By The Non-Von Team

There’s rarely a conversation where AI isn’t mentioned. It’s advancing fast.That’s no surprise. What gets missed in most conversations is the infrastructure and the fact that we’re still trying to run these next-gen models on last-gen infrastructure.

The result is a growing mismatch between what AI is capable of and what current hardware can actually support. This gap is becoming one of the biggest barriers to progress. And it’s time the conversation shifts to focus on proactive solutions, like new chip technology, vs. AI performance.

Why AI Progress Is Running Into a Wall

Over the last few years, AI models have changed in ways that aren’t just noticeable, they’re foundational. They’ve grown in size, taken on more sparsity, become more dynamic, and now run across a wider range of environments from massive cloud clusters to edge devices with strict power limits.

These aren’t surface-level updates or new product releases.  They reshape what compute infrastructure needs to support.

Today’s models don’t behave like the dense, uniform workloads existing hardware was built for. They include conditional logic, multimodal inputs, and irregular memory access patterns. They shift and adapt during inference. And yet, most of them are still running on systems optimized for raw throughput not for the complexity they actually bring.

The Hardware Fit Is All Wrong.

General-purpose GPUs are being relied upon to run advanced AI models. That’s a problem because these chips were built for graphics, not intelligence. They follow the classic Von Neumann architecture, (where memory and compute are physically separate) which means data has to move constantly just to get basic work done.

That back-and-forth burns energy, adds latency, and turns into a real bottleneck as models grow more complex. And because GPUs don’t natively support unstructured sparsity, they tend to ignore it, leaving efficiency gains on the table.

The degradation in performance is one of the issues AI operations today faces. More concerning is the fundamental mismatch between how today’s AI models behave and the capabilities of the hardware they run on was built to do.

How Infrastructure Gaps Show Up in the Real World

​These architectural mismatches show up quickly in the real world. Inference latency climbs even on high-end hardware. Power then becomes a constraint, especially in edge environments, mobile devices, or anywhere thermal budgets are tight.

Models are pruned to be lighter, but current hardware forces them back into bulky formats canceling those gains. And as developers, engineers, and researchers continue to innovate, they’re still limited by the reality of existing hardware unless AI chips built for this workload are readily available.

Why Non-Von Took a Different Approach
At Non-Von, we’ve been thinking about the need for improved hardware. Constanty. We’ve seen the shift in AI models and knew that existing hardware was not suitable for the AI revolution. Instead of squeezing more out of an outdated architecture, we started over, and designed a chip for how modern models actually compute.

Our chips are sparse-native, meaning they process unstructured sparsity directly. Pruned models don’t need to be reshaped or re-densified. They just run faster and more efficiently. They do what is required to support the AI boom. 

Non -Von chips also keep compute close to memory. That reduces the overhead of constant data movement and helps eliminate the Von Neumann bottleneck entirely. 

And we made it easy to use. Our software stack works with PyTorch, ONNX, and TensorFlow. With just a few lines of code, teams can plug into Non-Von’s acceleration without restructuring their models. We’re working on the solution for future AI operations. 

Where Progress Stalls

​It’s not just about speed or power efficiency. It’s about whether the next generation of AI breakthroughs will be possible at all or whether they’ll get stuck waiting on infrastructure that wasn’t built to support them.

If we want to keep pushing what’s possible, we can’t keep asking outdated systems to carry the load. We need to rethink what AI hardware should look like and align it with the models we’re actually building.

​

By The Non-Von Team

For centuries, the global economy inched forward, just 0.1% growth a year until 1700, doubling only once in a millennium. Then the Industrial Revolution rewrote the rules pushing growth to 2.8% annually in the 20th century  doubling output every 25 years. Now, AGI (Artificial Growth Intelligence) could be the next inflection point.

Today’s AI leaders believe we’re on the edge of something bigger than the Industrial Revolution. They see AGI driving global GDP growth to 20–30% a year or more, which is a  pace that would have been unthinkable in any other era. It sounds extreme, and by historical standards it is. But so did the idea of sustained economic growth… until steam, steel, and machines made it real.   (Section 1; The Economist, 2025)


AI could push GDP growth to unprecedented levels.

What could fuel growth on this scale? AGI’s ability to accelerate innovation without adding more people to the equation. In the past, more population meant more ideas. With AGI, the dynamic shifts. Progress can compound on its own, with technology refining and advancing at speeds far beyond human limits.

Some projections go well past simple automation. As AI approaches AGI, whole segments of the global economy could shift to machine-led production. But the real tipping point isn’t about replacing human labor in predictable, repetitive work, it’s when AI starts improving the very technology it runs on. 

