Related Articles
This is a segment from The Breakdown newsletter. To read more editions, subscribe
“There has been a lot of talk about an AI bubble. From our vantage point, we see something very different.”
Jensen Huang
Alphabet CEO Sundar Pichai said this week there are “elements of irrationality” in the current boom in AI infrastructure.
His reveal of Google Gemini’s latest iteration hinted that perhaps we weren’t so crazy after all.
Gemini 3: Numbers Don’t Lie (But Do They Tell the Whole Story?)
On paper, Gemini 3 trounces its predecessor, Gemini 2.5. The language model gurus are ecstatic, citing metrics understood only by algorithms and themselves. As a mere mortal, I, however, remain unconvinced. The emperor’s new clothes shimmer, but the emperor himself? Still feels strangely…naked.
Forget hitting a wall! Tomasz Tunguz just detonated the “scaling wall thesis” – the notion that LLMs were grinding to a halt even with more processing power.
Google, however, added better compute smarter algorithms, better training, newer chips and Gemini 3 got a lot better.
This seems like a green light for everyone to keep investing like mad in everything.
This week, Jensen Huang dropped a bombshell: the GPU gold rush is far from over. “Blackwell sales are off the charts,” he declared, “and cloud GPUs are sold out.” The insatiable demand paints a picture of an industry utterly consumed by the power of accelerated computing.
Even after six years, Nvidia’s A100 GPUs are workhorses, according to CFO Colette Kress, who stated they “are still running at full utilization today,” proving their enduring value.
Kress’s comment landed like a counterpunch, seemingly aimed squarely at Michael Burry’s recent salvo. Burry had accused Nvidia’s clients of cooking the books, alleging they were juicing their reported profits through suspiciously extended depreciation timelines.
GPUs aren’t just cutting-edge for a season; they’re workhorses with a multi-stage career. After a year of intense AI training, these silicon gladiators transition to inference tasks, flexing their processing muscles on real-world data. Only then, after years of dedicated service, do they finally retire to the digital farm, perhaps spending their twilight years serving up YouTube videos – a well-earned rest for these digital titans.
For investors, this could be the golden ticket: AI models are still evolving, driving demand for cutting-edge chips while older generations remain valuable. The real question is, are AI companies subtly sandbagging their earnings reports?
And yet, stocks were lower this week.
The market’s mood has shifted: from fretting over chip demand to sweating the energy needed to fuel their creation.
Demand isn’t just high; it’s a ravenous beast. One Google Cloud leader confessed they’d need todoubletheir computing power every six months for the next half-decade just to keep the monster fed.
But where the power will come from to provide it is a mystery.
Data centers are insatiable beasts, and their hunger for power has created a bottleneck: the gas turbines that keep them humming. These energy behemoths take half a decade, sometimes longer, to construct. The companies that forge them? Their order books are overflowing, backlogged solid until the next decade. Forget 2029, we’re talking 2030, minimum. The digital age is stalling, not from a lack of innovation, but a shortage of raw power.
Chasing the latest GPU only makes sense if your power supply can handle the thirst. Otherwise, you’re better off sticking with an older, more efficient chip. A wisely chosen, power-sipping classic can often outperform a cutting-edge guzzler starved for watts.
Forget silicon valleys sprouting without wind farms; data centers demand power, and building them without new energy sources is like building a house with no foundation.
All of which is to say, the AI bubble could pop even if the demand for AI is effectively unlimited.
This week, even Pichai cautioned that a potential AI bubble burst wouldn’t spare anyone, Google included.
He meant no AI company, but “no company” is not far from the truth, either.
The US economy dodged a recession bullet, and you can thank…data centers? These digital warehouses, seemingly unassuming, are quietly fueling growth. While representing just 4% of the overall economy, they single-handedly powered a staggering 93% of GDP growth in the first half of the year. Without this unexpected surge from the data center boom, America might be singing a very different economic tune.
Forget your niche – from Main Street to Wall Street, your fate’s now entangled with the “scaling wall” theory, data center depreciation, and their ripple effects.
I’ll keep you posted.
“One chart to rule them all”:
Michael Burry’s Red Flag for the AI Frenzy: One Chart. Forget the hype. Burry sees a bubble brewing, and it’s all in one simple metric: capital expenditure as a percentage of GDP. He argues the AI investment surge already mirrors the unsustainable booms that fueled the dot-com crash, the housing crisis, and the shale oil bubble. Is AI innovation, or just history rhyming… again?
Depreciating, yes, but slowly:
Contra Burry from a16z notes that demand for older, less-powerful A100 GPUs has held up surprisingly well.
Going all in:
Microsoft’s betting big. Capex now devours almost half its sales, signaling a tectonic shift. Forget network effects; Benedict Evans argues Microsoft’s new game is raw capital. A game, he warns, dangerously close to bubble territory.
Startups are all in, too:
Again from Benedict Evans, nearly all Y Combinator startups are now AI-related.
It could get bigger:
Emre Akcakmak’s overlaid chart, while the launch dates might be up for debate, delivers a stark message: If AI is a bubble, history whispers that the ride could be far from over – buckle up.
The scariest chart in the world?
ChatGPT’s launch coincided with a peculiar economic see-saw: the S&P 500 levitating skyward while US job openings plummeted. Is this AI alchemy or mere coincidence? The chart alone is mesmerizing, hinting at a connection that’s either profound or profoundly misleading.
The bear case for human employment:
So far, so good:
The delayed jobs data showed the US with a surprise gain of 119,000 jobs in September.
We need some Autotronic housebuilders:
So, the median age of a U.S. home buyer is 59, huh? Finally, I’m ahead of the curve on something! But as a millennial who braved the housing market gauntlet, let me tell you: the American Dream is pushing fifty and slightly delusional. Renting gets a bad rap, but honestly, skipping the mortgage payments and leaky-roof anxieties? Underrated bliss.
At least houses aren’t imported:
A new economic report shines a spotlight on the hidden cost of tariffs for American consumers. The study reveals that import prices are now 5.44% higher than they would be without these trade barriers, impacting everything from electronics to everyday household goods. Are tariffs truly protecting American industries, or are they simply padding foreign manufacturer’s pockets at the expense of the average US consumer?
Inflation’s bite? A new car now tops $50,000 on average. Ironically, in some locales, you’ll shell out more for H2O than for a gallon of gas.
Have a great weekend, non-depreciating readers.
Thanks for reading Friday charts: Elements of a bubble
