AI Solutions For Energy Management

Every cloud provider faces the same AI infrastructure challenge: chips need to be positioned close together to exchange data quickly, but they generate intense heat, creating unprecedented cooling demands. We needed a strategic solution that allowed us to use our existing air-cooled data centers to do liquid cooling without waiting for new construction. And it needed to be rapidly deployed so we could bring customers these powerful AI capabilities while we transition towards facility-level liquid cooling. Think of a home where only one sunny room needs AC, while the rest stays naturally cool – that’s what we wanted to achieve, allowing us to efficiently land both liquid and air-cooled racks in the same facilities with complete flexibility. The available options weren't great. Either we could wait to build specialized liquid-cooled facilities or adopt off-the-shelf solutions that didn't scale or meet our unique needs. Neither worked for our customers, so we did what we often do at Amazon… we invented our own solution. Our teams designed and delivered our In-Row Heat Exchanger (IRHX), which uses a direct-to-chip approach with a "cold plate" on the chips. The liquid runs through this sealed plate in a closed loop, continuously removing heat without increasing water use. This enables us to support traditional workloads and demanding AI applications in the same facilities. By 2026, our liquid-cooled capacity will grow to over 20% of our ML capacity, which is at multi-gigawatt scale today. While liquid cooling technology itself isn't unique, our approach was. Creating something this effective that could be deployed across our 120 Availability Zones in 38 Regions was significant. Because this solution didn't exist in the market, we developed a system that enables greater liquid cooling capacity with a smaller physical footprint, while maintaining flexibility and efficiency. Our IRHX can support a wide range of racks requiring liquid cooling, uses 9% less water than fully-air cooled sites, and offers a 20% improvement in power efficiency compared to off-the-shelf solutions. And because we invented it in-house, we can deploy it within months in any of our data centers, creating a flexible foundation to serve our customers for decades to come. Reimagining and innovating at scale has been something Amazon has done for a long time and one of the reasons we’ve been the leader in technology infrastructure and data center invention, sustainability, and resilience. We're not done… there's still so much more to invent for customers.

Microsoft and Meta Embrace New Power Design for AI Infrastructure: As data center rack densities rise to support more powerful GPUs for AI workloads, power distribution must also evolve. That's why Microsoft and Meta are collaborating on a design that will shift power conversion into a separate rack, laying the groundwork for denser and more configurable server racks. This disaggregated rack design, known as Mt Diablo, will initially use 48Vdc but will enable a shift to a 400Vdc power distribution system for AI data centers. The Mt Diablo project was disclosed at the recent Open Compute Project Foundation summit, and the architectural spec will be contributed to OCP to encourage further collaboration and development. "The need for scalability and future-proofing is driven by high-power server racks, which will exceed a few hundred kilowatts and are moving towards a megawatt," said Microsoft. "Our solution is to separate the single rack into an server rack and a power rack, each optimized for its primary function. With this approach, we can right-size the power shelf count to meet each configuration’s unique needs." The Meta team describes it as "a cutting-edge solution featuring a scalable 400 VDC unit that enhances efficiency and scalability. This innovative design allows more AI accelerators per IT rack, significantly advancing AI infrastructure." The companies say this approach will allow them to deploy 35% more accelerators in each rack, and the shift to 400Vdc will bring greater efficiency as data centers shift to extremely dense AI clusters. Mt Diablo has a modular design to support scalability and future-proofing as server racks grow denser, as well as different power configurations. Here's where you can learn more: Microsoft blog post: https://lnkd.in/e_tcGkEy Meta's blog post: https://lnkd.in/e6UeS86Q Open Compute presentation: https://lnkd.in/emjHAGji