AGI isn’t just about speeding up routine tasks.  It can take on the big, messy work humans spend years untangling. End-to-end project management. Driving complex research forward. Even running entire labs without a human in sight. That kind of capability sets up a chain reaction: breakthroughs building on breakthroughs, each one laying the groundwork for the next. Economists call it the “ideas create more ideas” effect - a compounding loop with no real ceiling on how fast progress can move. (Section 2; The Economist, 2025)

This leap only happens if hardware can keep pace.

Transformative growth comes with real-world limitations. A highly automated economy can’t run on ambition alone,  it needs “sufficient energy and infrastructure,” and that means heavy investment. For AGI to manage projects and drive scientific discovery, it will take a step-change in computational power and efficiency.  (Section 2; The Economist, 2025)

As economist Tyler Cowen puts it, “the stronger the AI is, the more the weaknesses of the other factors bind you,” with energy and data often topping the list. Hardware is squarely in that crosshairs. Models from Epoch AI estimate that optimal investment in AI for this year could hit $25 trillion,  roughly 50 times today’s spend.(Section 3; The Economist, 2025)

Most AI still runs on GPUs, chips built for graphics work, not AI workloads. They draw huge amounts of power and demand massive electrical budgets. That inefficiency is a growing restriction-one that stands between today’s capabilities and the kind of economic acceleration AGI promises.

But potential alone doesn’t guarantee progress. To turn AGI’s promise into reality, the hardware behind it has to scale just as fast and that’s where the real barriers begin to show.

How smart chips solve key scaling limits

This is exactly where advanced chip design steps in. Our novo architecture isn’t a retrofit. It’s built for AI from the start. By rethinking how data moves and how power is used, these chips deliver a leap in processing performance while slashing electrical consumption.

Memory Access:
Traditional computing keeps hitting the same wall: the von Neumann bottleneck. Non-Von’s novo architecture removes it from the equation. By giving each core its own dedicated memory instead of forcing them to share a single, centralized pool, it eliminates memory blocks, cuts power demands, and strips out the extra computational and electrical overhead that slows conventional designs.

Energy Efficiency:
Our architecture is engineered to cut energy use without compromising performance. It matches the throughput of a full-scale data center while running on the power footprint of an edge device-delivering up to 100x greater power efficiency on workloads ranging from deep-learning vision models to transformer-based chatbots. That kind of efficiency lets entire server farms run faster while drawing only a fraction of the power GPUs consume today. (See Hokenmaier et al., 2024)

In addition, the SCULPT toolset automatically adapts and streamlines models for novo hardware, removing the inefficiencies that slow GPU execution and unlocking significant cost reductions at scale.

Scalability:
Our chips are purpose-built for the latest wave of AI architectures. Sparse and quantized models like BitNet, HNet, sparse transformers, and FP4 design approaches that are quickly gaining traction across the industry. Because the architecture anticipates where software is headed, performance advantages grow as models get newer and more advanced.

A key differentiator is the ability to handle unstructured sparsity, an inherent feature of the development process for most AI models. This gives engineers the freedom to tap into any form of sparsity for substantial efficiency gains. This is a capability that has only recently become practical at scale. The novo platform also offers broad compatibility, able to take an existing trained model and automatically convert it into a sparse, highly compact form directly in silicon, with full support for major frameworks like PyTorch, ONNX, Keras, and TensorFlow. This flexibility ensures our  hardware stays aligned with evolving model complexity and compute demands.


Without these chips, growth projections remain theory.

If computing hardware doesn’t evolve to overcome today’s power demands, memory constraints, and scaling limits, the economic boom envisioned by AGI optimists may never materialize. The shift toward fully automated global production and the massive build-out of data centers it requires, depends on chips that can deliver exceptional processing performance while sharply reducing energy use.

Our chip is built to meet those exact challenges, positioning it as purpose-made hardware for advanced AI systems and a potential catalyst for realizing AGI’s full economic impact. 

​History offers a hint of what’s possible: AI’s progress has repeatedly outpaced industry predictions, and if the hardware keeps pace, the next wave of breakthroughs could arrive far sooner than expected.

Sources:

The Economist. (2025, July 24). What if AI made the world’s economic growth explode? The Economist. https://www.economist.com/briefing/2025/07/24/what-if-ai-made-the-worlds-economic-growth-explode

Hokenmaier, W., & Jurasek, R. (n.d.). Co-design of a novel CMOS highly parallel, low-power, multi-chip neural network accelerator. Semiconductor Hardware Development, Green Mountain Semiconductor Inc. https://arxiv.org/pdf/2409.19389

by: the Non-Von team

The age of general-purpose compute is over—or at least it should be. As AI becomes foundational to innovation in nearly every sector, our continued reliance on GPU-based architectures and traditional Von Neumann designs is fast becoming a critical bottleneck. From healthcare and climate modeling to edge AI and real-time personalization, we risk throttling transformative progress not due to lack of imagination—but because our hardware simply can't keep up.

