Positron Raises $230M Series B to Challenge Nvidia with Energy-Efficient AI Chips

Positron, the AI inference hardware startup, has raised $230 million in Series B funding at a valuation exceeding $1 billion. The round was co-led by Arena, Jump Trading, and Unless, with strategic participation from Qatar Investment Authority (QIA), Arm Holdings, and Helena.

This funding marks a significant milestone in the race to challenge Nvidia's dominance in AI infrastructure—and Positron's approach is refreshingly different.

The Funding Details

Investors and Valuation

Positron Series B:
├── Amount: $230 million
├── Valuation: >$1 billion
├── Lead Investors: Arena, Jump Trading, Unless
├── Strategic Investors: QIA, Arm Holdings, Helena
└── Timeline: 34 months from founding to unicorn

Notable Investor Mix

The investor composition tells a story:

The involvement of Jump Trading—one of the world's most sophisticated trading firms—signals that Positron's technology performs in the most demanding, latency-sensitive environments.

What Makes Positron Different

The Problem They're Solving

Current AI inference infrastructure is expensive and power-hungry:

• GPUs are optimized for training, not inference
• Power consumption is becoming unsustainable
• Costs are exploding as AI usage scales
• Supply constraints limit access to top hardware

As Positron's founders put it: "GPUs for inference are CRAP—Chips Reaping Absurd Profits."

Positron's Approach

Instead of general-purpose GPUs, Positron builds purpose-built inference accelerators:

1. Transformer-first design - Optimized specifically for transformer models
2. Memory-centric architecture - Large memory capacity per accelerator
3. Power efficiency - Dramatically lower energy consumption
4. Software simplicity - Direct HuggingFace model loading, no recompilation

Performance Claims

Positron's Atlas system delivers:

For a trading firm running AI inference, 3x lower latency isn't a nice-to-have—it's potentially worth billions in competitive advantage.

Product Portfolio

Atlas (Shipping Now)

Positron's current production product:

Specifications:

• 8x Positron Archer Transformer Accelerators
• 32 GB HBM per accelerator (256 GB total)
• Dual AMD EPYC Genoa 9374F processors
• 64 cores total, 3.85 GHz base / 4.3 GHz boost
• Up to 2TB system memory support
• 24-hour SLA response from US-based team

Key Features:

• Runs models up to 500B parameters
• Direct HuggingFace .pt/.safetensors loading
• OpenAI API-compatible endpoint
• No custom compiler required

Titan (Coming 2027)

The next-generation system:

• "Superintelligence-in-a-Box"
• 8TB+ memory per system
• Powered by 4x Asimov chips
• Designed for massive concurrent model hosting
• Near-limitless context capabilities

Asimov (Coming 2027)

Positron's custom silicon:

• Purpose-built AI inference accelerator
• 2TB+ memory per chip
• Designed from Atlas learnings
• US-manufactured

The Nvidia Challenge

Why Now?

Several factors create an opening for alternatives:

1. Supply constraints - Nvidia GPUs are still hard to get
2. Power crisis - Data centers hitting power limits
3. Cost pressure - AI budgets under scrutiny
4. Inference explosion - Training is a one-time cost; inference scales forever

Positron's Advantages

For Inference Specifically:

• GPUs carry training-focused design overhead
• Purpose-built hardware can be more efficient
• Memory capacity matters more for inference
• Latency optimization is different from throughput

Business Model:

• Lower total cost of ownership
• Reduced power infrastructure needs
• Made in America (supply chain security)
• No liquid cooling required

The Competitive Landscape

Positron isn't alone in challenging Nvidia:

What sets Positron apart: shipping production hardware now, with proven customer deployments.

Customer Traction

Who's Using Atlas

Positron has deployed Atlas to customers across:

• Major cloud providers - Production rack deployments
• Trading firms - Ultra-low-latency inference
• Networking companies - Edge inference
• Gaming companies - Real-time AI features
• Content moderation - High-volume classification
• CDN providers - Distributed inference
• Token-as-a-Service - API inference providers

Proven Results

The most compelling evidence: trading firms are using it in production.

When Jump Trading—a firm where microseconds matter—backs a hardware company and uses their products, it's a strong signal that the performance claims hold up.

Company Progress

Rapid Execution

Positron's timeline is remarkable:

From founding to unicorn in under 3 years, while shipping real products.

The Team

Leadership combines AI, silicon, and cloud expertise:

• Thomas Sohmers (CTO/Founder) - Hardware architecture visionary
• Mitesh Agrawal (CEO) - Scaled GPU neocloud to $500M+ ARR
• Edward Kmett (Science) - Applied mathematics and optimization

The team has "over 400 years of AI, systems, silicon, and cloud experience combined."

Use of Funds

Scaling the Roadmap

The $230M will fund:

1. Asimov silicon development - Custom chip design and manufacturing
2. Titan system production - Next-generation product launch
3. Manufacturing scale - Increased Atlas production
4. Team expansion - Engineering and go-to-market
5. Customer support - 24/7 enterprise support infrastructure

US Manufacturing

Positron emphasizes American manufacturing:

"Designed, fabricated, and assembled in the United States."

This matters for:

• Supply chain security
• Government and regulated customers
• Reduced geopolitical risk
• CHIPS Act alignment

How to Get Started

For Enterprises

If you're evaluating AI inference infrastructure:

1. Contact Positron at positron.ai/contact-sales
2. Assess your workload - What models, what scale, what latency?
3. Request benchmarks - Get performance data for your use case
4. Plan deployment - On-premises or cloud partner

For Developers

The developer experience is straightforward:

# Using Positron with OpenAI-compatible API
from openai import OpenAI

client = OpenAI(uri="api.positron.ai")

response = client.chat.completions.create(
    model="my_model"  # Your uploaded model
)

Model loading:

1. Upload .pt or .safetensors file to Positron Model Manager
2. Update API endpoint to Positron
3. Run inference—no recompilation needed

The Bigger Picture

AI Infrastructure Shift

Positron represents a broader trend: purpose-built AI infrastructure.

The general-purpose GPU era served AI well during the training phase. But as AI shifts to inference at scale, specialized hardware makes economic sense:

• Training = High compute, one-time per model
• Inference = Lower compute per request, billions of requests
• Optimal hardware = Different for each

Energy Crisis Catalyst

The AI industry's power consumption is becoming untenable:

• Major tech companies restarting nuclear plants
• Data center power requests exceeding grid capacity
• Sustainability commitments conflicting with AI growth

4x better performance per watt isn't just a cost advantage—it may become a necessity.

Supply Chain Diversification

Reliance on a single GPU vendor creates risk:

• Nvidia's dominance gives them pricing power
• Supply constraints limit AI deployment
• Geopolitical factors add uncertainty

Alternatives like Positron create a healthier market.

The Bottom Line

Positron's $230M Series B represents a significant bet on specialized AI inference hardware. With:

• Production products shipping to real customers
• Trading firm validation (the most demanding use case)
• Strategic backing from Arm and QIA
• Aggressive roadmap toward custom silicon

They're positioned as a credible Nvidia alternative for inference workloads.

Key Takeaways:

• $230M Series B at >$1B valuation
• Backed by Arm Holdings, Qatar Investment Authority, Jump Trading
• Atlas delivers 4x performance/watt vs GPUs
• 3x lower latency in production trading workloads
• Custom Asimov silicon coming in 2027
• Fully US-designed and manufactured

The AI chip market is no longer a monopoly. Positron just made that clear.

Is your organization evaluating AI inference alternatives? Share your experience in the comments.