Why investors like Amazon and Google are betting $500m on this Kiwi

Friday, 3 July 2026

Kiwi Jeff Hawke’s AI startup, Odyssey, secured $500 million NZD in a funding round that values the company at $2.5 billion NZD.

Global tech giants like Amazon and Google are backing him, alongside local venture capital firm Icehouse Ventures.

Odyssey is building system known as a 'world model,' designed to help computers understand and simulate the physical world.

Despite headlines promising AI is about to usher in either a sci-fi utopia or a Terminator-style armageddon, most of us still experience it through the fairly mundane medium of chatbots.

Sure, they’re great for drafting boring emails or summarising dense reports. But where is my virtual assistant like Iron Man’s Jarvis? Where is the fully reactive virtual world like Star Trek’s Holodeck?

And when’s my bloody robotic butler doing my laundry?

Kiwi Jeff Hawke, co-founder of AI company Odyssey, recently raised over $500m NZD at a $2.5b valuation to turn those sci-fi dreams into a reality—minus any murder bots.

He is building a new kind of foundational AI system known as a world model, which he is betting could transform everything from the robotics industry to the gaming sector.

“There's a whole host of human knowledge that is not encoded in language. And really that's where we come in … this is taking AI out of the classroom and into the world,” Hawke told Stuff.

He has some heavyweight investors in his corner. Kiwi VCs Icehouse Ventures backed the round alongside Natural Capital, Amazon, AMD Ventures and GV (formerly Google Ventures).

“The easiest way to think about it is that OpenAI built models that help computers understand language,” explains Icehouse Ventures principal Mason Bleakely.

“Odyssey is building models that help computers understand the physical world. If they get it right, that becomes core infrastructure for robotics, autonomy, gaming and any industry that needs to simulate real environments.”

Hawke is no stranger to the frontier of AI; he cut his teeth as a founding engineer at autonomous driving company Wayve - currently worth $15b.

He says self-driving was the ‘proving ground’, but now that the technical challenge of self-driving is largely solved and robotaxis are rolling out worldwide, he wants to take the lessons learned at Wayve and start applying them to the wider economy.

A lot could be riding on whether he pulls this off.

The multi-trillion-dollar bet

It’s hard to overstate how much artificial intelligence is driving the global economy.

Investment bank Goldman Sachs estimates $7.6 trillion USD will be spent globally on AI over the next five years, just on hard costs like data centre infrastructure.

The World Economic Forum estimates AI could contribute up to 14% of global GDP by 2030, equivalent to about $15.7 trillion USD.

To put that in perspective, the United States spent around $300 billion USD ($514 billion NZD) in total over a decade, inflation-adjusted, on the Apollo programme to get humans to the moon.

At the time, that was seen as a historic spend. Tech giants are currently spending an Apollo programme's worth of money on AI every 6 to 10 months.

But all of that investment only works out—avoiding a monumental stock crash—if AI continues to improve at an exponential rate and becomes practically useful across virtually every sector of the economy.

If the current AI boom is a race, the hypothetical finish line is ‘AGI’, or artificial general intelligence. The definition is fuzzy but generally refers to a kind of AI capable of doing most valuable human work.

That’s why bets like Hawke’s, which could unlock a new frontier of progress, are attracting so much money and attention.

The kind of AI you’re probably familiar with in ChatGPT, Claude, or Gemini are all Large Language Models (LLMs).

Put simply, an LLM turns natural language into a maths problem and then back again. See more details in Stuff’s explainer.

There are two general camps in how we scale up existing AI towards AGI. The first is betting that current, language-based models can eventually learn to build better versions of themselves through a process called ‘recursive self-improvement’.

There is some evidence this is already happening, with Anthropic claiming recent versions of Claude wrote 80% of its own code.

Unfortunately, there’s also some concern that full recursive self-improvement could end the world, but that’s another story you can check out here.

Assuming the above method doesn’t kill us all, the second camp, where Hawke sits, says we need AI that can simulate interactive representations of the world with all its physical complexity.

Basically, if we want AI that can do valuable work in the real economy, we need AI which can experience the world more like a person does and less like a chatbot. Not everything can be broken down into language problems - or at least, not efficiently.

Hawke says an obvious early adoption for Odyssey will be in video gaming. Imagine if Grand Theft Auto 7 wasn't pre-generated, but a world that built itself around you and responded to your inputs in real time.

You can check out an early version of the virtual worlds Odyssey is creating on their website. As a demo, they built a version of the classic Nintendo game GoldenEye being generated in real time with multiple players interacting.

But the ultimate version, according to Hawke, would be a system which can responsively produce any virtual world you want and understand the physical world around you, like Iron Man’s Jarvis or the Star Trek Holodeck.

And the data produced by such an AI system would be crucial in training robots which can usefully navigate physical spaces, with obvious implications for essentially all physical labour.

The head-spinning pace of progress

While Hawke is planning to disrupt the AI world with a new foundational model, he is quick to stress that it is a complementary technology to LLMs.

Rapid progress in that field is transforming how he builds his own AI, with advances in coding tools like Claude Code providing a massive accelerant.

“One of my colleagues went on parental leave in February for a couple of months. They came back after just two-and-a-half months and said, ‘I don't understand my job anymore’. This is a person who has a very deep background in machine learning research and is a very, very competent developer,' Hawke says.

As for how soon we can expect some of the benefits outlined above, Hawke says world models are currently at a stage equivalent to GPT-2, which was built by OpenAI in 2019.

Three years later, the much more user-friendly ChatGPT launched and kickstarted the current AI revolution.

So if progress heads on a similar track to what happened with LLMs, we can expect our ‘ChatGPT’ moment for world models around 2029.

Guess we’re stuck doing our own laundry till then.