The Rise—and Risk—of AI WorkloadsAI workloads have grown dramatically—both in how complex they are and how much computing power they need. Training large-scale AI models now demands compute measured in exaflops and runtime stretching across days or weeks (Choi et al., 2024). Despite their parallelism, GPUs are still bound to the legacy Von Neumann limitation, where computation and memory are physically separated—requiring vast amounts of data shuttling between the two.
This architecture leads to severe inefficiencies, especially for AI models with high sparsity or irregular data access patterns. In fact, many AI accelerators experience energy utilization rates below 20%, with over 80% of energy lost in data movement or unused compute pathways (Yu et al., 2024; Zaman et al., 2021).
Compute Waste Is Not Just Technical—It Has Real-World ConsequencesThese inefficiencies have broader consequences:

• Healthcare AI models can’t run effectively on portable devices due to power-hungry hardware (Zaman et al., 2021).
  
  
• Edge AI systems—like drones or autonomous robots—struggle with latency and thermal limits that GPUs can't overcome (Ni et al., 2020).
  

​

• ​Climate simulation and smart infrastructure are slowed down by data throughput limits and memory barriers (Gómez-Luna et al., 2023).

The Von Neumann constraint—once just a theoretical limitation—is now a critical barrier to progress.
What Makes Non-Von Different?Emerging chip architectures like the one created by  Non-Von are not just faster—they're architected for AI's unique needs.

• Unlike legacy chips, Non-Von embraces computing-in-memory (CIM) designs, eliminating the data shuttling problem and directly addressing the Von Neumann bottleneck.
  
  
  
• Its architecture is sparse-native, meaning it efficiently handles AI models with irregular patterns, achieving 10–100× better energy efficiency than standard GPUs
  
  
  
• Because it’s designed with distributed, memory-proximal compute blocks, Non-Von enables real-time inference on the edge—critical for robotics, AR, and mobile AI.
  
  
  

In other words, Non-Von isn’t trying to make yesterday’s architecture faster—it’s reimagining how AI compute should be done from the ground up.


The Innovation RoadblockIf scientists and developers don’t adopt these AI-native chips soon enough, we’ll face a compute-driven innovation ceiling. That can have direct impacts on daily life. Here are some examples:

• Personalized medicine delays  due to off-device processing constraints.
  
  
• Urban intelligence (traffic, energy, logistics) limited by cloud latency and power draw.

​

• ​Scientific modeling and simulation (climate, fusion, epidemiology) constrained by unscalable compute costs.

​​
As researchers Lee et al. (2020) and Zhang et al. (2023) emphasize, next-gen AI needs next-gen infrastructure—or it simply won’t happen.
Final ThoughtThe world is not running out of data. We’re not running out of models or brilliant ideas. What we’re running out of is hardware capable of realizing them.


The future won’t be limited by AI—it’ll be limited by compute.
And it’s time to choose architectures that are built for what’s coming, not what came before.
📚 Cited Sources

1. Choi, J., Kim, S., & Kim, D. (2024). PIM hardware accelerators for real-world problems. Advances in Computers, Elsevier. Link
   
   
   
2. Zaman, K. S., Reaz, M. B. I., & Ali, S. H. M. (2021). Custom hardware architectures for deep learning on portable devices: A review. IEEE. Link
   
   
   
3. Yu, K., Kim, S., & Choi, J. R. (2024). Computing-in-Memory for Neural Networks: Review. IEEE Access. Link
   
   
   
4. Ni, K., Keshavarzi, A., et al. (2020). Ferroelectronics for edge intelligence. IEEE. Link
   
   
   
5. Lee, J., Kang, S., et al. (2020). Energy-Efficient DNN Processor on Edge Devices. IEEE. Link
   
   
   
6. Zhang, C., Sun, H., et al. (2023). Survey of memory-centric computer architecture. IEEE. Link
   
   
   
7. Gómez-Luna, J., Guo, Y., & Brocard, S. (2023). Evaluating ML workloads on memory-centric computing systems. IEEE. Link
   

By: Non-Von team

The Compute Crunch
We’re living in an era shaped by AI. Every swipe, voice command, and AI generated search  requires an algorithm that needs serious compute power to function. The adoption of AI in almost every facet of daily life requires an advanced computing infrastructure that is hitting power and efficiency limits fast.

With AI models ballooning in size and cloud services demanding more processing, we need to ask: Is there a smarter, more efficient way to compute?

