Note: I am currently travelling. Hence the more limited postings.

Nvidia’s Upbeat Forecast Soothes Fears of AI Spending Bubble

Nvidia Corp. delivered a surprisingly strong revenue forecast and pushed back on the idea that the AI industry is in a bubble, easing concerns that had spread across the tech sector.

The world’s most valuable company expects sales of about $65 billion in the January quarter — roughly $3 billion more than analysts predicted. Nvidia also said that a half-trillion-dollar revenue bonanza due in coming quarters may be even bigger than anticipated.

The outlook signals that demand remains robust for Nvidia’s artificial intelligence accelerators, the pricey and powerful chips used to develop AI models. Nvidia had faced growing fears in recent weeks that the runaway spending on such equipment wasn’t sustainable. (…)

Nvidia’s CEO had said last month that the company has more than $500 billion of revenue coming over the next few quarters. Owners of large data centers will continue to spend on new gear because investments in AI have begun to pay off, he said.

Chief Financial Officer Colette Kress went further on Wednesday, indicating that Nvidia would likely eclipse the $500 billion target.

“There’s definitely an opportunity for us to have more on top of the $500 billion that we announced,” she said on the conference call. “The number will grow.”

The growing role of AI will help maintain demand for Nvidia’s products, Huang said. The technology is helping speed up existing computing work, such as search. And it’s about to come to the physical world in the form of robots and other devices.

Nvidia’s third-quarter results also topped analysts’ estimates. Revenue rose 62% to $57 billion in the period, which ended Oct. 26. Profit was $1.30 a share. Analysts had predicted sales of $55.2 billion and earnings of $1.26 a share.

Nvidia’s main data center unit had revenue of $51.2 billion in the quarter, compared with an average estimate of $49.3 billion. Chips used in gaming PCs — once the company’s chief source of revenue — delivered sales of $4.3 billion. That compares with an average estimate of $4.4 billion.

The forecast for the latest quarter reflects a staggering run for the company. Sales will be up more than 10-fold from where they were in the same period just three years ago. (…)

“Our forecast for China is zero,” Huang said in a Bloomberg Television interview. “We would love the opportunity to be able to reengage the Chinese market with excellent products.” (…)

On the conference call, Huang was questioned about the deals with OpenAI and Anthropic. Huang said Nvidia’s investment in OpenAI, which still hasn’t been finalized, will provide a good return, he said. Backing Anthropic, meanwhile, will help establish ties with a company that hasn’t been a big user of Nvidia’s technology, he said. (…)

Huang said Wednesday that the competitive pressure remains low. More customers are coming to Nvidia after trying out alternatives than ever before, he said. The complexity of AI computer systems has put Nvidia in a strong position, Huang said.

The CEO is also pushing to spread the use of AI across more of the worldwide economy. The CEO has embarked on a globe-trotting tour to persuade government bodies and corporations to deploy his technology. (…)

The Santa Clara, California-based company still has more than 90% of the market for AI accelerator chips. It’s added other products to that lineup to help solidify its edge, including networking, software and other services.

“Business is very strong,” Huang said in the interview. “We have done a good job planning for a very strong year.”

Nvidia Corp. Chief Executive Officer Jensen Huang said his company has enough new Blackwell chips to meet increasing demand and that business is “very, very strong.”

Speaking on Bloomberg Television, Huang offered a new insight into comments made earlier during his company’s third-quarter earnings report. His reference to the Blackwell product line being “sold out” meant that existing chips were being used at maximum capacity by customers, he said.

“We’ve planned our supply chain incredibly well,” Huang said. “We have a bunch of Blackwells to sell.” (…)

Huang also said Nvidia is getting an increasing portion of data center spending because its products are adding more capabilities. The forthcoming Vera Rubin generation will deliver about $35 billion of revenue for Nvidia out of the roughly $55 billion spent on each gigawatt of computing power, Huang said.

“Look Ma, no China last quarter!”

• Trump Plans ‘Genesis Mission’ to Boost US AI Development

President Donald Trump plans to roll out a “Genesis Mission” as part of an executive order to boost US artificial intelligence efforts on Monday at the White House, according to a Department of Energy official.

The effort is intended to signal that the Trump administration sees the coming AI race as important as the Manhattan Project or space race, Department of Energy Chief of Staff Carl Coe said Wednesday at the Opportunities in Energy Conference in Knoxville, Tennessee.

“We see the Genesis Mission as equivalent,” Coe said.

Coe declined to provide additional detail, but said the order would likely direct national labs to do more work on emerging AI technologies and could involve public-private partnerships.

The Trump administration is also preparing a separate executive order for the president’s signature that would allow the Department of Justice to sue states over artificial intelligence regulations it deems unconstitutional, and threaten funding cuts to states with AI laws considered too burdensome or restrictive. (…)

At a Saudi investment conference on Wednesday, Trump said he would work with partners “to build the largest, most powerful, most innovative AI ecosystem in the world.”

“And we are going to work it so that you’ll have a one approval process to not have to go through 50 states,” Trump said, adding that a patchwork of state-level regulations would be “a disaster” because business could be derailed by “one woke state.”

