PLUS: Merck’s AI cuts drug dev from 10 years to months, UK pension AI fails catastrophically, FDA redraws wearables rules, Philips trains 70K workers

Morning All, MIT has cracked AI's energy paradox, deploying smart grids that balance rising computational demands with renewable integration - turning a mounting crisis into sustainable infrastructure.

As energy needs from data centers prepare to double this decade, Merck provides a parallel breakthrough: slashing drug development timelines from 10 years to months by simulating molecular behaviours. Are we witnessing AI evolve from problem-creator to solution-engineer?

Today's dots:

• MIT's AI-powered grid optimisation
• Merck slashes decade-long drug development cycles
• UK pension catastrophic AI miscalculations
• FDA redraws rules for AI wearables
• Philips trains 70K workers in AI adoption

MIT's AI solves its own energy crisis

Here's the thing: MIT researchers just demonstrated how AI can optimise power grids to handle its own growing energy demands while boosting renewable integration - a critical breakthrough for sustainable AI scaling (MIT announcement).

Let's unpack that:

• The system maintains exact balance between energy supply/demand while accounting for unpredictable renewable outputs like solar and wind - crucial as grids decarbonise
• AI creates faster, more accurate approximations for grid optimisation problems that normally take hours to solve - enabling real-time adjustments during weather extremes
• This comes as data centre energy consumption is projected to double by 2030 (from 415TWh to 945TWh) according to IEA data
• Researchers are partnering with utilities to deploy smaller, application-specific AI models that respect power grid physics - avoiding blackout risks from less precise general AI

If you remember nothing else: AI's energy appetite is real, but targeted applications like this show the technology can also be part of the solution. As grids get cleaner and smarter, they'll enable sustainable scaling of both AI and renewables together.

Merck's AI slashes drug development from decade to months

Here's the thing: Pharma giant Merck now completes drug discovery in months instead of 10 years using AI foundation models - while slashing clinical trial dropout rates through predictive analytics. See their breakthrough announcement.

Let's unpack that:

• Their custom AI models design drug candidates 10x faster by simulating millions of molecular interactions before lab testing - already producing viable cyclic peptide treatments
• Predictive algorithms flag at-risk patients in clinical trials, enabling targeted interventions that reduced dropout rates by 18% last quarter (clinical trial details)
• 80% of staff now use internal AI tools to automate regulatory paperwork and medical reviews, freeing scientists for higher-value research
• Vaccine production lines use computer vision that spotted microscopic vial defects human inspectors missed - slashing waste by 7%

If you remember nothing else: This isn't just about faster lab work - it's about getting life-saving treatments to patients years sooner. Merck's proving AI can accelerate biology's most complex puzzles while maintaining rigorous safety standards.

Pension AI Fails Basic Financial Tests

Here's the thing: 65% of UK pension providers discovered dangerous AI calculation errors during recent audits, with systems hallucinating answers and misdirecting users - putting both retirees and firms at regulatory risk. Quietroom's research reveals why most financial AI deployments aren't ready for real-world use.

Let's unpack that:

• Basic website features like accordions (expandable info boxes) blocked OpenAI's Operator tool from reading pension details accurately - leading to wrong retirement age calculations
• When confused by poorly structured content, chatbots didn't admit uncertainty but hallucinated answers using other schemes' data - with 100% failure rates in controlled tests
• Firms now face FCA Consumer Duty violations when members make bad decisions based on AI's fictional outputs - accountability stays with providers
• Some tools dangerously directed users to unvetted financial advisors rather than scheme contacts, creating compliance minefields
• The fix? Human-first content design - clear sentences and logical structures help both people and AI understand critical financial details

If you remember nothing else: These flaws expose how supposedly 'smart' systems threaten real-world financial decisions when trained on messy data. Meanwhile, financial firms remain legally responsible for every AI-generated answer - making transparent, human-checked systems essential.

FDA sharpens rules for AI-driven wearables

Here's the thing: The FDA just clarified how it regulates AI-powered wellness wearables versus medical diagnostic tools - creating clearer paths for healthtech innovators while tightening oversight on devices crossing into healthcare.

Let's unpack that:

• Boundaries defined: General wellness devices (fitness trackers, step counters) get greenlight for minimal regulation if they don’t claim to diagnose/treat conditions – while tools making clinical-grade claims (like blood pressure monitoring) face stricter scrutiny.
• Real-world ripple: Companies got immediate market boosts (Garmin +3%, glucose monitor makers +4%) as investors reacted to predictable regulatory lanes.
• Agency intentions: FDA Commissioner Marty Makary told Fox Business they’re balancing “promoting innovation” with guarding against safety risks when AI tools like symptom checkers interface with healthcare decisions.
• Enforcement precedent: WHOOP’s 2025 warning letter shows consequences - its blood-pressure feature got flagged for estimating clinical metrics like diastolic values without medical approval.
• Consumer warnings: Guidance reinforces that wellness devices are decision-support tools only - never replacements for professional medical consultations.

If you remember nothing else: This framework gives healthtech startups critical certainty about product categorisation. Expect faster innovation in non-medical wearables - but stricter enforcement for devices blurring the wellness-healthcare divide.

Philips' 70K-Strong Workforce Masters AI - And It’s Paying Off

Here's the thing: Philips just achieved 91% AI adoption across its 70,000-strong workforce through a deliberate 'toy-to-tool-to-transformation' approach - and it's already saving clinicians precious patient time. Learn how they did it.

Let's unpack that:

• Executives got hands-on training first to model practical AI use, proving leadership wasn't just mandating change but living it
• They balanced top-down strategy with bottom-up innovation through company-wide challenges letting employees pitch-and-test use cases
• Started with low-risk internal workflows (like medical device documentation) to build trust before touching patient-facing systems
• Enterprise ChatGPT access created organic momentum - proving that when you remove barriers, people rapidly find high-value applications
• Measurable ROI comes from saving clinicians 40% documentation time - turning bureaucratic hours into life-saving minutes

If you remember nothing else: This isn't just about tech adoption - it's proof that cultural transformation comes when leadership learns alongside staff. The real win? Healthcare AI success gets measured in minutes returned to patient care, not just efficiency metrics.

The Shortlist

Dell unveiled disaggregated infrastructure solutions embedding AI governance into data center operations - a critical move as enterprises face 30% higher demand for GPU-accelerated analytics workloads.

Netomi detailed how its AI agents handle Fortune 500 financial workflows using OpenAI's GPT-5.2 for multi-step reasoning, cutting client incident resolution time by 37% while maintaining 98% audit compliance rates in heavily regulated environments.

CES revealed physical AI's growing dominance beyond chatbots, with NVIDIA's automotive AI processing 40,000 concurrent customer requests at sub-three-second latency - proving local AI can now handle safety-critical edge computing without cloud dependencies.

Researchers demonstrated machine learning's ability to predict mining exploration targets 45% more accurately than traditional methods by analysing spectral data and drill core logs - potentially unlocking $2.5B in operational savings across copper/gold exploration.

Omnicom deployed its AI-powered Omni platform across $73.5B in media spend, using generative models to optimize real-time ad creative while maintaining brand compliance - a blueprint for marrying artificial creativity with media economics at scale.