Where We Are: The Von Neumann Bottleneck

​The dominant architecture in computing, Von Neumann, requires data to shuttle back and forth between memory and processing units. This constant movement causes what engineers call the Von Neumann bottleneck—a major drag on speed and energy efficiency.

Every:

• memory fetch
• processing cycle
• and storage operation

…adds to the energy bill and slows down the pipeline.

Smarter Computing: Rethinking the Flow

This bottleneck isn’t a technical inevitability—it’s an architectural legacy. Researchers and chipmakers are now challenging this model to build something radically better.

This is how Innovation is redefining how we compute:

• In-memory computing: Reduces the distance data travels, making processing much faster and leaner.
  
  
• Neuromorphic chips: Mimic the brain to handle tasks in parallel, especially useful for perception and AI.
  
  
• Reconfigurable computing (FPGAs): Tailor chip behavior to a task, reducing overhead.
  
  
• Event-driven computing: Only wakes up the chip when needed, saving battery on edge devices.
  
  

These solutions eliminate unnecessary steps, compressing the compute pipeline and reducing power use dramatically. Advancements like these are necessary as AI continues to be more commonplace in today’s society. 

Why This Matters: Power, Cost, and the Planet

AI is rooted in innovation, therefore the chips required to support the AI algorithms must also stem from innovation. Creating a solution that can support AI operations and the planet is the only sustainable way to expand this technology. Let’s  ground this in reality:

• Data centers already consume ~3% of global electricity, and this number is projected to rise rapidly
  (Mo, R. et al., 2025, IEEE Transactions on Cloud Computing)
• Large AI models like GPT-4 can require hundreds of megawatt-hours to train.
• Cooling infrastructure in data centers is now a multi-billion dollar industry, driven by excess heat from inefficient compute architectures.
  
  

In short, the status quo is unsustainable. Smarter compute architectures can help:

• Reduce global power demand
• Extend battery life for mobile and IoT devices
• Cut infrastructure and cloud costs
• Lower carbon footprints across industries

The Awareness Gap: Why Aren’t More People Talking About This?

Most developers and product teams focus on frameworks and cloud providers. Rarely do conversations zoom in on chip architecture. That’s a problem.

Why the blind spot?

• Hardware choices are abstracted away in modern dev environments.
• Chip startups don't get as much airtime as cloud providers.
• Power and efficiency improvements are harder to market than speed benchmarks.
  
  

But efficiency isn’t just a backend concern anymore—it’s becoming mission-critical for everything from mobile UX to sustainability.

Compute Consciousness

We need a shift in mindset from “how much can we compute?” to “how well are we computing?”

That means:

• Asking vendors about architecture-level power optimizations.
• Supporting chipmakers that build in-memory and neuromorphic solutions.
• Investing in software stacks that play nicely with streamlined computing hardware.
  
  

The next great leap in computing won’t be more of the same—it’ll be less wasteful and more aware.

Streamlined Compute is the Future

Efficiency isn’t an optimization—it’s a strategic priority. Streamlined compute architectures that cut out unnecessary processing steps are quietly reshaping the future.

​The real breakthroughs aren’t always louder or bigger—they’re smarter.

Non-Von is such a breakthrough. We eliminate waste across the AI landscape.  Our novel architecture and supporting software platform make it easy to be more efficient.  We enable model builders to leverage any kind of sparsity, including unstructured sparsity (something GPU’s cannot do), this means that the entire post-pruning model conversion step can be eliminated! 
​

Our architecture and SDK also enable the elimination of wasted compute at the granular level of computations.  In the simple dot-product below, a common function in matrix algebra, we can eliminate 5 of the 9 operations.  Scaled to the AI model level we are talking about millions of calculations eliminated, saving compute across the board!

By: The Non-Von Team
​
In order to scale artificial intelligence, most companies default to buying off-the-shelf chips from the biggest players in the industry, or wait in line—sometimes indefinitely—for access to cloud based GPUs that are already in short supply.But there’s a third, smarter option gaining traction: partnering with small, R&D-driven companies to co-develop purpose-built AI technology that fits your needs exactly—rather than retrofitting your goals to someone else’s roadmap.

Why Small Companies Are Often Leading InnovationWhile big tech companies have size, funding, and brand recognition, smaller R&D-focused companies often lead in true innovation. Here’s why:

●      They move fast: Small companies aren’t weighed down by legacy systems, bureaucracy, or long internal approval cycles. They can test, iterate, and pivot faster—making them ideal environments for disruptive breakthroughs.

●      They take more risks: Without shareholders to appease or massive product portfolios to protect, small firms are freer to experiment—and often push boundaries where larger companies play it safe.