On Tuesday, Trump called on lawmakers to pass a federal standard governing artificial intelligence either in an upcoming defense spending bill or as standalone legislation.

“If we don’t, then China will easily catch us in the AI race,” Trump said in a social media post. (…)

Trump in July unveiled a sweeping AI policy blueprint designed to make it easier for AI companies to grow in the US, and easier for US allies to acquire crucial hardware and software.

That blueprint encouraged the Department of Energy and other agencies to invest in “automated cloud-enabled labs for a range of scientific fields, including engineering, materials science, chemistry, biology, and neuroscience” in collaboration with the private sector and national laboratories. It also directed the administration to expand AI research and training at the labs. (…)

That sounds smart and serious enough. Hopefully, there will be enough American scientists with sufficient research budgets …

• White House Asks Congress to Reject Bill Curbing Nvidia Exports

White House officials are urging members of Congress to reject a measure that would limit Nvidia Corp.’s ability to sell AI chips to China and other adversary nations, according to people familiar with the matter, dimming prospects for legislation opposed by the world’s most valuable company.

The so-called GAIN AI Act would create a system that requires chipmakers to give Americans first dibs on AI chips that are controlled for export to China and other arms-embargoed countries — an “America first” framing designed to appeal to the Trump administration. That would effectively bar Nvidia and Advanced Micro Devices Inc. from selling their best products to the Asian country, making GAIN AI something of a bipartisan congressional pushback to President Donald Trump’s suggestions that he is open to such shipments. (…)

Killing GAIN AI would not, however, mean the end of China chip curb efforts on Capitol Hill, where there is broad bipartisan support for limiting Beijing’s AI ambitions. Lawmakers have separately begun working on a measure that would codify existing limits on AI chip sales to the Asian country. That simpler legislation, which has not previously been reported, would require the Commerce Department, which oversees approvals of restricted technology shipments, to deny all applications for sales to China of any AI chips that are more powerful than what the US currently allows, effective for 30 months.

The fate of both bills remains undecided. Lawmakers are still considering whether to include GAIN AI in an annual defense bill that’s under discussion, while also determining when to introduce the second bill, which is called the Secure and Feasible Exports, or SAFE, Act of 2025. All told, the situation makes clear the significant appetite in Congress to play a bigger role in the wonky world of semiconductor export controls, a national security policy area that’s risen to the forefront of the tech and trade war between Washington and Beijing. (…)

Xi’s administration has discouraged Chinese companies from using even the AI chips that the US has permitted Nvidia to sell. (…)

• Google boss says trillion-dollar AI investment boom has ‘elements of irrationality’

Every company would be affected if the AI bubble were to burst, the head of Google’s parent firm Alphabet has told the BBC.

Speaking exclusively to BBC News, Sundar Pichai said while the growth of artificial intelligence (AI) investment had been an “extraordinary moment”, there was some “irrationality” in the current AI boom.

It comes amid fears in Silicon Valley and beyond of a bubble as the value of AI tech companies has soared in recent months and companies spend big on the burgeoning industry.

Asked whether Google would be immune to the impact of the AI bubble bursting, Mr Pichai said the tech giant could weather that potential storm, but also issued a warning.

“I think no company is going to be immune, including us,” he said.

In a wide-ranging exclusive interview at Google’s California headquarters, he also addressed energy needs, slowing down climate targets, UK investment, the accuracy of his AI models, and the effect of the AI revolution on jobs. (…)

Mr Pichai said the industry can “overshoot” in investment cycles like this.

“We can look back at the internet right now. There was clearly a lot of excess investment, but none of us would question whether the internet was profound,” he said.

“I expect AI to be the same. So I think it’s both rational and there are elements of irrationality through a moment like this.”

His comments follow a warning from Jamie Dimon, the boss of US bank JP Morgan, who told the BBC last month that investment in AI would pay off, but some of the money poured into the industry would “probably be lost”. (…)

Mr Pichai said action was needed, including in the UK, to develop new sources of energy and scale up energy infrastructure.

“You don’t want to constrain an economy based on energy, and I think that will have consequences,” he said. (…)

• US Faces Winter Blackout Risks From Data Centers’ Power Needs

Power consumption has grown 20 gigawatts from the previous winter, the North American Electric Reliability Corp. said Tuesday in its winter assessment. A gigawatt is the typical size of a nuclear power reactor. Supply hasn’t kept up.

As as result, a repeat of severe winter storms in North America that unleash a polar vortex, of which there have been several in recent years, could trigger energy shortfalls across the US from the Northwest to Texas to the Carolinas. All regions have adequate resources in normal conditions.

“Data centers are a main contributor to load growth in those areas where demand has risen substantially since last winter.” Mark Olson, manager of the reliability assessment, said in an emailed statement. (…)

From John Authers’s column:

(…) the Bureau of Labor Statistics announced Wednesday. September unemployment numbers will belatedly be available a few hours after this newsletter is published, but October’s will never appear. November’s will be published only after the Federal Open Market Committee meeting in December.

As the case for lowering rates rests on rising unemployment, and the Fed is data-dependent, this sharply reduces the chances of a cut.

