Ex-Google and Apple researchers raise $50M for Trajectory

Trajectory is a Palo Alto startup raising about $50 million to improve AI visual reasoning and feedback loops.

A new AI startup backed by former Google DeepMind and Apple researchers is betting that visual intelligence is the next bottleneck worth fixing. The company, Trajectory, is based in Palo Alto and is reportedly targeting roughly $50 million in seed funding.

That is a big check for a company that is still early, but the thesis is easy to understand: today’s models can write, summarize, and code with impressive speed, yet they still struggle with images, video, spatial relationships, and the physical world.

Why Trajectory is betting on visual feedback loops

Trajectory’s core idea is that AI should learn from visual data in faster, tighter loops. Instead of waiting for a giant retraining cycle, the company wants models that can keep improving as they see new examples from the real world.

This matters because visual understanding is still a weak spot for many systems. A chatbot can draft an email or summarize a paper, but it may miss that a box is blocking a hallway, a shelf is unstable, or a machine part is out of place.

The founders are borrowing a mindset from the software world, where rapid build-test-ship cycles helped software teams move faster. Applied to AI, the idea is simple: if the model is wrong, correct it quickly, feed the correction back in, and keep the loop moving.

• Text generation has improved fast, while visual reasoning still lags.
• Physical-world tasks need spatial awareness, timing, and context.
• Continuous feedback can help models adapt faster than old training pipelines.
• Robotics and industrial inspection are natural early markets.

Who is behind the startup

Trajectory’s founding team brings a mix of deep research and product experience. Andrew Dai spent more than 14 years at Google DeepMind and led data and pre-training work for the Gemini model family. Yinfei Yang, formerly Apple’s chief research scientist, adds heavyweight experience from one of the most secretive AI labs in the industry.

Seth Neel, who comes from Harvard’s AI research community, gives the company an academic angle. That mix matters because visual AI is not just a product problem; it is also a data, evaluation, and training problem.

“The future of AI is not just larger models, but better feedback loops,” Andrew Ng said at a Stanford HAI event in 2024.

That quote is useful here because it gets at the same basic bet Trajectory is making. The company is not trying to win by building another general chatbot. It is trying to win by improving how models learn from the world after deployment.

Why investors care about this angle

A $50 million seed round is large, but it is not absurd in 2026 AI fundraising. What makes Trajectory interesting is the narrowness of the problem it is attacking. Investors have seen plenty of generic assistants, coding copilots, and wrapper apps. A company focused on visual reasoning has a more specific technical target.

That target also maps to real businesses. Warehouses need robots that can avoid clutter. Factories need systems that can spot defects in real time. Self-driving systems need to read messy intersections and unpredictable motion. These are hard problems, but they are also problems with clear economic value.

• General-purpose chatbots are crowded.
• Visual AI remains underdeveloped relative to text AI.
• Robotics, manufacturing, and autonomous systems need better perception.
• Faster feedback loops could shorten the path from data to deployment.

For Trajectory, the pitch is less about hype and more about reducing a technical bottleneck. If the company can show that its approach improves visual reasoning in measurable ways, it could become relevant well beyond Silicon Valley.

What the reporting timeline tells us

The company’s fundraising activity first surfaced in The Information in January 2026, with Bloomberg following up in April. That kind of staggered reporting usually means a deal is real enough to attract multiple outlets, even if the final terms are still in motion.

There is also an important negative signal in the story: Trajectory has not indicated any involvement with blockchain or digital assets. That keeps it in the AI tooling and research lane, where the technical challenge is the product itself, not token design or on-chain incentives.

If Trajectory executes, the company could become a test case for a bigger idea: that the next wave of AI progress will come from better learning systems, not just bigger models. The real question is whether visual feedback loops can be made cheap and fast enough to matter outside demos.

My read is that this is the kind of startup worth watching closely over the next 12 to 18 months. If Trajectory can prove that continuous visual learning works in robotics or industrial settings, it may push other AI labs to rethink how they train models that interact with the physical world.

For readers tracking the space, the practical takeaway is simple: keep an eye on benchmarks, deployment partners, and whether the company can turn a research thesis into measurable gains in perception tasks.