SK Hynix Indiana plant: AI memory supply chain shifts

Samantha: Welcome to The State of Tech, The European Edition, Friday August twenty-eight, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: SK Hynix breaks ground on a four-billion-dollar memory plant in Indiana, more than a hundred organisations sound the alarm on AI-powered cyberattacks, the Trump administration weighs sweeping new chip tariffs, a US university teams up with Autodesk to build houses with robots, and to close out two things you can actually try today: the encrypted European messaging platform Rocket dot Chat, and the free open-source home dashboard Dashy. Let's start in Indiana.

SK Hynix breaks ground on its first US memory plant with a four-billion-dollar bet on AI chips.

Samantha: South Korean memory maker SK Hynix has just broken ground on its first ever US production facility, in West Lafayette, Indiana. The investment is more than four billion dollars, and it is aimed squarely at one thing: the specialised memory that sits next to every serious AI chip. The cleanroom is scheduled to open in October 2028, with mass production of the next generation of this memory starting in the second half of 2029. Around a thousand jobs are expected on site.

Bob: And this is not just any memory. High Bandwidth Memory is the stuff that pairs with Nvidia and AMD accelerators inside AI data centres. Without it, those chips simply cannot feed themselves fast enough. Right now SK Hynix and Samsung dominate that market, and almost all of it is made in South Korea. Moving part of that supply chain physically onto American soil is a strategic shift, not a routine factory expansion. The wafers will still start in Korea, but the advanced packaging, the tricky final step, will happen in Indiana.

Samantha: The timing is not accidental either. The US has been pushing hard, through subsidies and pressure, to bring chip production back home, and this fits that story. It also gives SK Hynix a hedge against exactly the kind of tariffs we will get to in a moment. For Korean-US industrial cooperation on AI hardware, this is a milestone. Whether one thousand jobs really transforms Indiana's economy is another question, but the symbolism matters to both governments.

Bob: What is worth noting is how long the timeline is. Mass production in late 2029 is three years away, and the AI hardware cycle moves fast. By then we will be two, maybe three generations further in accelerator design. So this plant is really a bet on the assumption that demand for high bandwidth memory keeps climbing right through the end of the decade. Given how deep the current AI capex wave runs, that is a reasonable bet, but it is a bet.

Samantha: For European buyers of AI infrastructure, from cloud providers to research institutes, a more geographically spread memory supply chain is quietly good news. Today, if anything disrupts Korean production, everyone feels it at once. A second major node in the US eases that single-point-of-failure risk. Europe itself still has no meaningful high bandwidth memory production of its own, and this announcement underlines that gap. Expect it to feed back into the ongoing debate in Brussels about whether the European Chips Act is thinking ambitiously enough about memory, not just logic.

More than a hundred organisations, including OpenAI and Anthropic, call for global action on AI-powered cyberattacks.

Bob: Over one hundred organisations from across the tech industry, including AI developers OpenAI and Anthropic, have signed an open letter urging governments and companies to work together on defending against AI-powered cyberattacks. The core message: there is a narrow window to act before attackers using AI at scale become significantly more sophisticated than the defences we have today. The letter asks for coordinated intelligence sharing, funding for defensive research, and more responsible access to frontier AI models.

Samantha: The concern behind this is concrete. Security researchers have been warning for a year that generative AI lowers the cost of writing convincing phishing emails, adapting malware, and probing networks automatically. What used to require a skilled human attacker can increasingly be done by a script wrapped around a capable model. When the companies building those models themselves put their names on a letter like this, it is worth reading as more than PR. They are effectively saying: the offensive side is moving faster than the defensive side.

Bob: The letter is careful in its wording. It talks about testing processes, model access controls, and shared threat intelligence between governments, cloud providers, and AI labs. What it does not do is call for specific regulation. That leaves a lot of room for interpretation. Cynics will note that voluntary industry letters often precede attempts to head off harder rules. Others will point out that international cyber cooperation genuinely is a mess, and any push to unify it is welcome. Both can be true at the same time.

Samantha: For Europe, this lands in the middle of ongoing work on the NIS2 directive and the AI Act's provisions for high-risk systems. European cybersecurity agencies have been asking for exactly this kind of cross-border intelligence sharing for years. The letter gives that push political weight. Expect ENISA and national CERTs to reference it in the coming weeks as they argue for tighter coordination between EU member states and with the US and UK on AI-driven threats.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

The Trump administration weighs sweeping new tariffs on semiconductors, with industry warning it could hurt US AI ambitions.

Bob: The Trump administration is reportedly considering a new round of broad tariffs on semiconductors. Details are still being worked out, but the shape being discussed would tie relief from the tariffs to whether a foreign company is investing in US chip manufacturing. In other words, build a fab in America, pay less. Do not, pay more. Industry groups are already warning this could raise costs across the AI supply chain, and paradoxically slow down the very US AI leadership the policy is meant to protect.

Samantha: It is worth noting these are proposals under discussion, not signed policy. But the direction of travel is clear, and companies are already reacting. The SK Hynix announcement we just covered fits neatly into this logic: build in Indiana, and future tariffs become less of a problem. That is not a coincidence. Whether this approach actually strengthens the US chip ecosystem or just raises prices for American cloud providers and AI startups is genuinely contested among economists.

Bob: The trickier issue is that modern chips cross borders many times before they end up in a server. A wafer might be made in Taiwan, packaged in Malaysia, tested in the US, and integrated into a system in Mexico. Blunt tariffs risk hitting every step of that journey. The proposed carve-outs for companies investing in US production try to soften that, but they also create winners and losers based on political calculations, not just economics.