The weather. A small talk topic for many, a (main) source of info during skiing season for others (🙋). And for our Hywind Tampen team: a challenge.   When the wind blows at Hywind Tampen, the world’s first – and largest floating wind farm, Equinor produces renewable energy for the Gullfaks and Snorre field platforms. That way, the use of traditional gas-powered turbines offshore are reduced – and so are our CO2 emissions.    But the wind is unpredictable.    Luckily, to artificial intelligence (AI), the weather and wind – like anything else, are only data. Data that can be structured and used to solve challenges.     Sure, AI won’t produce any wind (or snow, for that matter).    But by using our own data and a machine learning algorithm, Equinor have now developed a wind and weather prediction solution. This AI solution is based on historical and current weather data, and not least real-time wind measurements on our own installations in the area. It was launched late this October.    When the wind-measurements are telling us that the wind is blowing a few kilometers away, and the direction is right, we can expect wind at Hywind Tampen, even if the forecast says no. This makes the Hywind Tampen team (that operates the facility from onshore from Bergen) better at predicting and calculate just how much power we will get from the wind within the next 1-2 hours.     What does it mean in practice?    That we can beat traditional weather forecasts, in a way. And reduce the amount of idling power generators to a minimum.   I love the fact that we see more and more concrete examples on how AI is being used to optimize operations and solutions, like this one. Let me know below if you have other great examples 😀

🌬️ PART 3: AI, Wind Turbines, and LiDAR Tackling Yaw Misalignment 🛠️ Continuing our exploration of machine learning's role in enhancing wind turbine efficiency, let's talk about a common issue: yaw misalignment. When wind turbines aren't perfectly aligned with the wind, the consequences are two-fold. First, there's a significant dip in energy production, leading to lost revenue. Second, the misalignment causes increased loads on the turbines. This results in higher operational and maintenance costs. Addressing yaw misalignment is crucial for optimizing turbine efficiency and reliability. 🌪️ Understanding Yaw: The Wind Turbine's Compass 🌬️ Imagine the yaw system as the compass guiding a wind turbine, ensuring it faces the wind perfectly. It's like the brain behind the turbine, using a wind vane to detect where the wind is coming from. By adjusting the turbine's direction, it makes sure it's catching as much wind as possible, maximizing energy production.   🌀 Decoding Yaw Misalignment: Static vs. Dynamic 🌪️ Think of yaw misalignment in wind turbines as being off-target, either slightly or because of moving conditions. Static misalignment is like setting up your equipment with a slight offset from the start, due to human error or wear and tear. Dynamic misalignment happens as conditions change, like wind directions shifting, making the turbine sway and struggle to stay aligned. 🌬️ Nacelle LiDAR to the rescue... kind of: 📡 LiDAR technology measures wind speeds before they hit turbine blades, offering a preview that helps adjust the turbine's alignment for optimal efficiency. By detecting wind direction and speed early, LiDAR can fine-tune yaw alignment, reducing wear and enhancing power generation. Despite its benefits, high costs and data accuracy concerns temper widespread adoption. 🎛 Machine Learning + LiDAR = Yaw solution Because LiDAR is an expensive technology, we can leverage ML on real-time data to accurately predict the wind's approach, mimicking LIDAR's precision on a LiDAR-mounted turbine. This approach enhances turbine efficiency by precisely aligning with the incoming wind to maximize energy production and minimize stress. Now calibrating nacelle LiDAR and data extraction is another story. Until next time. Part1: https://lnkd.in/gqt89Q3G Part2: https://lnkd.in/drd8kAft #grid #windturbine #machinelearning #electricalengineering #iot #lidar #energy #energytransition #innovation #yycdata #yyctech #yyc

One of my all-time favorite A.I. researchers, Dr. Jason Yosinski, is my guest today! He details how his startup is using ML to collect wind energy more efficiently and digs into visualizing/understanding deep neural networks. Jason:  • Is Co-Founder and CEO of Windscape AI, a startup using ML to increase the efficiency of energy generation via wind turbines. • Is Co-Founder and President of the ML Collective, a research group that’s open to ML researchers anywhere. • Was a Co-Founder of the A.I. Lab at the ride-share company Uber. • Holds a PhD in Computer Science from Cornell, during which he worked at the NASA Jet Propulsion Laboratory, Google DeepMind and with the eminent Yoshua Bengio in Montreal. • His work has been featured in The Economist, on the BBC and, coolest of all, in an XKCD comic! Today’s episode gets fairly technical in parts so may be of greatest interest to hands-on practitioners like data scientists and ML engineers, although there are also parts that will appeal to anyone keen to hear how ML is being used to produce more clean energy. In today’s episode, Jason details:  • How ML can make wind direction more predictable, thereby making wind turbines and power grids in general more efficient. • How to infer what individual neurons in a deep learning model are doing by using visualizations. • Why freezing a particular layer of a neural net prior to doing any training at all can lead to better results. • How you can get involved in a cutting-edge research community no matter where you are in the world. • What traits make for successful A.I. entrepreneurs. Many thanks to Crawlbase for supporting this episode of Super Data Science, enabling the show to be freely available on all major podcasting platforms as well as the video version we publish on YouTube. This is Episode #789! #superdatascience #machinelearning #ai #climatechange #windenergy

