DeepSeek V4 Open-Source AI Challenges Closed Models

Samantha: Welcome to The State of Tech, The European Edition, Wednesday April twenty-nine, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI missing its targets while expanding on AWS, Meta's enormous new AI infrastructure bill, Google rethinking Gemini's voice features, Japan Airlines bringing humanoid robots to Haneda, DeepSeek launching its V4 model series, and quantum computing hitting what McKinsey calls a commercial tipping point. Let's start with OpenAI.

OpenAI misses revenue targets, expands to AWS, and rattles AI stocks worldwide.

Samantha: OpenAI is reportedly missing its own internal targets for new users and revenue growth. According to leaks from inside the company, some leaders are now openly questioning whether the massive data centre investments can be sustained at the current pace. AI-related stocks slid globally on the news.

Bob: And at the same time, OpenAI is widening its enterprise reach. The company is expanding its partnership with Amazon Web Services, putting its frontier models in front of a much bigger pool of AWS customers. That follows the end of the exclusivity deal with Microsoft Azure earlier this year.

Samantha: So you have two stories pulling in opposite directions. On one side, doubts about whether the revenue can ever match the spend. On the other, a clear push to get the models everywhere, not just on Azure.

Bob: The AWS move is essentially a distribution play. If you can't grow consumer revenue fast enough, you broaden enterprise channels. Sam Altman has said before that compute is the bottleneck, but now the question is whether demand justifies the bottleneck.

Samantha: This is the first real crack in the AI growth narrative since 2023. For two years markets assumed OpenAI's user numbers would keep climbing in a straight line. Missing internal targets shifts that assumption.

Bob: And the data centre numbers are eye-watering. Hundreds of billions committed across the industry, all betting on continued exponential adoption. If the curve flattens even slightly, somebody is left holding a lot of expensive silicon.

Samantha: For European pension funds and tech investors, the AI stock slide is a direct hit. A lot of European institutional money has flowed into US AI names through index funds.

Bob: But there's an upside too. European enterprises that were locked out of OpenAI through the Azure-only deal now get genuine choice. AWS Frankfurt, AWS Ireland, those regions matter for data residency. More competition usually means better pricing and better terms for European buyers.

Samantha: It also strengthens the case for European AI sovereignty projects. If even OpenAI is wobbling on its growth story, Mistral and Aleph Alpha suddenly look less like underdogs and more like reasonable alternatives.

Bob: Reasonable, yes. Cheap, no. They still need the same compute everyone else does.

Meta commits up to 135 billion dollars to AI infrastructure in 2026.

Bob: Meta dropped its Q1 2026 earnings yesterday and the headline number is the capex guidance. Between 115 and 135 billion dollars for this year alone, almost entirely going into AI infrastructure. That's a major step up from 2025.

Samantha: The supplier list reads like a who's who. Graviton5 chips from AWS, fibre-optic from Corning, and a long-term nuclear power agreement with Vistra to keep the data centres running. Mark Zuckerberg is still using the phrase "personal superintelligence" to justify it all.

Bob: The bet is straightforward. Advertising revenue keeps growing fast enough to cover the AI bill. So far, the ad business has cooperated. Q1 numbers came in above expectations on that front.

Samantha: Nuclear power is the part that catches my eye. Meta isn't buying electricity, it's locking in decades of baseload generation. That tells you they expect these data centres to run flat-out into the 2040s.

Bob: The scale is reshaping entire supply chains. Corning is sold out of optical fibre through 2027. Nuclear operators are getting calls they haven't had since the 1980s. This is industrial policy by hyperscaler.

Samantha: And it sets a benchmark. Once Meta commits 135 billion, Microsoft, Google and Amazon have to match or risk falling behind on training runs.

Bob: For European chip and component suppliers, this is opportunity. ASML obviously, but also smaller players in optical networking and power electronics. Demand isn't slowing down.

Samantha: The flip side is regulatory. The European Commission is going to look hard at how Meta's AI investments interact with the Digital Markets Act and the AI Act. Especially around training data, advertising targeting, and any superintelligence claims.

Bob: And there's an energy story for Europe too. If the US ties up nuclear capacity for AI, European data centres will be competing for a smaller global pool of clean baseload power. That pushes up costs here.

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.

Google plans to remove Gemini's voice features in pivot to proactive AI.

Bob: APK teardowns of the latest Gemini app suggest Google is preparing to remove the existing voice interaction options entirely. Not upgrade, remove. The plan is to replace them with a more proactive, predictive AI experience.

Samantha: The idea is that Gemini stops being something you summon and becomes something that anticipates. It surfaces information before you ask, acts across apps, and behaves more like a universal interface than an assistant. Google wants to shed the "search company" label.

Bob: It's a big bet. Voice was the original promise of AI assistants going back to Google Now and Siri. Walking away from that interaction model is a real strategic shift.

Samantha: And it puts pressure on Apple, which is still rebuilding Siri. If Gemini becomes genuinely proactive on Android, the gap on iOS becomes more obvious.

Bob: Proactive AI is fundamentally different from reactive AI. Reactive AI waits for a prompt. Proactive AI watches your context, your calendar, your messages, and acts. The privacy surface area is enormous.

Samantha: It also changes the business model. A proactive assistant doesn't need search queries, which means less of the traditional ad inventory Google has built its empire on.

Bob: This is where the AI Act gets interesting. A predictive system that acts autonomously on personal data sits much closer to the high-risk categories than a simple chatbot does.

Samantha: European data protection authorities will want to know what data Gemini is observing, how consent works, and whether users can meaningfully opt out. The Irish DPC will be involved given Google's European headquarters.

