The Enginuity Digest
Another week of developments and the world just keeps getting crazier (if Musk getting a $1 Trillion pay package agreement isn’t crazy then I don’t know what is). This week we take a focus on the energy challenges facing AI and the risks and opportunities this may present.
Here’s what will be covered in today’s newsletter:
News Update:
AI takes over process control for an entire gas treatment facility
Google shoots for the stars with project Suncatcher
China’s new AI model outperforms GPT-5
The AI Energy Crisis: The challenges, risks and paths forward
AI in Application: AI for Supply Chain Optimisation
What’s been happening with AI?
Aramco and Yokogawa Launch World’s First Fully Autonomous Gas Treatment Facility (Link) |
Fadhili Gas Plant (Credit: Aramco)
Saudi Aramco, the world’s largest integrated oil and gas company, commissioned the world’s largest fully autonomous control system at its existing Fadhili gas processing facility.
The system uses multiple collaborative agents that can independently determine optimal control strategies, including in contexts they have not previously encountered.
The implementation of this system involved three steps: firstly the development of a digital twin simulation of the plant operations, validation of the control system, and finally integration with the existing control system and safety infrastructure.
This advanced process control has been credited with a 10-15% reduction in amine and steam usage, along with a 5% reduction in power consumption.
Yokogawa has explicitly framed this project as moving from industrial automation to industrial autonomy.
Why this matters: This project is being hailed as a template for the next generation of self-governing industrial sites, demonstrating the importance of a simulation first approach and performance monitoring
Google’s Suncatcher Project Shoots for Space (Literally!) (Link) |
These projects really trying to make me feel like I’m already in the future
Google has announced “Project Suncatcher”, a bold (or crazy) idea to deploy modular, solar powered AI data centres in low Earth orbit.
Positioned in a sun-synchronous orbit (i.e. always in direct sunlight) these units will be operated with steady state generation, with energy output being about 8x higher per square metre than installations on land.
The extreme cold of space provides a passive source of cooling that can eliminate the need for water intensive thermal management, a key crisis facing current data centres.
Google’s AI chips survived radiation tests that equate to 5 years in space to address the hurdle most electronics face in space (where they typically fail within months).
The first trial run of this project involves two satellites being deployed as soon as 2027.
Why this matters: The buildout of AI has been massive across the globe and could soon reach the stars. A successful deployment of this project could unlock significant scale improvement with unlimited solar power, no grid limitations or community opposition.
China’s New AI Model Just Beat GPT-5 (Link) |
Meet Kimi K2 the new reasoning machine. This new open-source model (meaning people can tweak and change the internal parameters that make up models) can think through 300 steps straight without losing focus, outscoring GPT-5 on expert reasoning (44.9% vs 41.7%).
In a true David Vs Goliath moment, it was trained for just $4.6M, meaning K2 is competing with billion-dollar giants, and, in many benchmarks, winning.
Why this matters: Open-source Chinese models are now matching, and sometimes surpassing, closed US systems, signaling a major shift in AI innovation. Many are seeing this AI race between US and China as the new space race, but instead of rockets and egos on the line, it is intelligence.
The AI Energy Crisis
Current State of AI Energy
Artificial intelligence is driving an unprecedented surge in global electricity demand, reshaping grid pressures, community impacts, and engineering challenges worldwide. Data centers powering machine learning and generative AI now account for around 4% of global electricity use, and their demand is rising at a pace that few energy systems were designed to handle. If current trends continue, AI-related workloads could multiply total data center energy use tenfold or more by 2035, transforming energy grids forever.
Data centres come in different types depending on who owns them and how they’re used. Some are built and managed by one company just for their own needs (“enterprise”). Others let many organisations rent space, power and cooling in the same facility (“colocation”). Then there are huge cloud-provider sites built for global scale (“hyperscale”). Also emerging are “AI data centres”, facilities designed solely to train and run machine-learning and generative-AI models, featuring high-density compute, massive storage, and specialised cooling. Currently, AI workloads account for 5–15% of data centre energy consumption (20–60 TWh).
Challenges
AI-centric data centers are highly concentrated in regions with cheap power and favorable network access, imposing major stresses on local grids. In places like Northern Virginia, data centers consume 25% of the state’s electricity.
The grid struggles with the variability and spikes in AI demand, prompting utilities to rely on fossil-fuel-based backup generation, which drives up carbon emissions and water usage.
Communities near data center hubs face greater risks of outages, higher bills, and environmental burdens, unethically placing economic and environmental burdens on these communities.