●      They go deep: Big companies need solutions that work at massive scale, which often means general-purpose hardware. However, small teams tend to focus narrowly—building purpose-built solutions that outperform generalized systems in specific use cases.

●      They’re closer to the edge: Whether it’s a defense contractor working in contested environments or an energy company deploying AI in the field, small companies are more likely to understand niche needs and design around them directly.

●      They recruit specialists, not generalists: Research-driven companies often attract top-tier engineers, scientists, and technologists who are laser-focused on solving a specific class of problems—rather than maintaining legacy architectures.

Partnering with a Focused R&D Team Gives You:Whether you're a government agency, enterprise AI buyer, or advanced robotics company, collaborating with a small innovation partner offers strategic advantages:

●      Early access to specialized tech: Work directly with the engineers shaping the product roadmap. Gain access to performance features, form factors, or integrations not available off-the-shelf.

●      A collaborative build process: Rather than forcing your team to adapt to rigid hardware specs, you co-develop a solution that fits your unique constraints—whether that’s power, latency, size, or interoperability.

●      More flexible IP and licensing terms: With smaller teams, there’s more room to negotiate IP ownership, white-label agreements, or custom integrations tailored to your business model.

●      Faster time-to-value: Because smaller R&D teams are more agile, the feedback loop between idea and implementation is shorter. You’ll see results—and iterate—faster than you would with a traditional vendor relationship.

●      Shared innovation without vendor lock-in: You get access to cutting-edge performance and long-term support, without being tied to the roadmap of a massive, company that isn’t optimizing for your needs.

The Strategic Advantage: Beyond the Supply ChainEven AI giants are struggling to secure enough compute. Both Meta and OpenAI recently reported significant GPU shortages—slowing development and forcing teams to wait in line for access to critical infrastructure (Quartz).

If companies investing $60B+ annually in AI can’t get the chips they need, how will the rest of the industry compete?

It’s time to rethink the approach. Partnering with small, research-focused companies offers a chance to break free from infrastructure bottlenecks and develop AI chips on your own terms.

Non-Von: Built for What’s Next
At Non-Von, we combine rigorous research with real-world execution. Our architecture is designed to support advanced workloads where conventional chips fall short—especially in edge and real-time environments. Our goal is to partner with forward-thinking organizations to design and deploy solutions that not only work today, but scale for tomorrow.

If you’re tired of buying technology that’s not built for your needs—partner with a team that’s building with purpose.
Sources

1. Quartz: Meta and OpenAI say they don't have enough AI chips
   
   
2. McKinsey & Company: How data centers and the energy sector can sate AI’s hunger for power
3. IEEE Spectrum: Why custom AI chips are the future

February 2025
​By: The Non-Von Team
The rise of artificial intelligence (AI) is disrupting up industries, fuelling innovation, and changing the way organizations handle data. At the heart of this transformation are AI chips—specialized processors built to tackle the heavy lifting of modern AI applications. However, many organizations are still relying on traditional processors for their AI needs, and that’s a choice with risks. From edge computing failures to data bottlenecks, inefficient operations, and even power outages from overtaxed hardware, sticking with outdated processors is becoming more problematic every day.

1. Failed Edge Computing: A Roadblock to Real-Time Decisions
Edge computing brings data processing closer to the source—whether that’s an autonomous vehicle, an industrial machine, or an IoT device. This setup enables quick decision-making right where the data is being generated, without relying on distant servers. However, traditional CPUs and GPUs often fall short when it comes to the low-latency and power efficiency that edge applications demand. Without AI chips, systems can face:

●      Higher latency: Traditional processors struggle with the parallel processing needed for real-time tasks, causing delays in critical decisions (Google + Deloitte Impact Report).
●      Increased failure rates: Many edge devices can’t handle continuous computation on older processors, leading to downtime during important operations (Google + Deloitte Impact Report).

These limitations show just how outdated hardware can hold back edge devices, potentially causing costly errors or system failures. Upgrading to AI chips ensures the real-time performance needed for those critical edge applications to run smoothly.

2. Stranded Data: The Untapped Resource
With traditional processors, the sheer volume of data generated in modern systems often overwhelms the hardware’s ability to process it. This leads to stranded data—valuable information that remains unanalysed due to insufficient computational capacity.

●      Data Generation Outside Traditional Data Centers: Analysts predict that by 2025, a significant portion of enterprise-generated data will be created and processed outside traditional centralized data centers or clouds. This shift underscores the growing importance of edge computing in handling the increasing volume of data generated by enterprises (Comarch)
●      Impact on Industries Like Healthcare: The healthcare sector, among others, is expected to benefit from real-time data analytics enabled by advanced processing capabilities. AI algorithms can analyse electronic health records (EHRs) and genomic data to select the most suitable patients for trials, thereby increasing the success rate of the trials (MDPI).

