OpenBind Open AI Drug Discovery Data: What It Means

Samantha: Welcome to The State of Tech, The European Edition, Thursday May seven, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Microsoft wobbling on its climate pledge as AI guzzles power, a Wall Street semiconductor rally driven by AI spending, the UK-led OpenBind initiative dropping open data for drug discovery, XBOW raising thirty-five million to fight AI-powered hackers, Penn researchers cracking complex math with so-called Mollifier Layers, and finally a Samsung smartwatch that can predict when you're about to faint. Let's start with Microsoft.

Microsoft considers walking back its 2030 renewable energy pledge as AI data centres devour electricity.

Samantha: Microsoft is reportedly rethinking one of the most ambitious climate goals in big tech. The company committed to matching every hour of its electricity use with renewable energy purchases by 2030. Now, internal discussions suggest that target may be delayed, or scrapped entirely, because AI data centres are eating power at a pace nobody anticipated.

Bob: And the driver is exactly what you'd expect. Copilot, Azure, all the generative AI services Microsoft has been pushing aggressively, they need vast compute capacity. That means new data centres, and those data centres need power around the clock. Renewable supply simply isn't scaling fast enough to keep up.

Samantha: No final decision has been made, that's the important caveat. But the fact that this conversation is happening at all is significant. Microsoft has been one of the loudest voices on corporate climate leadership. If they pull back, it sends a signal to the entire industry about which pledges are realistic in the AI era and which were aspirational.

Bob: It also exposes a structural problem. The 24/7 hourly matching standard is much harder than the older annual matching model, where you just buy enough green energy across a year to cover your total use. Hourly matching means you need clean power available at every moment, including overnight, which is exactly when solar drops off.

Samantha: The bottom line is the AI boom and the energy transition are now visibly colliding. Hyperscalers are turning to nuclear, to gas, to anything that delivers reliable baseload. And every gigawatt that goes to training a model is a gigawatt that doesn't decarbonise something else.

Bob: There's a credibility issue too. Investors, customers, and regulators have been told that growth and sustainability can coexist. If the largest AI operator in the world admits the maths no longer works, that narrative takes a serious hit.

Samantha: For Europe this matters on multiple fronts. The EU has stricter sustainability reporting rules than most regions, and Microsoft operates major data centres in Ireland, the Netherlands, and the Nordics. National grids are already strained, and Ireland in particular has flagged data centre demand as a planning crisis.

Bob: European regulators will watch this closely. If Microsoft eases its target globally, expect pressure to enforce harder local commitments here. And European cloud customers, especially in regulated sectors, will start asking awkward questions about the carbon profile of every prompt they run.

Wall Street semiconductor stocks surge on AI spending as AMD jumps nineteen percent and valuations stretch.

Bob: Moving on to the markets, where AI is doing what AI does best right now, which is making chip stocks go vertical. The S&P 500 and Nasdaq both hit new highs, and the engine is semiconductors. AMD jumped nineteen percent on strong earnings. Intel, Super Micro, Nvidia, all riding the same wave.

Samantha: The mechanics are straightforward. Hyperscalers are spending unprecedented sums on AI infrastructure, that creates a shortage of compute and memory chips, which gives chipmakers pricing power, which lifts margins. It's a virtuous cycle for sellers, and an expensive one for buyers.

Bob: But here's where caution comes in. The Philadelphia Semiconductor Index is trading at 13.2 times price-to-sales. The S&P 500 sits at 3.5. That's nearly four times the broader market multiple. Historically, that kind of spread doesn't end gently.

Samantha: Analysts are split. The bulls argue this AI cycle is structurally different, that we're at the start of a multi-decade infrastructure build-out, and chipmakers will compound earnings for years. The bears point out that every previous semiconductor super-cycle ended with a brutal inventory correction.

Bob: And memory is particularly interesting. HBM, high-bandwidth memory, is now a critical bottleneck for AI training. Samsung, SK Hynix, Micron, they're effectively printing money. But memory has always been the most cyclical corner of the chip world.

