Google TurboQuant, Agentic AI Insurance & Meta Layoffs

Samantha: Welcome to The State of Tech — The European Edition. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Google drops a memory compression breakthrough that could slash AI costs dramatically. An agentic AI platform goes live with a major European insurer. Caris Life Sciences launches AI that predicts brain metastases in cancer patients. US states are racing ahead with a patchwork of AI legislation. Washington accuses China's biggest chipmaker of supplying tech to Iran's military. And Meta lays off hundreds while its executives line up a billion-dollar bonus package. Let's start with Google.

Google claims it can run massive AI models using a fraction of the memory — and without losing accuracy.

Samantha: So Google has unveiled something called TurboQuant, and on paper, the numbers are pretty wild. We're talking about a compression algorithm that reduces key-value cache memory — that's the working memory AI models need during inference — by at least six times. And processing speeds? Up to eight times faster. The kicker is they're claiming zero accuracy loss.

Bob: And that's the part that really matters here. Compression techniques have been around for a while, but there's always been a trade-off. You squeeze the model down, you lose some quality. Google is saying they've cracked that problem. Now, if that holds up in real-world deployment at scale, this changes the economics of running large language models fundamentally. Think about what it costs to run Gemini or any frontier model — the memory, the GPUs, the electricity. Cut that by a factor of six and suddenly a lot of applications that were too expensive become viable.

Samantha: Google is explicitly framing this as a solution to what they call the global memory shortage. And they're not wrong. The bottleneck for AI deployment right now isn't just chips — it's the memory those chips need to access.

Bob: For Europe specifically, this is quietly a big deal. European companies have been slower to deploy large-scale AI partly because of infrastructure costs. If TurboQuant or techniques like it bring those costs down sharply, you could see a wave of adoption across European enterprises that have been sitting on the sidelines. And there's an environmental angle too — less memory, less energy, smaller carbon footprint. That matters when you're operating under EU sustainability requirements.

Samantha: It also makes European cloud providers slightly more competitive. If you don't need as much hardware to deliver the same capability, the gap between hyperscalers and smaller providers narrows a bit.

Bob: A bit. Let's not get carried away. But directionally, yes — efficiency gains are democratising gains.

---

An agentic AI system is now live inside a major European insurance company — processing invoices autonomously.

Samantha: Next up, Global AI Inc. — ticker GLAI — has announced that its agentic AI platform is now live with a leading European insurer. And we're not talking about a pilot or a proof of concept. This thing is running in production, processing invoices through multiple daily cycles, end to end.

Bob: So let's unpack what "agentic" means in practice here. This isn't a chatbot answering questions. This is an AI system that receives invoices, reads them, understands them, validates them against policy rules, and pushes them into backend systems — all without a human in the loop for routine cases. It handles the full workflow autonomously.

Samantha: And the fact that it's running inside a highly regulated European insurance environment is significant. Insurance is one of the most compliance-heavy sectors in Europe. You've got Solvency II, GDPR, the AI Act coming into full enforcement. If an agentic system can operate there with full auditability, that's a serious validation.

Bob: The company says they're already in discussions to expand into other insurance workflows — claims processing, underwriting support. This is the pattern we've been watching. Once an agentic system proves it can handle one process reliably, it tends to spread fast within an organisation.

Samantha: The bigger picture for Europe is that this shows regulated industries aren't waiting anymore. There was this narrative that European regulation would slow AI adoption. But what we're seeing instead is that regulation is shaping the type of AI that gets adopted — auditable, transparent, process-driven. And agentic platforms that are built with those guardrails from the start actually have an advantage in Europe.

Bob: And that's a competitive edge, by the way. If you can demonstrate your AI works inside EU compliance frameworks, you can sell it anywhere. Europe's strict standards become a stamp of credibility.

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.

---

AI is now predicting which cancer patients are at risk of brain metastases — before they happen.

Bob: Caris Life Sciences has launched two new AI-powered diagnostic signatures that assess the risk of brain metastases in breast and lung cancer patients. These are now included in their Molecular Tumor Board Report, which means oncologists get this information as part of their standard workup.

Samantha: Let's put this in context. Brain metastases are devastating. When breast or lung cancer spreads to the brain, treatment options narrow dramatically and outcomes get much worse. Being able to predict that risk early — before it happens — could fundamentally change how doctors manage these patients. Maybe you monitor more aggressively. Maybe you adjust treatment earlier. Maybe you catch it when intervention is still effective.

Bob: The AI is trained on multimodal real-world datasets, which means it's pulling from genetic sequencing data — both whole exome and whole transcriptome — and generating personalised risk scores. This isn't a one-size-fits-all prediction. It's tailored to the individual patient's molecular profile.

Samantha: And this is where precision oncology has been heading for years, but the AI component is what makes it scalable. You can't have a team of geneticists manually analysing every patient's sequencing data. The algorithms do it consistently and fast.

Bob: For European healthcare, this is compelling. Europe has strong cancer research infrastructure — think of the Netherlands Cancer Institute, Gustave Roussy in France, Charité in Berlin. But access to cutting-edge diagnostic tools varies enormously across member states. AI-driven diagnostics like this could help level the playing field if they're adopted broadly.