As data generation continues to expand outside traditional data centers, industries like healthcare stand to gain immensely from the power of edge computing. The ability to process data in real time will not only reduce the burden on centralized systems but also enable more informed, timely decision-making.

3. Inefficient Operations: Ballooning CostsAs AI workloads continue to grow, relying on traditional processors like GPUs and CPUs is leading to operational inefficiencies. Organizations are facing rising energy consumption and maintenance costs as a result. These inefficiencies put a strain on resources and make it harder to scale and maintain sustainability.

●      Energy Consumption: Traditional GPUs and CPUs use a lot more power than specialized AI accelerators. For example, AMD's MI250 accelerators pull 500 W of power, with peaks reaching up to 560 W, while the MI300x uses a hefty 750 W (Forbes).
●      Over-Provisioning: To meet performance demands, organizations often allocate more resources than needed—this practice is called over-provisioning. It leads to higher costs, especially in cloud environments where it can result in significant overspending (The Register).

By switching to AI-specific chips, organizations can cut down on energy consumption, lower operational costs, and set themselves up for long-term efficiency and scalability.

4. Blackouts Due to Power ConsumptionThe rapid growth of artificial intelligence is driving up energy consumption, especially when it comes to training large AI models. Traditional processors, like GPUs, are major contributors to this energy demand, which has significant environmental consequences.

●      Carbon Emissions: A study from the University of Massachusetts Amherst found that training a single large AI model can emit over 626,000 pounds of CO₂—roughly equivalent to the lifetime emissions of five cars (MIT Technology Review).
●      Data Center Vulnerabilities: Data centers using outdated hardware are increasingly vulnerable to power outages and system failures. This is due to surges in energy consumption during heavy AI processing tasks. A report even predicts that U.S. data center power usage could nearly triple by 2028, highlighting the strain on existing infrastructure (Reuters).

By implementing specialized AI chips, like Non-Von’s, organizations can improve power efficiency, reduce their environmental impact, and boost the reliability of their data center operations.

The Future of AI Hinges on Specialized Chips
The risks of sticking with traditional processors instead of switching to AI chips are simply too big to overlook. Failed edge computing, stranded data, inefficient operations, and blackouts aren’t just technical problems—they’re business-critical issues that can throw entire industries off track.

Organizations that make the switch to AI chips now will gain a competitive edge, driving operational efficiency, cutting costs, and securing a more sustainable future.

Sources:

1. Google + Deloitte Impact Report. "Edge Computing: The Key to Real-Time Decisions." Google Cloud Blog, 2023, https://static.googleusercontent.com/media/www.google.com/en//about/datacenters/static/pdf/google-va-impact-report.pdf .
2. MDPI. "The Role of AI and Real-Time Analytics in Healthcare." MDPI Future Internet, vol. 16, no. 9, 2023, www.mdpi.com/1999-5903/16/9/308.
3. Comarch. "Top Data Center Trends 2025: Global Overview." Comarch ICT Solutions, 2023, https://www.comarch.com/trade-and-services/ict/news/top-data-center-trends-2025-global-overview/.
4. Kindig, Beth. "AI Power Consumption Rapidly Becoming Mission-Critical." Forbes, 20 June 2024, www.forbes.com/sites/bethkindig/2024/06/20/ai-power-consumption-rapidly-becoming-mission-critical
5. Robinson, Dan. "Cloud Customers Are Wasting Money by Overprovisioning Resources." The Register, 7 Dec. 2022, www.theregister.com/2022/12/07/cloud_customers_overprovision
6. Strubell, Emma, et al. "Energy and Policy Considerations for Deep Learning in NLP." MIT Technology Review, 6 June 2019, https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/.
7. Groom, Nichola. "US Data-Center Power Use Could Nearly Triple by 2028, DOE-Backed Report Says." Reuters, 20 Dec. 2024, https://www.reuters.com/business/energy/us-data-center-power-use-could-nearly-triple-by-2028-doe-backed-report-says-2024-12-20/.

Derek Odom, CEO Non-Von
As artificial intelligence (AI) technologies continue to advance and proliferate, the energy demands associated with current computing hardware (CPUs and GPUs) is also rapidly increasing. This growing energy consumption driven by AI leads to significant environmental challenges.  New AI focused chip designs, like Non-Von’s, are a large part of the solution and are the focus of this post.

Growing Energy Demands
The rise of AI applications, especially generative AI, has led to a dramatic increase in energy consumption, particularly within data centers that support these technologies. According to the International Energy Agency, electricity demand from data centers is expected to more than double by 2026, largely due to AI applications (TIME). This surge in energy needs poses significant challenges, especially when much of this energy is derived from fossil fuels, undermining global climate change efforts (McKinsey & Company).