Bob: For European Android users, this is also just a usability question. Hundreds of millions of people use voice on their phones daily. Removing it without a clear replacement is going to generate complaints.

Samantha: Especially in markets with older demographics or accessibility needs. Voice isn't a luxury feature for everyone.

Japan Airlines deploys humanoid robots at Haneda Airport for cargo handling.

Samantha: Japan Airlines is starting a two-year trial in May, putting Chinese-made humanoid robots to work at Tokyo's Haneda Airport. The first job: loading and unloading cargo containers on the ramp.

Bob: Japan's aviation sector is short of staff. Tourism is at record levels, the workforce is shrinking, and ground handling is physically brutal work. JAL is framing this as easing the load on humans rather than replacing them.

Samantha: It's one of the first large-scale humanoid deployments in actual airport operations, not a demo or a research lab. Real cargo, real schedules, real weather.

Bob: The fact that the robots are Chinese-made is also notable. Unitree and similar firms have been pushing hard on price, and Japanese operators are clearly willing to buy from them rather than wait for domestic options.

Samantha: Ground handling has been one of the toughest jobs to automate. Variable cargo, weather, tight time windows. If humanoids can do this reliably, a lot of other physical jobs are next.

Bob: And it's a real-world test of whether humanoids justify the hype. Two years of data from Haneda will tell us more than any number of lab demos.

Samantha: European airports face the same demographic squeeze. Schiphol, Frankfurt, Heathrow, all dealing with chronic ground staff shortages. If JAL's trial works, expect procurement conversations across Europe by 2027.

Bob: The union response will be different here, though. European labour rules and works councils mean any humanoid rollout has to be negotiated, not just deployed. That slows things down, but it also means workers get a seat at the table.

Samantha: And there's a sovereignty angle. Buying humanoids from Chinese suppliers for critical airport infrastructure raises the same questions Europe has been asking about Huawei in telecoms.

DeepSeek launches V4 Flash and V4 Pro, claiming open-source AI leadership.

Bob: DeepSeek announced its V4 Flash and V4 Pro models today. The company says these are the most powerful open-source AI models available, with significant gains in coding and reasoning. The architecture uses a hybrid attention mechanism that DeepSeek claims is more efficient than transformer-only designs.

Samantha: V4 Flash is the lighter, faster variant. V4 Pro is the heavyweight aimed at complex reasoning tasks. Both are released under an open licence, which is the part that matters most.

Bob: DeepSeek has been the loudest disruptor in open-source AI since their V3 release shocked the market. V4 doubles down on that strategy, undercutting OpenAI and Anthropic on cost while staying genuinely open.

Samantha: And the timing, on a day when OpenAI is missing revenue targets, is brutal. The open-source side is gaining ground exactly when the closed-source giants are showing strain.

Bob: Open weights mean European companies can run these models on their own infrastructure, in their own data centres, without sending data to a US or Chinese cloud. That's a meaningful sovereignty win.

Samantha: It also forces a pricing conversation. If V4 Pro performs close to GPT-class models at a fraction of the cost, enterprise procurement teams are going to ask hard questions about why they're paying premium API rates.

Bob: For European startups and research labs, this is genuinely useful. You can fine-tune V4 for medical, legal or industrial use cases without licensing constraints. That lowers the barrier to entry significantly.

Samantha: The caveat is provenance. DeepSeek is a Chinese company, and questions about training data, security audits, and supply chain trust will follow these models into European deployments.

Bob: Right. Open weights help with one kind of sovereignty, but they don't answer every question. European regulators will want assurances before V4 ends up in critical infrastructure.

Samantha: Still, the direction of travel is clear. Open source is closing the gap on closed source faster than most people predicted a year ago.

Quantum computing hits commercial tipping point as Cisco unveils universal quantum switch.

Samantha: A new McKinsey report says quantum computing has reached its commercial tipping point. Over 300 companies globally are now actively adopting quantum technology, and the market is projected to hit billions by 2028.

Bob: The eye-catcher is Cisco's announcement of a universal quantum switch. It's designed to let quantum computers actually talk to each other, which has been one of the field's biggest unsolved problems. Quantum networking has been theoretical for years. Cisco just put hardware on the table.

Samantha: And there's a third piece: scientists have developed a method to track quantum data loss, what's called decoherence, more than a hundred times faster than before. That sounds dry, but it's the kind of measurement breakthrough that makes stable quantum systems possible.

Bob: Three pieces of news that line up nicely. Commercial adoption, networking, stability. That's basically the trio you need for quantum to leave the lab.

Samantha: We've heard "quantum is five years away" for about fifteen years now. McKinsey's report doesn't claim quantum supremacy is here, but it does say the money has shifted from research grants to actual procurement budgets.

Bob: And once procurement gets involved, things move faster. Boring, but true.

Samantha: Europe's Quantum Flagship programme has been quietly funding this kind of work for years. IQM in Finland, Pasqal in France, Quantinuum's UK operations, they all benefit when interoperability and stability improve.

Bob: And quantum is one of the rare tech areas where Europe genuinely competes at the frontier rather than catching up. A universal switch standard would let European quantum machines plug into global networks without losing sovereignty.

Samantha: Boring infrastructure news that turns out to be quietly thrilling. My favourite kind.

Bob: Quantum networking, the most exciting thing nobody at dinner wants to hear about.

Samantha: Today we covered: OpenAI's revenue worries and the AWS expansion, Meta's massive AI infrastructure spend, Google reworking Gemini's voice features, Japan Airlines bringing humanoids to Haneda, DeepSeek's V4 launch, and quantum computing's commercial tipping point.

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.