they are locked out of the domain due to adoption and user lock in of higher level services in the stack required for development of ML/AI. Basically from the PCB (hardware) up to Python(software layer a human interfaces with) Nvidia controls the keys to the kingdom. Go check out CUDA, the middleware. PyTorch, TensorFlow, Kerras - all uses CUDA.

quuxquxbazbarfoo

I could probably make one myself. Lightweight PCB. Write a chatgpt hook (probably a couple dozen lines of code at the most), interface to the audio driver, wifi driver, flash chip, memory. On off switch. What am I missing? It literally would be the same production costs as todays bluetooth toys. Or just use bluetooth, that's what they're all using today anyway. All you're adding is a few lines of code to send prompts to chatgpt.

rankaleefanclub

>This is because when you make a chip out of chiplets, if one of the components goes wrong you don´t have to throw the entire thing away. Hence, yields are much higher than in monolithic architectures. Patently untrue. It is standard practice in the chip world to laser off sections of the chip that are unusable/not up to specification to produce lower tier products. It's called chip binning. A lot of the consumer hardware sharing chips with the enterprise hardware is often a binned version of the enterprise chip. You could make the argument that Nvidia doesn't sell its H100/H200 (among others) silicon at the consumer level, and you would be correct. However, consider Nvidia's gross and net margins - in spite of wastage, the company is still raking in cash hand over fist and will continue to do so for as long as CUDA is viable. The questions to ask then are: 1. Just how good/bad are Nvidia's yields? 2. How much money is Nvidia losing on these bad chips? (clearly not much, given their huge margins) 3. What kind of hurdles can we expect in newer process nodes? How will they drag down Nvidia's balance sheets? You say that the industry has moved to pytorch. However you seem to have made a huge misinterpretation on what Pytorch is and what CUDA is. CUDA is not an AI framework. It is an API used to program shader processors to work Nvidia's graphics hardware. Instead of programming graphics shaders, CUDA is used to program general purpose compute tasks in a highly parallel manner. This language can be used literally on any Nvidia chip starting with the G8x series of chips released all the way back in 2007-2008, all the way to today. So the backend are guaranteed to be mature and mostly trouble-free, and developers with sufficient experience should be easier to find. At that point, if you need CUDA you have no choice but to buy Nvidia chips. By comparison, Pytorch is a higher level framework that sits on top of CUDA and other lower level APIs. So for the customer deciding between AMD and Nvidia chips the following considerations will be made: 1. Is it more efficient to roll our own AI technology or to use Pytorch or some other library? Most of the times, Pytorch or its peers will make a strong case for themselves. 2. Now that we know what we want to use, do we buy Nvidia or AMD hardware? Over here the customer, depending on their position in the value chain, will consider performance, price and space. The order of priority will depend on their needs. * For example, a company making huge datacentres will want to fit the highest performance in the least amount of space at the lowest cost possible. * A company renting hardware from one of these datacentres to develop AI will only care about the performance:cost ratio, because they can just rent more capacity as needed. Companies that run AI-assisted services but don't do any in-house AI development will likely go with the cheapest options that service their needs. In this space I believe AMD and Nvidia may both be at a disadvantage because Google already has excellent AI inference chips and are renting them out to clients via their cloud services. Amazon and Microsoft have also made significant investments in this sphere and have deep pockets so they cannot be ruled out in the long term. However, in spite of all this, Nvidia still will lead because it will take a lot of work to actually produce products that can comprehensively beat Nvidia at their price points. Another thing people seem to forget is the performance bottlenecks that chiplets introduce. When you have multiple compute chips tied together using a chiplet design, you necessarily need to wire them to each other either via a PCB or via another chip (as is the case with AMD's Infinity Fabric). This can actually cause noticeable bottlenecks in inter-chip communication. Given that GPUs are by design highly parallel and AI LLMs necessarily take a lot of parallel processing to be trained efficiently, the chiplet design may or may not be the best way forward.

ColdStoryBro

Its beacuse the soft modifications they make to weaking the chip can be overriden. Possibly by firmware, possibly by extracting the gpu and reinstalling it into another high memory PCB. The chinese are plenty capable enough to do that. If they want to sell to china its got to be from the ground up a design which doesnt have AI acceleration, or has a in silicon theoretical hard cap.

Educational-Year4108

I got out of Bayer and took the L. The US waited till some foreign company was dumb enough to buy Monsanto. Then the lawsuits started rolling in. Now they lost a PCB case which Monsanto produced in the 70s. The school was warned in the 90s and Bayer gets sued.

Fmbounce

A great pipeline? The stock just dropped because of inefficacy of their late stage blood thinner. Its current blood thinner - Xarelto - is set to lose European patents in 2026. Then you still haven’t added in the litigation pending on PCB and glyphosate. You give no reason why it can just magically overcome these issues.

torrrrrgo

Quote: * *"According to a spot by our sister site, Tom's Hardware, the workers could repurpose these components onto a "special reference" PCB with a blower-style cooler. Tom's Hardware states that the entire process is similar to "organ traffic" since the factory only preserves the GPU and GDDR6X chips during disassembling, dumping the rest of the components to second-hand markets."*

FarrisAT

The HBM alone is a significant amount of die space larger than the H100 GPU die alone. ASML scanners produce everything you see on the H100 except the PCB. As for the logic that this is a quantity not price issue, ASML & TSMC & Intel & Samsung all raise prices when they can.

Golfuckyerself

I’m hostile… as a rule this is the internet… it’s a hostile environment. Very genuinely; I work in electronics, PCB population and manufacturing but I’m in the supplier side equipment. If you can think of cool electronics my company is almost certainly involved. It’s been busy a long time, we don’t have enough bandwidth to keep up. I’m literally heavily burdened since mid 2020. No fluff. And there’s no sign of slowdown.

AReallyGoodName

I don't see that as a blocker to quickly pivoting. Yes the A100 has different packaging, a different PCB and more memory but none of that has supply constraints that prevent them quickly pivoting towards this opportunity. The point is that Nvidia don't have to ship 10x to 10x. A common misconception on this topic. They just need to pivot.