The overwhelming energy requirements of AI can strain local infrastructures, resulting in brownouts and additional resource shortages, such as water (used to cool the data centers), exacerbating the urgency of addressing AI's energy consumption sustainably (TIME). The increasing energy demands point toward a steadily increasing carbon footprint for AI computing industry and the processors currently powering its growth.  These environmental impacts, and their associated economic and national security implications, highlight the need for mitigating innovative solutions.(Yale E360).

Environmental Challenges
The environmental consequences of AI's energy demands are substantial. Critics emphasize that the growing carbon footprint associated with current AI processors could further exacerbate climate change (Goldman Sachs). While AI holds the potential to optimize energy usage and reduce emissions, the current trajectory of energy consumption may outpace available resources in many regions (Yale E360). The challenge lies in ensuring that the expansion of AI technologies does not compromise environmental sustainability.

The Need for Sustainable Solutions
To tackle these pressing challenges, collaboration between the energy and data center sectors is vital. Developing sustainable solutions, such as transitioning to renewable energy sources and enhancing energy efficiency, is critical (McKinsey & Company). 

Interestingly, AI can also contribute to its own solution. By leveraging AI to optimize energy use across various industries, companies and researchers can enhance overall energy efficiency, emphasizing the need for energy-efficient AI chips (Yale E360).

Advantages of Non-Von’s Hardware/Software Solution
Non-Von’s chips and supporting software ecosystem are engineered to overcome the limitations of traditional computing hardware. These chips boast enhanced computing power and improved energy efficiency, directly addressing the energy consumption challenges posed by standard chips. Non-Von uses advanced parallel processing capabilities, allowing them to perform complex AI functions while minimizing energy use.  While easily implementing today’s popular models through its advanced compiler, Non-Von’s hardware also opens up computing to a whole new class of sparse, unstructured models.  These types of models can enable additional efficiency gains currently unrealized due to constraints in today’s hardware. 

By prioritizing power efficiency, Non-Von chips are crucial in meeting the increasing energy demands associated with AI technologies.

Future Implications
As the demand for AI technologies continues to grow, the environmental impact of standard chips will become increasingly unsustainable. Transitioning to energy-efficient hardware/software solutions like Non-Von’s is essential for mitigating these impacts and achieving sustainability goals.  Continued innovation and investment in AI chip technology is a necessity in aligning technological advancements with environmental sustainability. By emphasizing energy-efficient designs and harnessing AI to optimize energy use, the industry can ensure that the benefits of AI do not come at the expense of our planet's health.

Hope for a sustainable future with AI
The environmental impact of standard chips presents significant challenges as the world moves toward an AI-driven future. However, with innovative solutions like Non-Von’s, there is hope for a more sustainable technological landscape. By addressing energy consumption and prioritizing efficiency, we can harness the power of AI while protecting our environment for future generations (not to mention the economic benefits of reduced electrical requirements). Sign up for our newsletter at the bottom of our homepage (non-von.com) and become part of shaping the future of AI with us.

Sources:

1. McKinsey & Company. "How data centers and the energy sector can sate AI’s hunger for power."
2. TIME. "How AI Is Fueling a Boom in Data Centers and Energy Demand." 
   
3. Goldman Sachs. "AI is poised to drive 160% increase in data center power demand." 
4. Yale E360. "As Use of A.I. Soars, So Does the Energy and Water It Requires."
   

Richard Granger, PhD, President Non-Von
Eli Bowen, PhD, CTO Non-Von

Non-Von chips reduce power costs by two orders of magnitude (100x) for the growing set of sparse, quantized AI systems.  Non-Von's power efficient chips both i) enable low-power untethered edge systems such as sensors, and ii) provide enormous power savings for server-level AI.
 
Current AI systems typically run on GPUs, first designed decades ago for screen graphics processing.  To work fast, GPUs rely on high power consumption, burdening current AI systems with immense electric budgets (1). Currently, many companies are scrambling to hire AI and data science coders, to construct or adapt large models, and then *lose money* on the resulting power bills.
 
Small parts of this burden can sometimes be alleviated with software changes alone.  Many companies are making systems sparse (reducing redundant data), and quantized or compacted (using lowest-precision encodings), such as BitNet and Hamiltonian Net (2).  
 
Orders of magnitude more savings arise by using non-GPU hardware that is directly fitted to these sparse and compacted software methods.  Some chips are designed for particular software examples, such as convolutional nets or transformers, but Non-Von's NV1 chip specifically targets the large and growing class of sparse compacted systems.
 
