No matter. By that time, Hawke was back in school, first earning a master’s degree in robotics from the Georgia Institute of Technology then, after graduating in 2013, hopping the pond for a PhD in Applied AI at Oxford.
This was years before artificial hit the mainstream, but another Kiwi expat in the UK, fellow Auckland University engineering school mechatronics grad Alex Kendall, was already cooking up a way for AI to learn how to navigate traffic lights and roundabouts for his self-driving car technology start-up, Wayve.
Jeff Hawke, who was raised in Mt Albert, was last in Auckland for Easter. He and his family visit once a year.
The idea was that a car would “see” the world via camera, radar or LiDAR (Laser Imaging, Detection and Ranging) or other sensors, then the hardware-neutral Wayve AI would crunch the data so the car could drive itself, making decisions by itself as it navigated new streets rather than rote-learning routes.
That all went extremely well. Wayve recently raised US$1.2 billion in a Series D round which valued the company at US$8.6b, with Microsoft, AI chip-making king Nvidia and three major car makers (Mercedes-Benz, Nissan and Fiat, Peugeot and Jeep owner Stellantis) investing.
Large language models like ChatGPT use text scraped from websites to teach their AIs. Odyssey created a wearable rig with six cameras, two lidar sensors and an inertial measurement unit to collect data that would teach its platform about the real world.
Uber put in another US$300m on top, and announced that Wayve’s technology would be used for its first robotaxi trial in London later this year.
But Hawke departed mid-flow, in 2023, for a stealth start-up – now revealed as Odyssey, which he co-founded with Yorkshireman Oliver Cameron, a self-styled “university dropout” who learned to code as a child.
Hawke and Cameron (who had been working for a Wayve competitor) had the same idea: that the “world simulation” technology they were building to help cars drive by themselves could be decoupled from video and used to learn not just Give Way signs but the physics of the whole world and be used for everything from robotics to video games to interactive video.
“Instead of ChatGPT learning text and spitting out text, we developed a new category of foundational models called ‘world models’ – which you can think of as an intelligent stream of pixels that you can interact with,” Hawke says.
This is AI that can learn from sight and sound to understand how the real world works and then simulate it.
Instead of delivering a screed of text like the large language models (LLMs) that popularised AI three years ago, Odyssey’s model simulates the real world, allowing robots that learn like humans, and games that feel like living worlds, Hawke says.
Odyssey’s real-world simulation software.
See the kitten and baby demo on Odyssey’s website here.
The AI giants have already dabbled in video – most famously ChatGPT maker OpenAI with Sora, launched to the public (or at least ChatGPT Pro subscribers) in December 2024.
You could deliver a natural language prompt – “a stylish woman walking down a city street” – and Sora would produce a video that seemed out of Hollywood, at least if you didn’t look at the details too closely.
OpenAI’s short-lived Sora could create video from plain English prompts – but it wasn’t interactive and was restricted to a few seconds – yet still took 20 to 40 minutes to generate, turning it into a money pit for the ChatGPT maker.
In December 2025, Disney invested $1b into OpenAI to create a brave new world where Sora users could create their own videos using Luke Skywalker, Darth Vader or one of more than 200 other characters from the Disney-owned Star Wars franchise.
Hawke says Sora was delivering a finished product, whereas Odyssey can be used for creating a video – or a virtual world – that you can interact with, because it knows the laws of physics and how the world works.
You’ll also notice he’s talking about Sora in the past tense.
It all went to heck. Disney pulled the plug on its Sora partnership in March this year.
No product materialised for the public – likely because Sora took between 20 and 40 minutes to process a prompt for its maximum 10 seconds of video.
Sora was not only slow but a money pit. Cantor Fitzgerald estimated OpenAI was losing US$15m per day on the GPU-intensive video generation service, which would have worked out to US$5.5b over the year if the ChatGPT maker hadn’t pulled the plug in April.
“Sora was shut down largely due to cost reasons,” Hawke says.
Given Odyssey is not just video but interactive video, won’t it be quantumly more compute-intensive again?
Odyssey co-founders Jeff Hawke (left) and Oliver Cameron.
Especially given Hawke is not just talking about interacting with a piece of video, but multiple people interacting in the same scene at once – plus possibly some AI, agents too?
“One thing that is unique about world models is that they are streaming.
“You’re not generating a whole video clip at once, with a determined start and end. Rather, you’re simulating forward in time, which changes the equation a lot. You can structure it as a real-time model, which is much more computationally efficient.”
Some heavy hitters reckon that Hawke knows what he’s talking about.
When Odyssey staged a US$18m Series A raise in 2024, it drew investment from Nvidia and OpenAI.
A much larger Series B round that just closed brought in US$310 million ($529m) at a US$1.45 billion which saw Nvidia rival AMD plough in funds, along with Amazon, which has also become a partner.
Odyssey will use the tech giant’s Amazon Web Services (AWS) as its preferred cloud provider and Amazon’s inhouse-designed Tranium chips.
Other Series B backers included the CIA’s venture capital wing, In-Q-Tel and New Zealand’s Icehouse Ventures, which got its foot in the door with a $4.7m contribution.
The new funds will be used for “people and CPUs”, Hawke says.
How far is Odyssey from commercialising its product?
That will depend in part on how soon partners can take its technology and use it to create products that consumers can get their mitts on.
“I’d bet good money on seeing a world model video game within a year and I suspect a number of other industries will find early-use cases as well.
“AI moves fast.”
The expat crew
Odyssey has around half its 35 staff in Kings Cross, London, which Hawke says has emerged as the city’s AI hub.
He’s just minutes from Alex Kendall’s Wayve. The other half are in Palo Alto, Silicon Valley.
The UK and US are home to a raft of Kiwi expats making it big in AI.
Others making waves include Shane Legg, also based in Kings Cross, who is leading Google’s DeepMind project – possibly the closest thing that Odyssey has to a direct competitor (although in the context that Google also invested in Odyssey’s Series B).
“I have huge respect for them and a lot of DeepMind staff are friends,” Hawke says.
“It’s a small community.”
There’s also Dave Ferguson, the San Francisco-based co-founder of autonomous driving start-up, Nuro, which last August raised US$203m at a US$6b valuation from backers including Nvidia (Icehouse again got its foot in the door, chipping in $5m).
Plus the Christchurch-raised, now Singapore-based Paul Copplestone, co-founder of Supabase, which this month raised US$500m at a US$10b valuation for its technology that helps enable “vibe coding” (with Icehouse also chipping in).
Those in the cheaper seats include Adrian Macneil, the former Treasury software developer, now relocated to Silicon Valley, who recently raised $40m at an undisclosed valuation for his physical world AI start-up, Foxglove, Harry Mellsop, whose early stage firm. Antioch, which operates in a similar field and recently raised $15m at a $102m valuation, and Nic Lane, co-founder of UK AI-training start-up Flower Lab, which raised US$20m at a US$100m valuation.
Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.