Google’s new chip in the works may run AI up to 10x cheaper — why this efficiency breakthrough makes the stock a screaming buy

Alphabet (NASDAQ:GOOG) shares are taking several backward steps after reporting some pretty strong quarterly earnings results. Despite the exceptionally strong results, though, the stock took quite a dive in the after-hours session, now off just shy of 3% as of the time of this writing, just a few hours following the big reveal.

It feels like Alphabet is finally on the road to making a profit on its extraordinary CapEx, but that wasn't quite enough, especially since many investors are still just a bit shocked over the pace of spend, with quarterly CapEx coming in just shy of $45 billion — that's a lot of money being spent in three short months.

In any case, the cloud is flying higher, and search has continued to prove resilient amid the continued ascent in AI. As a wave of AI agents comes online and constraints are dealt with, perhaps there's still ample upside for cloud growth. In the meantime, expect shares of Alphabet to be sent to the penalty box for no good reason.

As the valuation starts coming in again and investors look past the incredible innovations, especially on the hardware side, that Google is investing big money into, I do think the window to buy at a meaningful discount has opened.

The company isn't just scaling up; it's tackling some very hard problems behind the scenes to get over some of the hurdles (including energy and memory) that rivals in the AI race might stumble into.

Google's new chip sounds seriously impressive — it's a real driver that makes recent delays and departures forgivable

Take Google's "Frozen v2" custom AI chip, which is reported by The Information to be 6-10x more efficient. The chip, which is in the works, might just help Google win serious market share as inference hits an inflection point.

Of course, it's hard to know what to make of the hardware breakthroughs going on behind the scenes, especially following a series of discouraging developments, from big-name AI researchers choosing to leave just a few weeks ago to delays hitting the release of Gemini's latest Pro model.

Who would have thought that we'd get Gemini Flash 3.6 before Pro 3.5 landed?

I don't fault Google for taking its time, especially since it's looking to raise the bar in the AI race, rather than just keep up. In my view, a month or so difference in release dates isn't all too meaningful if it means Gemini will be in a spot to top rivals in a range of metrics later on.

At this point in the AI race, it feels like enterprise users are more than willing to make the switch to the very best that's available if it means getting the upper hand. In that regard, I'd argue it makes less sense to release something that isn't quite a disruptive force that beckons in customers.

Frozen v2 could set a new high bar for ASICs

While it'll be quite some time before Frozen v2 hits the ground (another two years or so), I do think that it might not take all too long before Gemini packs Pro-level smarts at the speed and cost of its Flash model, especially if Google's coming ASIC lives up to the hype as "efficiency-maxxing" inference becomes the name of the game.

Whether such unprecedented efficiencies help tear down memory bottlenecks remains the big question. Combined with algorithmic innovations such as TurboQuant, I'd be willing to bet that Google might rise as one of the biggest custom silicon winners.

At the end of the day, Google isn't just making breakthroughs on the hardware side, but the software side as well. The result may very well be AI compute efficiencies that might be tough to keep up with in this looming "inference explosion" era of the AI revolution.

Contact [email protected] for any questions or corrections.