A new wave of American startups is attempting to break Nvidia’s dominance in AI chip manufacturing by partnering with US-based fabrication facilities, moving away from the industry’s reliance on Taiwan Semiconductor Manufacturing Company (TSMC). Positron AI, founded by Thomas Sohmers in 2023, has designed a chip architecture specifically optimized for transformer models—the foundation of OpenAI’s GPT systems—focusing on AI inference computation.
Positron manufactures its chips in Chandler, Arizona through Intel-owned Altera, leveraging relationships established when early employees came from Altera in 2023. While Sohmers acknowledges his chips have “woefully less FLOPS” than Nvidia GPUs, he argues that Positron’s architecture compensates through efficiency and cost-effectiveness, delivering better performance per dollar and per watt.
The strategy of using smaller US fabs offers several advantages. These facilities are “hungrier” for business and offer more favorable terms to startups that lack the high-volume orders needed for leverage with TSMC. Surprisingly, US semiconductor production can actually be cheaper for AI applications compared to other industries. The approach has kept Positron’s funding requirements reasonable while enabling easier scaling.
Positron isn’t alone in this strategy—Groq, another Nvidia challenger, partners with GlobalFoundries in upstate New York, pursuing similar goals of competitive performance at lower prices. Both companies aim to capitalize on growing industry discomfort with Nvidia’s 90-plus percent market share.
However, the approach carries risks. Some investors remain skeptical, and engineers naturally question choosing less-proven fabs over TSMC’s world-leading consistency and yield rates. The geopolitical advantages are significant though—avoiding the “constantly simmering tension between Taiwan, China, and the US” provides strategic security. Positron is also working to source more components from North America or outside China and Taiwan, with Mexican sourcing offering both geopolitical safety and faster shipping for rapid prototyping.
The movement represents a broader industry awakening to the need for diversification in AI chip manufacturing, challenging the TSMC-Nvidia duopoly that has defined the sector.
Key Quotes
Fundamentally, Positron is trying to provide the best performance per dollar and performance per watt
Thomas Sohmers, founder of Positron AI, explains his company’s core value proposition in competing against Nvidia. Rather than matching raw computational power, Positron focuses on efficiency metrics that matter for practical AI deployment costs.
In most other industries, made in the USA actually means that it’s going to be more expensive. That’s not the case for semiconductors — at least for now
Sohmers highlights a counterintuitive advantage of US chip manufacturing for AI applications, challenging conventional wisdom about domestic production costs and providing economic justification for the reshoring strategy.
If I have some optionality going with someone that is behind but has the ambition to get ahead, it’s always good from a customer or partner perspective. It gives both leverage
Sohmers explains the strategic benefit of partnering with smaller, hungrier US fabs rather than industry leader TSMC, describing a mutually beneficial relationship that provides startups with negotiating power they wouldn’t otherwise have.
People are finally getting uncomfortable with Nvidia having 90-plus percent market share
Sohmers observes a growing industry recognition that Nvidia’s dominance poses risks, suggesting market conditions are becoming more favorable for alternative chip providers and indicating potential shifts in customer purchasing behavior.
Our Take
This story represents more than just startup competition—it’s a fundamental challenge to the AI infrastructure status quo. The Positron approach is particularly clever: rather than trying to out-engineer Nvidia on raw performance, they’re optimizing for the specific workloads (transformer inference) that dominate real-world AI applications while leveraging cost and supply chain advantages.
The geopolitical dimension cannot be overstated. Taiwan’s precarious position makes TSMC dependence a systemic risk for the entire AI industry. By proving US fabs can be competitive for AI chips, these startups are essentially creating insurance against supply chain disruption.
What’s most intriguing is the architectural specialization strategy. As AI models diversify beyond transformers and inference workloads vary, there’s growing room for specialized chips that sacrifice general-purpose computing for efficiency in specific tasks. If Positron succeeds, it could validate a new competitive model: domain-specific optimization plus domestic manufacturing as a viable alternative to the pursuit of maximum FLOPS. This could reshape how we think about AI chip competition entirely.
Why This Matters
This development signals a critical shift in the AI hardware landscape as the industry confronts the risks of over-concentration in both chip design (Nvidia) and manufacturing (TSMC). The geopolitical vulnerabilities surrounding Taiwan have become increasingly apparent, making domestic US chip production not just a matter of industrial policy but of national security and AI infrastructure resilience.
For businesses building AI applications, the emergence of competitive alternatives could reduce costs and increase supply chain reliability. If startups like Positron and Groq succeed in delivering comparable performance at lower prices, it could democratize access to AI computing power, enabling smaller companies to compete more effectively.
The trend also reflects broader reshoring efforts in semiconductor manufacturing, supported by initiatives like the CHIPS Act. Success here could establish a template for other critical technology sectors. Most significantly, breaking Nvidia’s near-monopoly could accelerate innovation by forcing competition on price, efficiency, and specialized architectures rather than raw computational power alone. The industry’s willingness to embrace these alternatives suggests growing recognition that the current concentration of power is unsustainable for the long-term health of AI development.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- Jensen Huang: TSMC Helped Fix Design Flaw with Nvidia’s Blackwell AI Chip
- Biden hails $20B investment by computer chip maker in Arizona plant
- Encharge AI Secures $21.7M Series A Funding to Revolutionize AI Chip Efficiency
- Pitch Deck: TensorWave raises $10M to build safer AI compute chips for Nvidia and AMD
Source: https://www.businessinsider.com/us-startups-american-fabs-ai-chip-production-2025-1