Electricity management is increasingly an analytics problem where AI needs to step in. Decarbonization, variable demand, regenerative energy, and complex infrastructure make it impossible to rely on static rules or occasional reporting. Value comes from analyzing operational data continuously and turning it into decisions. The usual analytics setup does not scale. Work is often done in silos, with data pulled into notebooks, results shared as static reports, and little reuse across projects. Domain experts are separated from the analysis, cycles are slow, and each new use case starts largely from scratch. A collaborative model is a catalyst enabling AI to change the economics. At Mitsubishi Electric, data scientists work directly with domain experts on shared workflows. Analytics is used to identify concrete issues and opportunities. In railways, analysis showed where braking generates surplus energy and how it could be reused. In thermal energy management, a full year of building data was analyzed in 20 business days to optimize heating and cooling. Platform efficiency matters. By running the full AI lifecycle in Dataiku, Mitsubishi Electric reduced their time to produce new projects by about 60 percent. That translates into delivering value roughly 2.5 times faster, which means more use cases delivered and quicker operational impact. This is what AI Success looks like in energy and industrial systems. Read the full story on our website: https://lnkd.in/evhhuQNF 

After optimizing costs for many AI systems, I've developed a systematic approach that consistently delivers cost reductions of 60-80%. Here's my playbook, in order of least to most effort: Step 1: Optimizing Inference Throughput Start here for the biggest wins with least effort. Enabling caching (LiteLLM (YC W23), Zilliz) and strategic batch processing can reduce costs by a lot with very little effort. I have seen teams cut costs by half simply by implementing caching and batching requests that don't require real-time results. Step 2: Maximizing Token Efficiency This can give you an additional 50% cost savings. Prompt engineering, automated compression (ScaleDown), and structured outputs can cut token usage without sacrificing quality. Small changes in how you craft prompts can lead to massive savings at scale. Step 3: Model Orchestration Use routers and cascades to send prompts to the cheapest and most effective model for that prompt (OpenRouter, Martian). Why use GPT-4 for simple classification when GPT-3.5 will do? Smart routing ensures you're not overpaying for intelligence you don't need. Step 4: Self-Hosting I only suggest self-hosting for teams at scale because of the complexities involved. This requires more technical investment upfront but pays dividends for high-volume applications. The key is tackling these layers systematically. Most teams jump straight to self-hosting or model switching, but the real savings come from optimizing throughput and token efficiency first. What's your experience with AI cost optimization?

In a recent roundtable with fellow CXOs, a recurring theme emerged: the staggering costs associated with artificial intelligence (AI) implementation. While AI promises transformative benefits, many organizations find themselves grappling with unexpectedly high Total Cost of Ownership (TCO). Businesses are seeking innovative ways to optimize AI spending without compromising performance. Two pain points stood out in our discussion: module customization and production-readiness costs. AI isn't just about implementation; it's about sustainable integration. The real challenge lies in making AI cost-effective throughout its lifecycle. The real value of AI is not in the model, but in the data and infrastructure that supports it. As AI becomes increasingly essential for competitive advantage, how can businesses optimize costs to make it more accessible? Strategies for AI Cost Optimization 1.Efficient Customization - Leverage low-code/no-code platforms can reduce development time - Utilize pre-trained models and transfer learning to cut down on customization needs 2. Streamlined Production Deployment - Implement MLOps practices for faster time-to-market for AI projects - Adopt containerization and orchestration tools to improve resource utilization 3. Cloud Cost Management -Use spot instances and auto-scaling to reduce cloud costs for non-critical workloads. - Leverage reserved instances For predictable, long-term usage. These savings can reach good dollars compared to on-demand pricing. 4.Hardware Optimization - Implement edge computing to reduce data transfer costs - Invest in specialized AI chips that can offer better performance per watt compared to general-purpose processors. 5.Software Efficiency - Right LLMS for all queries rather than single big LLM is being tried by many - Apply model compression techniques such as Pruning and quantization that can reduce model size without significant accuracy loss. - Adopt efficient training algorithms Techniques like mixed precision training to speed up the process -By streamlining repetitive tasks, organizations can reallocate resources to more strategic initiatives 6.Data Optimization - Focus on data quality since it can reduce training iterations - Utilize synthetic data to supplement expensive real-world data, potentially cutting data acquisition costs. In conclusion, embracing AI-driven strategies for cost optimization is not just a trend; it is a necessity for organizations looking to thrive in today's competitive landscape. By leveraging AI, businesses can not only optimize their costs but also enhance their operational efficiency, paving the way for sustainable growth. What other AI cost optimization strategies have you found effective? Share your insights below! #MachineLearning #DataScience #CostEfficiency #Business #Technology #Innovation #ganitinc #AIOptimization #CostEfficiency #EnterpriseAI #TechInnovation #AITCO

