The Road to AGI: Building the Compute Foundation for Tomorrow’s Intelligence
What if the road to artificial general intelligence (AGI) mirrors human development?
Ian Bratt, vice president of Machine Learning (ML) Technology and Fellow at Arm, believes it might. In early life, our brains move from sensing to speaking, then toward thinking and planning. This is known as the neuroplasticity curve, and Bratt says AI is following a similar path, and that we’re now at the “cusp of cognition.”
We are seeing that evolution today. Current AI models are more capable than ever. They can simulate reasoning, complete complex tasks, and process text, vision, and sound, with many already being embedded in everyday tools and systems. As AGI matures, it will operate across cloud servers, mobile devices, vehicles, and industrial systems. That requires compute platforms that scale efficiently and run intelligently, wherever needed.
To understand what comes next, The Road to Artificial General Intelligence Report by MIT Technology Review Insights and Arm explores the path forward, and why it may look more human than we expect. The report leverages insights from thought leaders in AI research and governance to map the strategic challenges and opportunities ahead.
AGI is Closer than it Seems, But Still Far from Easy
According to global aggregate forecasts led by the nonprofit organization AI Impacts, AI systems may begin achieving AGI milestones as soon as 2028. These capabilities include problem-solving, goal-directed reasoning, and the ability to outperform humans in certain tasks unaided.
But even the most advanced models today struggle with adaptability. As the MIT report notes, they still fall short in areas like spatial reasoning, motor control, social awareness, and creativity. True AGI will need to bridge these gaps and unlock new ways to process, transfer, and scale intelligence across systems.
Intelligence Demands Architectural Efficiency
Progress toward AGI is accelerating, but so is the cost. Before 2010, AI compute requirements doubled every 21 months and since the rise of deep learning, that rate has increased to once every 5.7 months. Models will become more power-hungry as they grow, with one estimate placing the future energy demand for AI-scale workloads at more than 20 terawatts, which is close to the entire global electricity generation capacity today.
As Bratt says, “if we want to enable that far future where everything is cognitively enhanced, it’s going to be a huge step function increase in compute.” Meeting that demand will require architectural advances that optimize latency, bandwidth, and energy use, and not just raw speed.
Heterogeneous Compute: A Practical Path to AGI
Rather than focusing on one type of processor, the MIT report highlights heterogeneous compute as a scalable and balanced approach to AGI. This model combines CPUs, GPUs, NPUs, and accelerators, each tuned for different AI tasks. It allows compute to happen where it makes the most sense, whether on-device, at the edge, or in the cloud.
Across the technology ecosystem, the Arm compute platform is already powering billions of connected devices across the globe, offering the power-efficiency, performance and scalability required to enable intelligence at every level of the AI stack. For instance, close to 50 percent of the compute shipped to top hyperscalers in 2025 will be based on Arm. Arm technology is also enabling intelligence to run in smartphones, vehicles and industrial systems around the world – reducing latency, preserving privacy, and lowering bandwidth costs.
Software and Standards will Unlock the Next Wave
Hardware alone will not be enough. AGI will also require orchestration – the ability to manage complex, distributed AI workloads across compute environments. That means software frameworks, scheduling tools, and open standards that allow models to adapt, scale, and interoperate.
According to the MIT report, AI systems will be required to become context-aware and dynamically distributed. Developers need platforms that help simplify development across many chip types without the need to rewrite the code. Ecosystem collaboration will be key to this vision.
Arm’s approach is already aligned in this direction, democratizing AI for developers and cementing the Arm compute platform as a capable backend for common AI frameworks, enabling billions of Arm-based systems to run inference workloads efficiently to bring AI to the masses.
A Smarter Future Needs Smarter Foundations
Tomorrow’s breakthrough in AI will depend on the architectural choices we make today. Arm is delivering the foundations that let AI evolve, scale, and reach its full potential, responsibly, and efficiently.
The future of AI is being built on Arm.
The Road to AGI
Learn more on how the architecture will shape the path to AGI in the latest MIT Technology Review Report
Any re-use permitted for informational and non-commercial or personal use only.
Editorial Contact
Latest on X
A huge thank you to everyone who joined us at the grand opening of our new office space in Bengaluru. 🎉
From inspiring messages to vibrant celebrations, it was truly a day to remember, marking a milestone for Arm India and the future of global innovation.
We’re proud to be a technology partner for Stargate UK, marking a major step in expanding Britain’s AI computing power and digital infrastructure.
We look forward to continuing to deliver scalable, power-efficient AI that enables breakthroughs from the UK to the world.…
A new benchmark report from @Signal_65 just dropped!
In this analysis, Arm-based @awscloud Graviton4 instances didn’t just outperform x86 on key general compute and AI workloads, but they did so at a much better price-performance ratio.
⬇️⬇️⬇️
https://signal65.com/research/arm-neoverse-enables-leading-cloud-performance-and-cost-efficiency-with-aws-graviton4/
Since 2015, the Arm + @UNICEF_uk partnership has supported more than 770,000 children, young people, and their communities worldwide. 🤝
This video captures 10 years of collaboration, impact, and shared commitment to building a better future. 🎥
Today, PSA Certified enters a new era.
From 7 founding partners in 2019 to 250+ certified products, it’s now under @GlobalPlatform_'s governance, ready to scale and evolve.