Samantha: There's also the cost argument. Catching metastatic progression early is far cheaper than treating it late. European health systems running on tight budgets should pay close attention here.

Bob: And from a regulatory standpoint, the EU's Medical Device Regulation and the AI Act will both touch this kind of tool. Getting the framework right so these innovations reach patients quickly without cutting corners on safety — that's the balancing act European regulators face.

---

While Europe has one AI Act, US states are writing dozens of their own AI laws — and the patchwork is growing fast.

Samantha: Let's talk about what's happening with AI legislation in the United States, because it is getting messy. State legislatures across the country are advancing all kinds of AI-related bills. Chatbot safety laws aimed at protecting minors. AI provenance requirements so people can tell when content is generated by a machine. Social media harm bills that target AI-driven addictive features. It's a grab bag.

Bob: And there's no federal framework tying any of this together. So you end up with potentially fifty different regulatory regimes. Some states are further along — bills signed into law. Others are still in committee. But the direction is clear: states aren't waiting for Congress.

Samantha: This creates a genuinely complicated environment for tech companies. If you're a European AI firm trying to sell into the US market, you now have to think about compliance state by state. That's expensive and unpredictable.

Bob: The interesting contrast is with Europe. The EU AI Act is one framework for twenty-seven countries. You can criticise it for being complex or slow, but at least it's one set of rules. The American approach is fragmented by design — and that fragmentation has real costs.

Samantha: It also means global standards are harder to achieve. If the US and EU can't even align their approaches internally, aligning with each other becomes even more of a stretch. For companies operating on both sides of the Atlantic, the compliance burden is stacking up.

Bob: One thing to watch — some of these US state bills, particularly around AI provenance and content labelling, actually mirror provisions in the EU AI Act. So there's convergence happening at the edges, even if the overall structures are very different.

---

Washington says China's top chipmaker has been funnelling semiconductor technology to Iran's military for about a year.

Samantha: This is a serious escalation. Senior US officials are alleging that SMIC — Semiconductor Manufacturing International Corporation, China's largest chipmaker — has been supplying chipmaking tools and possibly technical training to Iran's military. And they say this has been going on for roughly a year.

Bob: SMIC is already under US sanctions because of its alleged ties to the Chinese military. So if these allegations are confirmed, it suggests the existing sanctions regime has significant gaps. You've got a sanctioned Chinese company allegedly helping Iran build its own chip fabrication capabilities while the US and Israel are in active conflict with Iran.

Samantha: The geopolitical implications are enormous. This isn't just about trade controls — it's about military technology reaching an adversary through a third country. And it raises the question of whether current export controls are fit for purpose.

Bob: For Europe, this hits close to home. European companies are deeply embedded in the semiconductor supply chain. ASML in the Netherlands makes the lithography machines that are essential for advanced chip production. If there's evidence that controlled technology — from any source — is being redirected to sanctioned entities, European governments will face pressure to tighten their own export controls further.

Samantha: And Europe's own push for tech sovereignty gets another argument in its favour. If global supply chains can be exploited for military purposes, having more domestic capability isn't just an economic goal — it's a security imperative.

Bob: The next few weeks will be critical. If the US presents hard evidence, expect coordinated pressure on European and Asian allies to close whatever loopholes exist.

---

Meta is cutting hundreds of jobs while top executives stand to earn nearly a billion dollars in incentives.

Bob: And finally — Meta. The company has laid off hundreds of employees this week. At the same time, reports emerged of a new executive incentive plan worth close to one billion dollars. And hovering over all of this is a landmark lawsuit alleging Meta's products are designed to be addictive and have harmed children.

Samantha: The optics are brutal. You're telling staff they're out of a job while the C-suite is being offered life-changing bonuses. Meta says the restructuring is about optimising teams, which is the standard line. But the timing makes it very hard to sell that story.

Bob: The lawsuit is potentially more consequential long term. If the court finds that Meta deliberately engineered addictive features and that this caused measurable harm to minors, it sets a legal precedent that every social media company in the world will have to reckon with.

Samantha: And in Europe, regulators have been ahead of the curve on this. The Digital Services Act already imposes obligations around algorithmic transparency and protecting minors. If this US lawsuit goes badly for Meta, it could embolden European regulators to push even harder — bigger fines, stricter enforcement, maybe even structural remedies.

Bob: Meta has significant European operations — Ireland is their EU headquarters. Any global restructuring will hit European employees directly. And if the legal and regulatory walls keep closing in, the cost of operating in Europe goes up for Meta, which could mean further cuts or strategic retreats from certain features.

Samantha: It's a company under pressure from every direction. Shareholders want efficiency. Courts want accountability. Regulators want compliance. And employees just want to know if they still have a job next quarter.

Bob: Not a comfortable position for anyone at Meta right now.

Samantha: Today we covered: Google's TurboQuant memory compression breakthrough, an agentic AI going live with a European insurer, AI predicting brain metastases in cancer patients, the patchwork of US state AI laws, allegations that SMIC supplied chip tech to Iran, and Meta's layoffs alongside billion-dollar executive bonuses.

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.