🌀 From Predictive Models to Agentic AI — in Just a Few Hours I wanted to experience what it’s like to build an agentic pipeline firsthand. So I did. Use case? Predictive maintenance for wind turbines — minimizing downtime and maximizing efficiency. Here’s the flow I created in Dataiku: 🛠️ Agents in Action: Data Collector Agent → pulls live sensor data (temperature, vibration, performance). Data Processor Agent → cleans, formats, and normalizes the inputs. Predictive Model Agent → Deploys ML models to forecast failures (Offshore, Onshore Small, and Onshore large turbines). Maintenance Scheduler Agent → prioritizes turbine maintenance based on predicted risks. The result? A conversational interface powered by Agentic AI — One place. One entry point. One orchestration layer. And it was built in just a few hours, thanks to the reusable descriptive and predictive artifacts I already had in Dataiku. Here’s what I learned: ✅ Agents get complex fast ✅ Visibility, governance, and usability are critical ✅ If you can’t trust or trace your agents, you’re not scaling — you’re gambling 🔍 With Dataiku, building and debugging agents is possible and straightforward. 📣 Curious how this works in your industry? The Dataiku team will be talking about this stuff live, bring your questions https://lnkd.in/gJ-qJi8s #AgenticAI #PredictiveMaintenance #WindEnergy #DataScience #Dataiku #MLops #AIatScale #ConversationalAI

I have wrote recently about how Elon Musk powered his 150 MW Colossus #LLM training data center in Memphis. Now I’m taking on Meta’s 2.2 GW data center announcement in #Louisiana: Meta to Build a $10 Billion, 2.2 GW #AI Data Center in Louisiana Meta has announced plans to construct its largest data center to date in northeast Louisiana. The $10 billion facility, spanning four million square feet, will focus on training large language models (LLMs), requiring up to 2.2 GW of power at full build-out by 2030. LLM training data centers can be located in remote areas since latency is not a critical factor, making northeast Louisiana an ideal choice due to its access to reliable power infrastructure. The 2.2 GW of power for the facility will be supplied by Mitsubishi Power Americas, a long-time supplier to Entergy. The project will utilize three J-Series Air Cooled (#JAC) combined cycle power islands, known for their high efficiency, reliability and low emissions. Entergy Louisiana will invest $3.75 billion to provide the power infrastructure, showcasing the collaboration between Meta, Entergy, and Mitsubishi Power to support the energy-intensive needs of #hyperscale AI data centers. Five Key Takeaways from the Entergy-Meta Agreement 1. Prioritizing Power Access Over Other Considerations: Meta’s choice of northeast Louisiana underscores a shift in priorities for #AI-focused data centers, with fast access to power taking precedence over proximity to customers, robust digital infrastructure, or natural disaster risks. 2. Utility and Technology Collaboration: Mitsubishi Power Americas’ role in supplying three JAC combined cycle power islands demonstrates how established partnerships between utilities and energy providers can accelerate power generation capacity. These high-efficiency power islands will provide reliable and sustainable energy to meet Meta’s massive power needs. 3. Innovative Cost-Sharing Model: To accelerate power supply, Meta has agreed to cover the costs of new generation capacity while sharing the expense of maintaining existing system infrastructure. This cost-sharing model not only reduces rates for current customers but also sets a potential precedent for future agreements between utilities and data centers. 4. Remote Data Center Buildouts Enabled by Latency Tolerance: AI training centers like this one can tolerate higher latency, allowing them to be built far from traditional data center hubs and population centers. This opens the door for more geographically diversified data center expansion, which could alleviate congestion in major tech hubs.