Dojo decommission. Tesla’s AI hopes in tatters? Or triumph?

Welcome to AI Collision 💥,

In today’s collision between AI and our world:

• Tesla drops Dojo
• Best buds with Samsung
• Learning or inference?

If that’s enough to get the AI plans changing, read on…

AI Collision 💥

One minute Tesla was building the world’s most ambitious AI training supercomputer.

The next, it’s on the phone to Samsung, designing the chips that will actually run the AI in real products that “think”.

Dojo, Elon’s moonshot AI supercomputer and chip project to create a wafer-scale, Tesla-built training platform, is finished. The team has been broken up. Some of the top engineers have already left to start their own AI chip companies.

However as surprising and shocking as this might appear on the surface, this isn’t a retreat. It’s a shift in focus to the next major growth area in AI hardware… inference.

And that tells us something about the direction the whole AI industry is moving.

Training chips like Dojo’s were built to chew through billions of video frames from Tesla’s cars and robots. They were the muscle for building AI models, the kind of silicon that only makes sense in a climate-controlled data centre surrounded by racks of supporting equipment and infrastructure.

Inference chips are different.

They take the finished model and run it in the real world, they act more like a brain would with perception and “thought”. They make the decisions that keep a car in its lane or a robot steady on its feet. They use less power, work faster and need to be everywhere the AI is deployed.

Tesla has decided that building one chip that is very good at inference and good enough at training is better than maintaining two complete separate hardware lines. That means less wasted engineering and greater focus on the chips that are going to be relevant and important to their future products.

It also means, lower long-term costs and a single architecture that can scale from factory floor to public roads to the home within their humanoid robots.

To top all this off Tesla has signed a multi-billion-dollar manufacturing deal with Samsung to produce the new AI6 inference chips in Texas. AI5 is currently being done by TSMC and Samsung had already been working on the AI4, so they are familiar with Tesla’s operations.

These AI6 processors will sit inside Tesla’s next-generation self-driving systems, in Optimus humanoid robots and in Tesla’s own AI data centres. The same chip design will be deployed at scale, across cars, robotics and infrastructure.

It’s really an all-in-one play and Tesla’s making a huge bet that it’s the smart way to go.

The deal also anchors Samsung firmly in the high-performance inference market, a space where it has ambitions to challenge Taiwan’s TSMC and ensure it fends off other competition on the rise. For Samsung, this is both a manufacturing contract and a showcase for its capabilities in advanced AI silicon.

The AI headlines of the last few years have been about ever-larger models and record-breaking training runs.

The next phase is about deploying those models at scale and entering the age of inference.

Inference is where AI moves from a research project to a commercial product.

It is what makes autonomous driving possible, what powers robots in factories and what will bring AI assistants into everyday devices. It is the difference between AI being an impressive demo and AI being part of daily life.

There is also a practical reason for the shift. Training is costly, energy-intensive and centralised. Inference happens everywhere and can be monetised in every unit shipped. Whether it’s a car sold, a robot leased or a subscription service powered by AI, inference chips are revenue makers from day one.

I’d also suggest that you keep an eye on what the Dojo engineers and alumni move onto next. You can bet that while they might not be required at Tesla anymore, they’ll very quickly launch their own AI and robotics companies, raise millions, billions in capital and become hugely successful away from Tesla too.

Who knows, maybe even Mark Zuckerberg might pick up a few of them on the cheap…

But Tesla’s move tells the market that inference is the way forward. If your AI can respond faster, more reliably and in more places than a rival, you have an edge that compounds with every new deployment and generation of chip.

By focusing its engineering resources on AI6 and this partnership with Samsung, Tesla is betting that the next wave of AI value will come from making intelligence even more relevant to the human world.

Dojo’s shutdown is not the end of Tesla’s AI ambitions. It is a decision to prioritise Ai to the parts that matter most for mass-market adoption.

#AD

WATCH TODAY: NVIDIA’s Next Big Move

If you’re reading this, there’s still time to catch James Altucher’s landmark presentation. He reveals his breakthrough AI system that would have predicted on average an astounding 84% of the biggest buyouts over four years. Now this same technology is pinpointing a massive move from NVIDIA could take the market completely by surprise – and make smart investors a lot of money…. As soon as August 14^th

Trading options is high risk. Back tested results. Simulated past performance is not a reliable indicator of future results.

Boomers & Busters 💰

AI and AI-related stocks moving and shaking up the markets this week. (All performance data below over the rolling week).

Boom 📈

• Micron (NASDAQ:MU) up 14%
• Palantir (NASDAQ:PLTR) up 13%
• Apple (NASDAQ:AAPL) up 11%

Bust 📉

• Microsoft (NASDAQ:MSFT) down 2%
• IBM (NYSE:IBM) down 5%
• UiPath (NYSE:PATH) down 8%

From the hive mind 🧠

• AMC Networks is a big media and entertainment organisation. So when they start talking about how things are only scratching the surface with AI, you take note. The future of content will involve AI, it’s just a matter of whether it’s complimentary not a takeover.
• China has invented a breeding robot. Now before you get to eyes-open about that, it’s not the kind of breeding you first think of. It’s the other one. Although maybe they could do with the first one…
• The more you think about this, the wilder it gets. The idea of having to pay the US government just to sell into China is unbelievable. Yet… here we are.

Artificial Polltelligence 🗳️

Weirdest AI image of the day

ChatGPT’s random quote of the day

Thanks for reading, and don’t forget to leave comments and questions below,

Sam Volkering

Editor-in-Chief

AI Collision