Samantha: The bottom line is index performance is now dangerously concentrated. If a handful of AI-linked semiconductor names correct sharply, they drag the entire market down with them. That's a portfolio risk even for investors who think they're diversified.

Bob: For Europe, this lands in a few places. ASML in the Netherlands sells the lithography machines that make all of this possible, so the order book there is a live indicator of how real the AI demand is. ASM International, BE Semiconductor, similar story.

Samantha: And European pension funds hold significant exposure to US tech through index products. A sharp semiconductor correction would ripple straight into Dutch, Nordic, and German retirement portfolios. Risk officers across the continent should be running stress tests right now.

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.

UK-led OpenBind initiative releases open AI model and dataset to accelerate global drug discovery.

Bob: The OpenBind initiative, led out of the UK, has just published its first dataset and a predictive AI model called OpenBind v1. The goal is to crack one of the hardest bottlenecks in pharmaceutical research, which is the chronic lack of high-quality data on how drug molecules actually bind to disease-related proteins.

Samantha: This is a real gap. Most published AI work in drug discovery uses small or proprietary datasets, which limits how well the models generalise. OpenBind is generating binding data on an industrial scale and putting it in the public domain. That's a meaningful change in how this field operates.

Bob: The model itself predicts binding affinity, essentially how tightly a candidate molecule sticks to a target protein. That's the first filter in drug screening. If you can do that computationally with high accuracy, you skip enormous amounts of expensive wet-lab work.

Samantha: The dataset matters as much as the model. AI in pharmacology has been talked up for a decade, but the lack of consistent, high-quality experimental ground truth has held it back. Releasing standardised data lets every research group, academic or commercial, train better models.

Bob: The bottom line is this lowers the barrier to entry for AI-driven pharma research. Smaller biotechs and academic labs can now compete with the data resources that used to belong only to the largest companies. That tends to accelerate innovation.

Samantha: And it sets a precedent. Open infrastructure has transformed other AI fields, language models, protein structure prediction with AlphaFold. Drug-target binding could be next.

Bob: For Europe, the timing is excellent. The UK, Switzerland, Denmark and Germany all host major pharma R&D operations, and European universities are strong in computational biology. Open access means companies like Roche, Novartis, Novo Nordisk, and the smaller biotech ecosystem in Cambridge and Munich can integrate this immediately.

Samantha: It also reinforces the UK's positioning as a serious AI-for-science hub post-Brexit. Initiatives like this are how the UK keeps influence in European research networks even outside the EU framework.

XBOW raises thirty-five million dollars as AI-powered cyber attacks force a defensive arms race.

Samantha: XBOW, which builds continuous offensive security tools, has closed a thirty-five million dollar round. The investors include Accenture Ventures, DNX Ventures, Liberty Global Tech Ventures, NVentures, Samsung Ventures, and SentinelOne S Ventures. That's a notable mix of strategic backers.

Bob: The premise is simple and a bit uncomfortable. Attackers are using AI to find vulnerabilities faster, write exploits faster, and scale operations in ways that overwhelm traditional security teams. XBOW essentially fights AI with AI, running continuous automated offensive testing against client systems.

Samantha: CEO Oege de Moor framed it as turning insights from large-scale offensive operations into faster defensive responses. The company already serves more than a hundred customers globally, which is a real footprint for a company at this stage.

Bob: The category itself, continuous automated red teaming, is one of the fastest-growing parts of cybersecurity. Manual penetration testing happens once or twice a year. Automated AI-driven testing happens constantly, which matches the tempo of automated attacks.

Samantha: Key takeaway: the asymmetry between attackers and defenders is widening, and AI is the lever on both sides. Companies that don't automate defence at machine speed will simply be outpaced.

Bob: And the financial stakes are enormous. Ransomware payouts, regulatory fines, downtime costs, they're all rising. Thirty-five million for a defensive AI platform looks small against the size of the underlying problem.

