Follow Ups: Intel AI PC, Qualcomm, Apple Intelligence, Microsoft Recall, Nuclear Power

How Microsoft's Stumbles Erased Qualcomm's Lead, More Nuclear Thoughts

We previously covered Intel's misguided AI PC strategy. Contrary to Pat Gelsinger's claims, the AI PC is not analogous to Centrino, and there are not compelling local AI use cases yet. Babbage added to the conversation with an interesting observation: Microsoft's Copilot+ hardware requirements likely reinforced Intel's overenthusiastic belief in the AI PC.

Microsoft requires 40 TOPS NPU → Intel Product team builds to the spec → Product and Marketing teams are excited about the NPU → AI PC hype train picks up steam.

Babbage’s Substack note

Here’s the corresponding die shot from Reddit. The NPU probably occupies 15-20% of the chip.

Lunar Lake’s huge NPU shows how important AI is to Intel

A friendly nitpick of Babbage’s poke at Apple:

Intel isn’t alone in overhyping AI features (🍎)

Yes, Apple announced Apple Intelligence before it was shippable, which was a questionable tactic and has similarities to Microsoft’s Recall rescindment. The difference is that Apple has compelling local AI functionality coming soon, but Intel does not.

The underlying value of Apple Intelligence is quickly apparent, especially on the iPhone. Using our thumbs as the primary input for a phone is an error-prone and slow user experience; instead, I’d rather use my voice as the input control as much as possible. Siri under delivered on that promise for the past decade+, but local Gen AI means we’re finally getting a useful Siri! Siri 2.0? 

Apple also showed that it can innovate with GenAI and build features we didn’t know we wanted, like Genmoji.

Chipstrat readers might not be the target audience for Genmoji, but their kids are!

Microsoft Recall, Take 2

Qualcomm made a big splash back in May as Microsoft's exclusive Copilot+ PC supplier, but the Copilot+ excitement quickly faded when Recall was rescinded over security concerns. Microsoft finally released Recall on October 1st, but that’s too little, too late for Qualcomm; Copilot+ is coming to Intel and AMD in November, just in time for holiday shopping.

This shines a light on the pains of not being vertically integrated. Qualcomm built a compliant NPU and OEMs raced to integrate the SoC in various laptops, but the software provider (Microsoft) stumbled and erased everyone’s head start. On the other hand, vertically integrated Apple designs its NPU and the software that runs on it, so they have complete control over the release timelines for AI on their laptops and phones.

Maybe this is why Qualcomm is investing in the Qualcomm AI Orchestrator, which allows them to control the software and hardware for smartphone generative AI.

By the way, I wonder if Qualcomm will share any meaningful PC sales numbers in their November quarterly earnings; the last earnings were “too early” to share an update.

Pro Nuclear

I’ve been thinking about nuclear a lot after last week’s post.

Data Center Energy, National Security, Nuclear Power

If training and inference scaling laws hold, all signs point to a future where energy demands exceed our current capabilities. Efficiency gains—whether chip, networking, or cooling—will only delay the inevitable. Since nuclear takes time and regulation, it’s probably best to start on it now.

Relatedly, Doug O’Laughlin is starting a fantastic primer series on nuclear here:

Nuclear Zeitgeist Part 1: A Reactor Primer

Speaking of Zeitgest, Jensen is pro-nuclear. From Bloomberg,

Nvidia Corp. Chief Executive Officer Jensen Huang, who helped create the technology at the heart of the explosion in artificial intelligence computing, said nuclear power is a good option for the renewable energy needed for the growing number of data centers.

“Nuclear is wonderful as one of the sources of energy, one of the sources of sustainable energy,” Huang said Friday in an interview with Bloomberg Television. “It won’t be the only one. We’re going to need energy from all sources and balance the availability and the cost of energy as well as the sustainability over time.”

And so is Eric Schmidt, who recently stated his belief that any gains from power infrastructure efficiency improvements are only temporary and will be quickly gobbled up by infinite demand.

The fact of the matter is the demand is infinite.

… We've already discussed better materials for batteries, that takes a while, but that's important. You move to DC power lines for better transmission loss… you work on power line losses, you put the data centers next to the big power lines as I described earlier. Architecturally, there huge improvements to be made in the cost effectiveness per transaction, but all of that will be swamped by the enormous needs of this new technology 

As far as other energy sources, I’m all for more renewable energy. Of course we’ll need battery innovation so we can stabilize this power. Regardless, I see no reason not to start building nuclear power immediately. It’s stable and clean, and there will be significant demand for it when it turns on. It sure seems like “no regrets” work.

Sure, the economics of nuclear may not be there yet, but surely the opportunity cost of the United States having the world’s foremost intelligence outweighs the economic cost of nuclear. Yes, maybe it will require government incentives, but intelligence is a national security concern.

I’m also bullish that access to said intelligence will help us solve this economic problem. By the time these nuclear power sources turn on, I’ll have kids old enough to seriously contribute to solving these problems, and they’re growing up as augmented intelligence natives. If we cast the right vision and build up this next generation, why couldn’t they solve this? After all, many of the best minds of some generations were distracted with building social networking and ads. But signs are pointing to the next generation rebuilding America’s defense technologies and tackling nuclear.

America can build the future we want; I’m glad to learn many have already started.