From first principles, Non-Von has designed a hardware-software solution called SCULPT - sparse compacted ultra low-power toolset for AI, "sculpting away" whatever the model does not need, refactoring it to take maximal advantage of Non-Von hardware.  Any existing model software can be SCULPTed with just a few lines of code, yielding large gains when implemented on Non-Von hardware.
 
By contrast, standard GPUs and von-Neumann architectures prevent efficient code execution, greatly adding to the costs of implementing and running large AI systems.  The Non-Von hardware-software SCULPT architecture allows AI designs that immediately run efficiently, with enormous cost savings.
 
Non-Von can take any trained model, including all common model development packages (Pytorch, ONNX, Tensorflow, Keras, SKLearn, and many more), and automatically convert it to a sparse compacted system implemented on silicon.  These can be printed as custom Non-Von silicon, or imprinted into Non-Von's mass-produced programmable AI accelerator boards.
 
The results are 100 times more power-efficient on tasks ranging from deep-learning-based visual recognition to transformer-based chatbots (3).  
 
 
Sources:

1. AI electrical power usage:
   1. https://www.nature.com/articles/d41586-024-00478-x
   2. https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand    https://www.wsj.com/business/energy-oil/ai-is-about-to-boost-power-billswholl-take-heat-for-that-c527f27b?mod=panda_wsj_custom_topic_alert
   3. https://www.theatlantic.com/technology/archive/2024/03/ai-water-climate-microsoft/677602/ 
2. BitNet and Hamiltonian HNet (Quantization and sparsity)   
   • https://huggingface.co/blog/1_58_llm_extreme_quantization
   • https://openreview.net/pdf/2dab642ad93b17ac60d51f6399c0f7f260b184e1.pdf
3. Hokenmeier et al. (2024) IEEE 33rd Microelectronics Design & Test Symposium (MDTS)
   1. doi.org/10.1109/MDTS61600.2024.10570137
   2. https://arxiv.org/abs/2409.19389

 

Derek Odom, CEO Non-Von
     The limits of current hardware prove particularly stark in the military and national security context. Not only are today’s tactical edge devices limited by the generally low computational capacity of current hardware, but any edge device on the battlefield also must limit its use of cloud communication, to evade enemy targeting of the device. Furthermore, the high power budgets of current edge AI hardware limits the duration of all types of autonomous devices from stand-alone sensors to air/land/sea/space unmanned vehicles.  These compute and power constraints often mean the ability to use advanced AI generally requires a cloud connection. This requires commanders to assess the risk of leveraging AI capabilities with the risk of compromise.  Together, these needs constitute an enormous demand for new approaches to AI at the edge.
 
     The recognition that today’s hardware technology must advance is not a novel idea.  Billions of dollars are pouring into developing advanced processors for AI applications from the edge to the data center.  But most of these "new" solutions are still just incremental fixes to existing technology.  The underlying architectural principal remains von Neumann’s 80-year-old design.  Right now, the industry reminds us of software companies that refuse to start over and instead build ever more convoluted solutions to advance their products. (Windows anyone?).
 
     We propose a different solution, a processor-and-software ecosystem intrinsically designed for AI.  We have constructed a new paradigm to break away from the constraining GPU world: a new chip architecture that enables efficiencies that the GPU ecosystem cannot take advantage of.  We start with the NV1 architecture, that already outperforms current edge GPUs on AI tasks [3], and encompasses a software ecosystem to enable algorithms to work far more effectively and efficiently than GPUs allow.  An innovative communication protocol allows Non-Von to minimize power usage, and data transport costs among nodes are vastly reduced by eliminating the address bus, through local target address matching [3].  Non-Von's new chips open the door for expanded classes of models, particularly suited to efficient edge processing.  Join our mailing list at the bottom of our home page (non-von.com), and help us build the future of AI. 
 
Sources:

1. “AI Is Poised to Drive 160% Increase in Data Center Power Demand.”
   
   1. https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand
      
   2. https://www.nature.com/articles/d41586-024-00478-x
      
   3. https://www.theatlantic.com/technology/archive/2024/03/ai-water-climate-microsoft/677602/
      
   4. https://www.wsj.com/business/energy-oil/ai-is-about-to-boost-power-billswholl-take-heat-for-that-c527f27b?mod=panda_wsj_custom_topic_alert
      
2. Singh and Gill, “Edge AI.”
   1. https://www.sciencedirect.com/science/article/pii/S2667345223000196
      
3. Hokenmeier et al. (2024) IEEE 33rd Microelectronics Design & Test Symposium (MDTS)
   1. doi.org/10.1109/MDTS61600.2024.10570137