Andrew Dai, a key researcher who helped build Google's Gemini AI model, has launched a new startup called Elorian to tackle what he sees as a fundamental limitation in today's most advanced AI systems: they still struggle to truly understand visual information the way humans do. The Palo Alto-based company emerged from stealth mode on Thursday with $55 million in funding and a $300 million valuation, backed by investors including Striker Venture Partners, Menlo Ventures, Altimeter Capital, Nvidia, and prominent AI researcher Jeff Dean .

Dai's departure from DeepMind signals a broader trend of specialized AI startups emerging to solve problems that general-purpose models from tech giants haven't fully cracked. While companies like Google and OpenAI have invested billions into massive AI models, Dai argues these systems still fall short when it comes to reasoning about visual data in real-world applications .

Why Does Visual AI Reasoning Matter So Much?

The distinction Dai makes is crucial: generating images that look impressive is relatively easy for modern AI, but understanding what those images mean, analyzing them deeply, and reasoning about them is far harder. This gap has real consequences for industries that depend on visual analysis and design .

Consider the practical applications Elorian is targeting. In architecture, engineers need AI that can analyze building designs and suggest improvements. In automotive manufacturing, companies need systems that can reason about how to make cars lighter and more efficient. In robotics, AI must understand spatial relationships and physical constraints to manipulate objects safely. These aren't tasks you can solve with text-based reasoning alone .

"These are not things that you can just express with code and have a faster rocket. You actually need to design the physical thing, and that design lives in the physical world," said Andrew Dai.

Andrew Dai, Cofounder of Elorian and Former Google DeepMind Researcher

Dai also noted that while OpenAI recently shut down Sora, its video generation tool, Elorian's focus is different. The startup isn't primarily interested in creating media; it's focused on building reasoning capabilities that allow AI to understand and explain what it sees .

How Is Elorian Building Better Visual AI Models?

• Foundation Approach: Elorian is developing its early AI products on top of open source models that can be freely used and modified, giving the company flexibility and access to community innovations
• Specialized Design: Rather than relying on massive general-purpose models, Elorian is building models specifically designed to understand visual information and reason about it in complex ways
• Hybrid Release Strategy: The company is considering releasing smaller versions of its models to the open source community while keeping its flagship model proprietary to maintain competitive advantage

The company has already hired more than a dozen researchers and plans to release its first publicly available reasoning model within approximately the next 12 months . Though Elorian isn't yet generating revenue, it's already in discussions with potential customers across multiple industries.

What Makes This Startup Different From Other AI Labs?

Dai's background building Gemini at DeepMind gives Elorian a significant advantage over other frontier AI labs. According to Brian Zhan, a partner at Striker Venture Partners, Dai's experience means the company can operate more efficiently than competitors who are primarily focused on running expensive experiments .

"Andrew knows the Gemini recipe, he's not wasting a single dollar," said Brian Zhan.

Brian Zhan, Partner at Striker Venture Partners

Dai also took an unusual approach to the company's valuation strategy. The funding round was raised in two tranches, with the first capital coming in at a $120 million valuation and the second portion at a $300 million level. Dai deliberately chose a lower initial valuation to ensure early employees could see meaningful financial gains as the company grows, rather than starting with an inflated valuation that would limit upside potential for the team .

Elorian's cofounders bring complementary expertise. Yinfei Yang previously worked on AI research at both Google and Apple, while Seth Neel is a former Harvard professor who researched data and artificial intelligence . This combination of industry experience and academic rigor positions the team to tackle the complex challenge of visual reasoning.

Dai's move reflects a larger pattern in AI research, where specialized startups are attracting top talent and significant funding to solve specific problems that general-purpose models haven't fully addressed. Other examples include You.com founder Richard Socher raising capital at a $4 billion valuation for a more advanced AI system, and Periodic Labs, cofounded by former researchers at Google and OpenAI, currently fundraising at a $7 billion valuation .

The success of Elorian could reshape how the AI industry approaches visual understanding, moving away from the "bigger is better" philosophy that has dominated recent years toward more specialized, efficient models designed for specific reasoning tasks. For industries like robotics, automotive design, and architecture, that shift could unlock capabilities that today's general-purpose AI models simply cannot provide.