Jeff Bezos and Vik Bajaj Launch Prometheus to Bring AI to Hardware Engineering
Jeff Bezos and Vik Bajaj Launch Prometheus to Bring AI to Hardware Engineering
Jeff Bezos and Vik Bajaj have co-founded Prometheus, a new company building AI tools to help engineers design and manufacture physical products more quickly, according to CNBC. Both serve as co-CEOs, and the announcement came on 11 June 2026.
Prometheus is focused on the physical economy — the world of atoms rather than bits. The company's AI tools are designed for engineering workflows in hardware: mechanical design, manufacturing processes, and crucially, semiconductor chip design. That focus on chips is telling. Semiconductor design is one of the most complex and expensive types of engineering work that exists. Designing a new chip requires months of careful iteration, and the masks used to print them cost tens of millions of dollars. Any tool that meaningfully shortens that timeline could affect nearly every part of modern technology.
Bajaj previously co-founded Verily Life Sciences and was a managing partner at The Column Group, giving him deep experience in both science and finance. Bezos is known for building large-scale physical infrastructure — Amazon's AWS cloud business, its logistics networks, and Blue Origin's launch operations all represent long-term bets on complex systems. Prometheus is their shared bet on AI-powered tools for engineers as an independent venture.
In their CNBC interview with David Faber, Bezos and Bajaj framed Prometheus as infrastructure for the entire physical economy rather than a single product company. Chip design is their flagship example, but they see the same opportunity across hardware engineering broadly.
Why This Matters
AI has already transformed software engineering. Tools like GitHub Copilot help programmers write code faster; language models speed up documentation and testing. But hardware engineering has been slower to benefit from AI, for two practical reasons. First, feedback is slower: you cannot test a physical silicon chip design in milliseconds the way you can test code. Second, the data that trains these AI models — proprietary designs from chip makers and manufacturers — is kept secret and scattered across many companies, making it hard to create general-purpose tools.
Other companies have tried to solve this problem. Cadence and Synopsys, the major vendors of chip design software, have both pushed AI features into their tools. Startups backed by major investors have explored similar ground in materials science. But so far, no company has created a dominant platform that brings AI to hardware engineering the way GitHub Copilot or ChatGPT have for software.
The Open Questions
What Prometheus will actually build remains unclear from the information released so far. The company could add AI assistants on top of existing design software — the way Copilot layers on top of code editors. Or it could take a more fundamental approach, rebuilding how design tools represent physical objects from the ground up. That choice will shape both how competing companies respond and how quickly engineers adopt the new tools. Incumbents like Cadence and Synopsys have decades of customer relationships and proprietary software; beating them either requires a dramatically better workflow or a partnership strategy that does not trigger a defensive response.
The co-CEO arrangement is also worth watching as the company matures. Shared leadership between two founders works best when each person has truly different responsibilities — one handling strategy and external relationships, the other running science and product, for example. Salesforce's early co-CEO structure followed that pattern. If both Bezos and Bajaj are making the same decisions, tensions typically emerge within a year and a half.
Prometheus is still very early. No product has been shown publicly, no customer has been announced, and no separate funding round has been disclosed beyond what the founders themselves are putting in. What exists right now is a mission and two founders with the resources and reputation to attract top talent from the chip design, EDA (electronic design automation), and AI research worlds. In a field where progress depends on the quality of the team building the underlying models, that recruiting advantage is substantial.
The promise is straightforward: AI can do for hardware design what tools like Copilot have done for writing code. But the real work is in the difficult details — figuring out how to teach an AI system about physical constraints, manufacturing tolerances, cost, and supply chain realities in ways that let it suggest designs that actually work in the real world. The problem is solvable. It is also genuinely hard.