Engineers are pressed to deliver new transmission solutions, more dynamic grid management, and rapid deployment that balance both peak demand and sustainability goals
Possible Paths Forward
One of the most ambitious ideas gaining traction is the concept of orbital data centers, like the Suncatcher project pictured above. While still experimental, major aerospace and energy firms are already conducting feasibility studies, aiming for small-scale prototypes by the 2030s.
Back on Earth, fusion power is once again being hailed as a potential long-term game changer. Breakthroughs from the Lawrence Livermore National Laboratory, Helion Energy, and Commonwealth Fusion Systems are pushing fusion closer to commercial reality than ever before.
However, these groundbreaking technologies won’t solve our energy challenges overnight. Commercial deployment remains years away, and AI’s rapid growth demands action now. In the meantime, the focus is shifting toward scaling up existing renewable and low-carbon energy sources, including wind, solar, nuclear, and geothermal, while simultaneously upgrading transmission lines, substations, and cooling infrastructure to handle the growing load. In the short term, natural gas will likely remain a necessary bridge, helping stabilize energy supply as cleaner technologies scale up. The true challenge will be ensuring that this expansion not only meets reliability and efficiency needs but also aligns with environmental standards and emission targets.
Ultimately, technology alone isn’t enough. Meeting AI’s soaring energy demand will require collaboration between communities, utilities, engineers, and policymakers to build smart grids, enhance energy storage, and embed sustainability at every stage of the AI supply chain. This represents one of the greatest challenges and opportunities of the AI era. With its abundant natural resources and innovation capacity, Australia is uniquely positioned to lead in the next generation of energy development. In the end, everything from politics and economics to now technology reverts around energy development.
AI in Application: AI for Supply Chain Optimisation
In modern chemical manufacturing, supply coordination is notoriously complex, with thousands of open orders, strict delivery windows, variable raw material availability, and production lines shared across products. Traditionally, production planners have had to rely on SAP reports, Excel macros, and late-night calls to keep shipments on track. JO.AI, however, has sought to tackle this challenge.
Enter Multi-Agent AI Systems:
Instead of one “super AI,” forward-thinking companies are deploying teams of specialised AI agents (individual AI models tackling individual tasks), each focused on a particular operational problem. Imagine an environment where:
One agent continuously monitors for supply chain disruptions (delays, material shortages, transport bottlenecks).
Another evaluates plant capacity constraints in real-time, checking for clashes in production orders.
A third reviews contractual deadlines, flagging any risk of missing milestones.
Others run simulations to optimise sequencing, proposing re-prioritisation to avoid bottlenecks.
Daily Life in an Agentic Supply Chain:
A practical example: At 7:45 am, as a planner logs in, the system notifies them—
"Potential delay for Order 100 due to a raw material shortfall. Estimated risk of breach: 78%. Recommended action: prioritise order 110, which uses available stock. Execute?"
Behind the scenes, multiple agents have worked together overnight, identifying threats and simulating solutions. The human oversees and approves actions, spending less time on firefighting and more on strategic oversight.
Benefits and Challenges:
Multi-agent AI makes operations:
Proactive instead of reactive: agents detect and solve problems before they escalate.
Collaborative: AI never works alone; humans always approve and guide decisions. AI is simply powerful
Scalable: Once proven, these agent teams can be rolled out across dozens of sites, as with JO.AI’s deployment in 50+ plants.
Through this all however, it is critical to ensure robust data governance, maintain accuracy and thus ensure operator trust. In the chemical sector, where safety and compliance reign, strong human oversight is, as always, critical.
In Other News
Amazon Web Services to launch a subsea cable linking US and Ireland by 2028 to deliver 320 terabits per second of capacity, following their recent outage (Link).
Apple to pay Google $1B a year for AI models to upgrade the Siri product, a much-awaited overhaul (Link)
Following a shareholder vote with over 75% approval, Tesla approved a $1T pay package for Elon Musk. Across 12 market-cap steps (from $1.4T today to $8.5T), and delivering millions of robots, robotaxis and vehicles, this makes the world’s richest man possibly even richer (Link)
Google DeepMind Ai finds new solutions to previously unsolved math problems relating to Nikodym sets (Link)
This newsletter seeks to engage and challenge the way engineers see AI and its potential for application in industry. Any thoughts, questions or arguments are welcome! Finally, if you enjoy the content, please refer it on to your friends and colleagues.