Samantha: For Europe, this is timely. The NIS2 directive is now in force across member states, raising cybersecurity obligations for thousands of organisations. Many of those companies don't have mature security teams, so AI-powered tooling is one of the few realistic ways to comply.

Bob: And European critical infrastructure, energy, water, transport, has been hit by a steady drumbeat of attacks. Tools like XBOW are exactly what regulators are pushing operators toward. Expect more European procurement to flow this way.

University of Pennsylvania researchers unveil Mollifier Layers to solve notoriously hard inverse partial differential equations.

Bob: Researchers at the University of Pennsylvania have introduced a technique they call Mollifier Layers. The target is inverse partial differential equations, which are a foundational tool across physics, biology, and engineering, but notoriously hard to solve when data is noisy or incomplete.

Samantha: The clever part is that they didn't just throw more compute at the problem. They refined the mathematical formulation itself, adding a smoothing layer that reduces noise sensitivity. The result is more stable solutions with much lower computational demand.

Bob: Inverse problems are everywhere. You see something, and you want to work backwards to figure out what caused it. Medical imaging, seismic analysis, climate modelling, fluid dynamics, gene regulation, they all reduce to inverse PDE problems at some level.

Samantha: The challenge has always been that small measurement errors get amplified into huge errors in the recovered parameters. Mollifier Layers damp that amplification by construction, rather than relying on brute force regularisation.

Bob: The bottom line is this is the kind of foundational tooling that quietly accelerates entire scientific fields. Better inverse solvers mean better fluid simulations, better material design, better biological models.

Samantha: And it pushes back on the idea that AI progress only comes from scale. Sometimes a smarter mathematical formulation beats a bigger GPU cluster.

Bob: For Europe, this aligns well with strengths in places like ETH Zurich, INRIA in France, the Max Planck institutes, and the Alan Turing Institute. European computational science has always emphasised mathematical rigour. Tools like this fit that culture.

Samantha: And applied fields benefit too. European climate modelling consortia, automotive simulation teams, pharma modelling groups, they all run inverse problems daily. A more stable solver is a direct productivity gain.

Samsung's Galaxy Watch6 predicts fainting spells five minutes early with eighty-five percent accuracy.

Samantha: And to close, something genuinely surprising. Samsung, working with Chung-Ang University Gwangmyeong Hospital, has validated that the Galaxy Watch6 can predict fainting. Specifically vasovagal syncope, the most common type of fainting spell.

Bob: And the numbers are striking. Up to five minutes of warning, with 84.6 percent accuracy. The watch uses its photoplethysmography sensor, which is the optical heart-rate sensor on the back, and an AI algorithm analyses heart rate variability patterns that precede a fainting episode.

Samantha: What makes this clever is they didn't add new hardware. The PPG sensor is already in millions of watches. The breakthrough is in what the algorithm extracts from the signal.

Bob: Vasovagal syncope sounds minor, but the secondary injuries are not. People hit their heads, fall down stairs, faint while driving. A five-minute warning gives someone time to sit down, pull over, get to a safe spot.

Samantha: This is part of a broader pattern. Wearables started as step counters, then added heart rate, then ECG, then sleep apnoea detection. Now predictive health alerts. Each step pulls the wrist into territory that used to require a hospital.

Bob: And the economics are interesting. If a consumer device prevents one serious fall, it has more or less paid for itself versus the healthcare system. That argument will increasingly drive insurer involvement.

Samantha: For European users, the catch is regulatory. Predictive health features need to clear MDR, the EU Medical Device Regulation, before Samsung can market them as medical claims here. That process is slower than in some other markets.

Bob: But once approved, features like this could become genuinely useful for patients with diagnosed syncope or cardiac conditions. And it gives European health systems a low-cost way to extend monitoring outside the clinic.

Samantha: Today we covered: Microsoft rethinking its climate pledge under AI pressure, the AI-driven semiconductor rally and stretched valuations, the OpenBind open dataset for drug discovery, XBOW's funding round against AI-powered cyber threats, Penn's Mollifier Layers for hard math problems, and a Samsung smartwatch that predicts fainting.

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.