AI News: Amazon, Microsoft, Meta and Broadcom Are Making Big Moves
Today, we're doing something a little different: a quick rundown of what's happening in the world of AI. Because over the past few days, the tech giants have been making some interesting moves.
Amazon is discussing a $7 billion deal for a company founded only in 2023. Microsoft is starting to rely on its own chips. Meta is charging a subscription fee for its digital assistant. And Broadcom's CEO says the biggest obstacle to AI is no longer chips. It's money.
Four different stories. But put them side by side, and they tell the same story.
MARKETS AND SUBSCRIPTIONS
Let's start with Amazon. According to the Wall Street Journal, the company is discussing an acquisition of Decart AI for approximately $7 billion. The company was founded by three Israeli engineers, including two brothers.
"$7 billion for a company founded in 2023, seriously, Christos?" you might be wondering.
Yes. And Amazon wasn't the only company interested. Anthropic, the company behind Claude, also considered acquiring it for approximately $6 billion. After examining the company closely, Anthropic decided in September not to proceed.
Of course, the deal with Amazon hasn't been finalized either, and there's no guarantee it will go through.
Next up is Meta, which has introduced Muse Wonder, its new model for digital assistants. Its codename? Watermelon. Yes, watermelon.
But there's a catch: it's not free. To use it through the Muse app, you have to pay $20 or $100 per month. In other words, Meta is looking for revenue streams beyond advertising.
So, what's different about it?
According to Alexandr Wang, Meta's head of AI, the assistant doesn't get stuck when it encounters an obstacle, such as making a dinner reservation. Instead of stopping and asking you what to do, it tries another approach.
Of course, not everything is rosy. Some users have alleged that Muse read their private messages without permission, an accusation Meta disputes.
MICROSOFT IS MAKING ITS OWN CHIPS
This brings us to Microsoft. For years, the company has relied on Nvidia's graphics processors. With Maia 300, the third generation of its own AI chip, Microsoft will primarily rely on its own hardware to handle requests coming into its data centers.
"And why go to all that trouble?" you might be wondering.
One major reason is cost. In July, Satya Nadella said that the previous chip, Maia 200, delivers 30% better performance per dollar than the most advanced hardware Microsoft already had.
Microsoft CFO Amy Hood also said that the company's own chips had helped improve profit margins.
Maia 300 is scheduled to enter production in 2027. Compared with graphics processors, it delivers 40% more work for the same amount of electricity.
Evercore ISI estimates that by the first quarter of 2027, Maia will account for approximately 10% of Microsoft's AI chip spending.
Nevertheless, Nvidia isn't disappearing from the picture. Microsoft will continue using Nvidia and AMD chips to train its models.
There's also a catch. Unlike Google's and Amazon's chips, Maia chips aren't rented out to other companies. As analyst Patrick Moorhead put it, "Nobody has tested them outside Microsoft yet."
THE PROBLEM IS MONEY
Now things get even more interesting.
Hock Tan, Broadcom's CEO, said that shortages of silicon wafers and memory are being overcome. From this point forward, the biggest obstacle to AI isn't chips. It's financing.
To understand the scale of the numbers involved, Broadcom is in preliminary discussions to arrange more than $50 billion in debt financing for OpenAI, allowing it to purchase the chips the two companies are designing together.
At the same time, Broadcom is lending Anthropic up to $42 billion for infrastructure. Its Wall Street banking partners are also beginning to put together a new $60 billion financing package for chips, aimed at Anthropic and other companies.
Yes, you read that correctly. The supplier is helping its customer find the money to buy its products.
According to Reuters, Anthropic is expected to become Broadcom's largest chip-design customer in 2027.
"But will they ever make that money back, Christos?" you might be wondering.
Tan has an answer. He says OpenAI and Anthropic have demonstrated that their new models, including those powering digital assistants, can generate substantial revenue.
THE REAL BATTLE
So, what do all these developments tell us?
The AI battle is no longer just about who has the best model.
It's about who controls the software and who can make money from it. That's why Amazon is discussing a $7 billion acquisition of a startup. And that's why Meta is now charging subscription fees.
It's also about who controls the chips. That's why Microsoft is developing its own, while Broadcom is designing chips together with OpenAI.
But above all, it's about who has the enormous amounts of capital required to build infrastructure and data centers.
As Microsoft's Alistair Speirs puts it, energy translates almost directly into operating costs for a data center. And Microsoft is targeting facilities with power capacity measured in multiple gigawatts.
In short, having the best model isn't enough. You also need the chips to run it. And, most importantly, the money to pay for them.
