Qualcomm Buys Some Plumbing

Interconnects

A thing that happens in any complicated business is that the bottleneck moves. Let's say you invent a pizza oven that can cook a perfect pizza in ten seconds. This is revolutionary. You are the pizza oven king. You sell your oven to every pizzeria in the world. But then your customers notice a problem: while they can cook a thousand pizzas an hour, they can only deliver ten, because all their delivery drivers are on unicycles. The world's fastest pizza oven is constrained by the world's slowest delivery fleet. The bottleneck is no longer the oven; it's the unicycles.

So, if you are the pizza oven king, your next move is obvious. You start a new division that makes delivery helicopters. You sell your customers a complete, integrated system of 10-second ovens and hypersonic delivery drones. You don't just sell the best oven; you sell the best pizza-making-and-delivery-system. Anyone who wants to compete with you now can't just build a better oven. They have to build better helicopters, too.

This is, more or less, what is happening in the market for AI chips. For years, the race was all about the oven: building the fastest possible GPU. Nvidia won that race spectacularly. But now that AI models are too big to run on a single chip, the bottleneck has moved. The problem isn't just the speed of the processor; it's the speed of the connections between the processors. The industry is full of Ferrari engines running on roads designed for unicycles.

And so, to compete, you need to buy a helicopter company. Here is the news on Qualcomm:

Qualcomm Inc. has agreed to buy the London-listed semiconductor firm Alphawave for $2.4 billion. The move is a clear signal of Qualcomm's ambition to push beyond mobile and into the booming market for artificial intelligence data centers. But the deal isn't about building a faster processor to rival Nvidia's GPUs; it's about acquiring a different, but equally critical, piece of the puzzle: high-speed connectivity. The acquisition highlights a new front in the AI hardware race, where the ability to link thousands of chips together is becoming as important as the power of the chips themselves.

This is the game now. Nvidia figured this out ages ago when it bought its own helicopter company, Mellanox, in 2019, giving it the proprietary NVLink technology that is a core part of its dominance. It doesn't just sell you a chip; it sells you a whole rack of chips that are already wired together to be as fast as possible.

For a company like Qualcomm, which wants to break into the lucrative data center market, showing up with just a fast chip is no longer enough. That’s like trying to sell a pizzeria a really great oven without a solution for the unicycle problem. To be a serious player, you need a story about the whole system. Buying Alphawave is Qualcomm's attempt to buy its way into the helicopter business. It’s a shortcut to acquiring the critical interconnect technology required to compete not just on processor performance, but on system-level performance, which is where the real bottlenecks—and the real money—are now.

Is Apple Too Good at Making iPhones?

A thing that is famously hard in business is for a company that is very, very good at making a thing to then make a new, different thing, especially if the new thing threatens the old thing.

Imagine you are the world's best maker of internal combustion engines. Your engine is a masterpiece of precision engineering, beloved by millions, and it makes you an astronomical amount of money. Then someone in a garage invents a personal teleporter. Your first instinct is not to build a teleporter. It is to try to bolt a teleporter onto your existing engine block. Your best engineers, the ones who are experts in fuel injection and piston timing, will tell you how they can make the engine-teleporter work. It will be complicated, and buggy, and you'll have to make sure it doesn't violate your core principle of not exploding. Meanwhile, the people in the garage, unburdened by a car, just build a teleporter.

This is, more or less, the problem facing Apple Inc. and its artificial intelligence efforts, particularly with its voice assistant, Siri. Here is the Financial Times on Apple's struggle:

Recently departed employees told the Financial Times that the Silicon Valley giant has been hit by challenges with updating Siri using cutting-edge large language models that can deliver more sophisticated responses to spoken prompts.

...Former executives said that the process of integrating the technologies has led to bugs, an issue not faced by competitors such as OpenAI which have built generative AI-based voice assistants from scratch.

One former Apple executive said: “It was obvious that you were not going to revamp Siri by doing what executives called ‘climbing the hill’,” meaning to incrementally develop the product rather than rebuilding it from the ground up.

When you are the world's most valuable company, and you have to outsource a core future technology to a startup that you are also competing with, it suggests that things are not going perfectly. Apple has done just that, leaning on OpenAI to fill in the gaps for the queries Siri cannot handle.

This is not generally what you do when you are winning.

The situation is further complicated by Apple’s own strengths. Its famous focus on user privacy and on-device processing—core to the iPhone's appeal—adds another layer of complexity that cloud-native competitors simply don't have. Your teleporter has to work without knowing where anyone is, which is tricky.

And in a final, beautiful irony, the legendary designer who crafted the look and feel of Apple's world-dominating hardware, Jony Ive, is now working with OpenAI to build a new generation of AI-powered devices. The person who designed your perfect car is now working with the teleporter people. Which, again, is not ideal. The reward for building the world’s most successful product is that you are now stuck with it.

The Scoreboard

• AI: Why Meta Invests Billions in Labeling (ARPU)
• SaaS: DocuSign Stock Tanks 18% After Company Cuts Billings Outlook (CNBC)
• Semiconductor: Taiwan Exports Hit Record on AI Demand (Reuters)
• Robotics: Tesla's Head of Humanoid Robot Program Leaving the Company (Reuters)
• Media: WPP Chief Steps Down as Advertising Group Struggles With Rise of AI (FT)