Samantha: European technology companies sit in an awkward middle position here. Many rely on chips that pass through the US at some point, and higher costs there will ripple into European data centre bills, into consumer electronics prices, and into the cost of running AI services. On the other hand, if the US pulls foreign investment into its own soil, Europe has to ask harder questions about whether its own Chips Act, and its incentives for companies like ASML customers, are competitive enough. This story will keep resurfacing in Brussels in the months ahead.

The University of Florida and Autodesk open the most advanced robotics construction lab in the US to tackle the housing crisis.

Bob: Now for something concretely not about AI. The University of Florida and design software company Autodesk have opened what they call the most advanced robotics-industrialised construction lab in the United States. The goal is straightforward: build houses faster and cheaper by shifting construction from muddy building sites to factory floors, where robots handle a lot of the repetitive work. The lab focuses on prefabricated components: walls, floors, roof sections, made with robotic arms and automated assembly lines.

Samantha: This matters because the housing crisis is not just a European story. In the US, labour shortages and rising materials costs have made new homes unaffordable for a big chunk of the population. Prefabrication is not new, but combining it with modern robotics, computer-controlled cutting, robotic welding, automated finishing, could genuinely change the productivity of the sector. The lab's own claim is that this makes existing workers more productive rather than replacing them. That is the political framing, and it will be tested in practice.

Bob: The interesting technical piece is that Autodesk brings the design software layer. If you design a house in their tools, you can send those designs straight to the robots on the factory floor. That closes the gap between architectural drawing and physical construction, which today involves a lot of manual translation and errors. Whether this scales beyond a research lab into actual mass housing depends on regulation, on unions, and on whether developers trust the finished product. Those are big ifs.

Samantha: European countries facing severe housing shortages, the Netherlands, Germany, Ireland, the UK, will watch this closely. Sweden and Germany already have strong prefabrication traditions, but the robotics layer is less developed. If Florida's approach demonstrates measurable cost reductions, expect European housing ministers to start asking their own construction sectors uncomfortable questions about productivity. There is also a labour dimension: European construction unions have historically been open to automation if it targets shortages rather than jobs. That conversation is coming.

A European-made team messaging platform to try today: Rocket dot Chat.

Bob: To close out, two things you can actually use today. The first is Rocket dot Chat, an open-source team communication platform developed largely out of Brazil and Europe, with servers you can host yourself or use through their cloud. Think Slack or Microsoft Teams, but without the vendor lock-in, and with the option to keep every message on infrastructure you control.

Samantha: What makes Rocket dot Chat interesting is the flexibility. You get channels, direct messages, voice and video calls, file sharing, and integrations with tools like GitHub, Jira, and email, all in one place. Small teams can sign up for the free cloud version and be running in about ten minutes. Larger organisations, and this is where it gets popular in Europe, can install it on their own servers and keep every conversation entirely inside their own network. That is a real selling point for schools, healthcare organisations, and public bodies that cannot legally put staff chats on American cloud services.

Bob: The app runs on Windows, macOS, Linux, iOS, and Android, and there is a web version too. Users report that setup on the cloud version is genuinely quick, and the mobile apps feel modern rather than an afterthought. End-to-end encryption is available for direct messages and private channels. It is not a magic bullet, but for teams frustrated with the cost and the data location of the big American platforms, it is a serious alternative worth an afternoon's evaluation.

Samantha: For European teams, especially in regulated sectors, Rocket dot Chat's self-hosting option is the practical answer to a lot of GDPR questions. You know exactly where your data sits, because you put it there. The free cloud tier is enough to test with a small team, and if you outgrow it, you can migrate to your own server without changing habits. Available today on every major platform, no credit card required to start.

A free open-source home dashboard to organise everything you self-host: Dashy.

Bob: The second pick is Dashy, a free open-source dashboard for anyone who runs services at home or in a small office. If you have a home server, a Raspberry Pi, a router with a web interface, a smart home hub, and half a dozen browser bookmarks for all of them, Dashy pulls them into one clean, organised page you can open from any device on your network.

Samantha: It is highly customisable. You get sections, icons, live status checks that tell you if a service is up or down, and even simple widgets for weather, calendar, or system stats. Users praise how much easier it makes daily use of a home lab: instead of remembering a dozen local addresses, you just open one page. It runs in a Docker container, which sounds technical but is really a one-line install on any Linux machine or Raspberry Pi.

Bob: You do not need to be a system administrator to appreciate it. Even someone with a Synology or a QNAP network drive, a smart home controller, and a few web services can benefit. It is entirely free, has an active community, and there are pre-made themes so it looks good without design work. A calm, useful little project for a weekend afternoon.

Samantha: Because Dashy runs entirely on your own hardware, none of your dashboard data ever leaves your home network. For anyone in Europe experimenting with self-hosted alternatives to cloud services, this is the piece that ties it all together. Install it this evening on a Raspberry Pi or an old laptop, and by tomorrow you will have a single page for everything you run yourself.

Samantha: Today we covered: SK Hynix's four-billion-dollar memory plant in Indiana, more than a hundred organisations calling for global action on AI cyberattacks, the Trump administration's proposed new chip tariffs, robotics tackling the US housing crisis with the University of Florida and Autodesk, and two things to try yourself: the team messaging platform Rocket dot Chat and the home dashboard Dashy.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.