We look forward to continuing to support PSA as part of a new working group: https://okt.to/zbwupr
Last week Arm SVP & GM of Infrastructure, Mohamed Awad, spoke at #AIInfrASummit to share our vision for the future of compute: performance, efficiency & an open ecosystem driving the next generation of AI experiences for everyone, built on Arm.
👀:
Great conversation with @electronicspec about the growing importance of vehicle updateability in the era of software-defined vehicles (SDVs) and what this means for modern automotive compute platforms.🎙️
🎧 Listen to “Arm-ing the software-defined car" episode here:…
The @Arm Keil Studio Webinar Series covers everything from project set up to ML at the edge, helping you build, debug, and deploy faster.
✅ Watch 3 sessions on-demand
✅ Don’t miss our upcoming live sessions (Sept 16 & Sept 30)
Register now:
Celebrating the launch of the Arm Lumex CSS Platform. 🎉
Chris Bergey kicked off Arm Unlocked Shanghai by introducing our brand new platform, built for the next era of on-device AI, joined by @AliPay and vivo to share more on SME2 benefits.
A full house at @IAAmobility for a powerful panel on Software-Defined Vehicles 🚗
Dipti Vachani joined leaders from @BMWGroup, @Google, Sonatus, @Accenture, and @Automotive_News to explore how SDVs are reshaping the industry.
📸 Highlights from the session below. #IAA25
🆕 Meet Arm Lumex, our most advanced CSS platform for flagship smartphones and next-gen PCs.
Uniting SME2-enabled CPUs with new GPUs and system IP, it delivers up to 5x faster AI performance, fueling smarter, quicker experiences on billions of devices: https://okt.to/Uc3koJ
Hello #AIInfraSummit 2025📢
Come say hello to the team at booth 408 to check out our demos and take a closer look at how we're creating the foundation for AI infrastructure - focused on efficiency and performance. 💪
With an ecosystem of innovators building powerful and efficient AI Infrastructure solutions - the future potential of AI is astounding!
In this episode of TechArena, Mohamed Awad breaks down the datacenter transformation ahead of the #AIInfraSummit.
👂: https://okt.to/iSuVd6
AI is reshaping the data center. Arm is at its core.
This week we're excited to be at the 2025 #AIInfraSummit in Santa Clara - for those attending be sure to catch our sessions and join us as we talk about data center infrastructure for AI!
https://okt.to/tA1CG9
Congratulations to @alifsemi, the first silicon provider to offer the Arm Ethos-U85 NPU supporting transformer-based AI at the edge.
Generative AI is raising the bar for intelligence beyond the cloud, demanding more performance, privacy, and efficiency. 💪…
Discover how generative AI and Arm technology are transforming industrial performance.
In factories, every second counts. Siemens is using Arm-powered AI at the edge to predict issues and keep critical assets running at peak performance.
https://okt.to/QBG1Nd
The Cloud Forward Arm Virtual Summit is happening next week! 🥳
Register now and get excited to reimagine your cloud strategy. You'll hear about the powerful innovations enabled by Arm architecture from the innovators and leaders leading the way.
🔗:https://okt.to/N95qDo
Arm is building the foundation for tomorrow's technologists.
After joining the White House's Pledge to America’s Youth - today we're announcing the Arm EducateAI Coalition to empower American students with world-class AI education. 🧑🎓🧠
Announcing the Arm EducateAI Coalition
Arm launches the EducateAI Coalition to expand AI education across the U.S., training teachers and empowering students...
okt.to
Excited to see @Acer’s new Chromebook Plus Spin 514 🙌
Powered by @MediaTek’s Kompanio Ultra SoC which is built on the Armv9 architecture, the laptop boasts fast AI processing, richer graphics with the Arm Immortalis-G925 GPU, and up to 17 hours of battery life.…
Next week, Dipti Vachani will take the stage at @IAAMobility alongside leaders from @BMWGroup, @Google, Sonatus, @Accenture, and @Automotive_News to discuss how the industry is unlocking the full potential of software-defined vehicles.
Join us! 🚘 #IAA25 https://okt.to/TYmnMl
👏 Congratulations to our partner @Acer on the launch of the new Chromebook Plus Spin 514.
Built on the Arm compute platform with MediaTek’s Kompanio Ultra, it delivers AI-ready performance, immersive graphics, and industry-leading battery life, all in Acer’s thinnest Chromebook…
Meet the NEW #AcerChromebookPlusSpin514 🤩 Powered by MediaTek Kompanio Ultrafor next-level AI productivity, with a brilliant 2.8K touch display, and up to 17 hours on a single charge, it was built for speed and style. EPIC 💥
👉 https://www.acer.com/acer-chromebook-plus-spin-514
In F1®, precision starts long before race day. 🏁
At the @AstonMartinF1 HQ, the CoreWeave wind tunnel uses Arm-powered, state of the art technology to help interface with over 1,000 sensors simultaneously to turn data into faster, smarter decisions.
Less time validating. More…
We're proud to provide partners like @awscloud with cost efficiency and performance for the moments that rely on it most.
Like Arm-based AWS Graviton, which powered more than 40 percent of the Amazon EC2 compute used for Amazon Prime Day 2025. ⚡🛍️
https://okt.to/XNG0Cv
Our CEO Rene Haas joined @CarnegieMellon President Farnam Jahanian to open the 2025–26 President’s Lecture Series.
They discussed leadership, career lessons, and how AI will drive the next wave of innovation.
📸 Highlights from the event below
