How Positron AI is driving sales ahead of product | Mitesh Agrawal

Positron AI is a 2+ year old silicon company targeting decode-heavy AI inference workloads where memory bandwidth, not compute, is the bottleneck. Launching end of 2025/early 2026, their architecture delivers 2TB of on-chip memory capacity versus Nvidia Rubin's 0.4TB—enabling 3-5x better performance per dollar and per watt for reasoning models, code generation, and video generation. In this episode, ⁠Mitesh Agrawal⁠ shares how ⁠Positron⁠ identified the memory bandwidth gap in a market where Nvidia controls 90%+ share, why they're prioritizing anchor customer commitments over product completion, and the hard lessons from Lambda Labs about rapid iteration and customer-driven optionality.Topics Discussed:Positron's technical approach: focusing on memory bandwidth and capacity over compute for inference workloadsWhy decode-heavy applications (reasoning models, video generation, code generation) are becoming memory-boundThe challenge of selling silicon to hyperscalers when Nvidia controls 90%+ of the marketBuilding optionality into product strategy: air cooling vs. liquid cooling as unexpected GTM advantageLearning to sell hardware before the product ships and why anchor customers matterLambda Labs experience: lessons on rapid iteration and thoughtful hiring during hypergrowthMaintaining engineering-centricity: 47 of 50 employees focused on product developmentGTM Lessons For B2B Founders:Find technical bottlenecks in high-growth markets: Positron identified that memory bandwidth wasn't scaling as fast as compute, creating a bottleneck for inference workloads. While Nvidia dominates with 90%+ market share, they optimize for training revenue. B2B founders should analyze where dominant players are constrained by their own economics or existing roadmaps, then build specifically for those underserved segments.Markets default to oligopoly, not monopoly: Mitesh observed that customers actively seek alternatives even when one vendor is superior. "Markets want oligopoly structure to exist," he explained. B2B founders shouldn't be discouraged by dominant incumbents—customers want optionality for leverage, supply chain resilience, and risk management. Position yourself as the credible alternative in specific use cases.Discover optionality through customer conversations: Positron initially pitched performance per watt without realizing air cooling capability was a major advantage. Only after selling their first product did they learn customers valued deploying in existing data centers without infrastructure overhauls. B2B founders should systematically debrief early customers to uncover which features solve problems you didn't anticipate.Sell before shipping in hardware: The biggest priority between now and product launch is securing anchor customers willing to commit purchase orders. "If you have someone to build for, the fillip it gives the engineering team, the confidence it gives operations and supply chain vendors—we underwrite that," Mitesh emphasized. Pre-sales derisk production, prove demand, and create momentum. Build storytelling into technical sales: Convincing customers to buy unshipped hardware requires months of narrative work. "It becomes like, if I sell it to you, why will it be useful to you? Is it going to save cost? Attract new customers? Drive growth?" Success means co-creating the internal business case your champion will present. Maintain rapid iteration cadence: Nvidia ships every 12-15 months versus the industry standard of 3-4 years. "If you tell me that in 10 years you've launched 10-12 products in silicon, I will give much more probability we will be successful," Mitesh stated. Delay non-engineering hires until product proves itself: With 47 of 50 people in engineering, Positron has consciously prioritized product over go-to-market. "It was a very conscious decision," Mitesh emphasized. For deep-tech companies, this focus ensures you can actually deliver before scaling sales.