A cut next month is unlikely but not impossible. Really poor data over the next couple of weeks might just sway enough votes on the committee. But recent comments from members make it look less likely. That point was amplified by Wednesday’s publication of the minutes for the last FOMC meeting, which confirmed that it had been very divided and revealed that “many” thought a December rate cut would not be appropriate. It made clear that the drift toward giving unemployment greater importance relative to inflation had halted.

China Weighs New Property Stimulus Package as Crisis Lingers

China is considering new measures to turn around its struggling property market, as concerns mount that a further weakening of the sector will threaten to destabilize its financial system, according to people familiar with the matter.

Policymakers including the housing ministry are considering a slew of options, such as providing new homebuyers mortgage subsidies for the first time nationwide, said the people, asking not to be identified discussing a private matter. Other measures being floated include raising income tax rebates for mortgage borrowers and lowering home transaction costs, one of the people said.

“The relaxation of fiscal policy is in line with our previous expectations, and reducing taxes and fees will moderately boost home buying activities,” said Jeff Zhang, a property equity analyst at Morningstar Inc. “We believe that the confidence of homebuyers still needs further stabilizing property prices to recover.” (…)

The dim outlook for the property market, coupled with households’ weakened ability to repay mortgages and other personal loans, means that banks’ asset quality could deteriorate next year, Fitch Ratings analysts warned last month. Chinese banks’ bad loans surged to a record 3.5 trillion yuan ($492 billion) at the end of September.

The plan to subsidize interest costs on new mortgages is intended to lure back homebuyers, who have been reluctant to enter a free-falling market.

While they may give a short-term boost, the steps are “probably not bold enough” to fix the supply-demand imbalance in the property market, Eric Zhu of Bloomberg Economics wrote in a note. “Cheaper mortgages may not help much if people don’t want to borrow.”

The average mortgage rate for buyers’ first homes in 42 big cities has hovered around 3.06% in recent months (…). 

Meanwhile, Chinese consumers remain firmly in deleveraging mode, hindered by soft income expectations and growing uncertainties in a slowing economy. Outstanding residential mortgages shrank in the second and third quarter to 37.4 trillion yuan and are now down 3.9% from a peak in early 2023.

Hundreds of billions of yuan of mortgages are likely in negative equity — a trend likely to intensify — which will weigh on buyers’ confidence and contribute to further home-sale declines. That risks deeper erosion of Chinese property developers’ inventory as well as the recovery value for bondholders.

They really need to get serious soon…

The USA also has its own housing challenges. The yellow line on this chart from Liz Ann Sonders shows the continued lack of interest in new housing, much like in 2006-10.

Shale Oil’s Next Revolution Should Worry OPEC

Even after years of technological breakthroughs, the shale industry still leaves most of the oil underground. At best, American drillers siphon away 15% to 10% of what’s potentially available; the rest has remained thousands of feet under the surface. Until now.

The next phase of the revolution — call it shale 4.0 — is an engineering arms race to improve the so-called recovery factor. Increasing the ratio even by a single percentage point is a prize worth billions of dollars over the lifetime of thousands of wells in Texas, New Mexico, North Dakota and Colorado. “The best place to find oil is where you already know you’ve got oil,” Chevron Corp. Chief Executive Officer Mike Wirth tells me in an interview in New York. “We know where the oil is. If we left 90% of the oil behind, it would be the first time in history that we didn’t figure out how to do it.”

If engineers are successful, it would turn shale from a sprinter into a marathon runner. The impact won’t be another gusher, but a steady flow of barrels far longer into the future than the industry anticipated. And the more the US provides, the less other sources — above all, the OPEC+ cartel — can pump without undermining prices. (…)

Lightweight proppants help, but until recently their high cost, at roughly $1 million per well, outweighed the profit of the extra oil. Exxon is experimenting with a new formula that the company says is cheaper, using particles of petroleum coke, a byproduct of its own refineries. The company claims that well recovery can improve by as much as 20%; the industry remains skeptical. Exxon is using its newly patented proppant in a quarter of all its wells in the Permian basin, and plans to expand it to roughly 50% by the end of next year. Others are playing with their own lightweight formulas, hoping to mirror the results.

Chevron, meantime, is trying a form of soap. The company already has a big business making petrochemicals such as lubricants. Thus, it’s tapping its in-house engineering talent to find cheap surfactants that can reduce friction inside the oil reservoirs. As with proppants, the problem in the past has been cost. But Chevron believes it’s developing formulas that work and are cheap. (…)

It’s not “if” but “when”. Shale 4.0 will happen.

Note: I am currently travelling. Hence the more limited postings.

Notices of Impending Layoffs by US Companies Surged in October

Some 39,006 Americans were given advance notice as required under the Worker Adjustment and Retraining Notification Act last month, the preliminary Cleveland Fed measure showed. In monthly data from 2006, that number has only ever been higher in 2008, 2009, 2020 and May 2025. (…)

China’s Rare-Earth Product Exports Falter as Talks Go On With US

China’s exports of rare-earth products edged lower in October from a month earlier, as Beijing and Washington continue to hash out details of supply arrangements under their trade truce.

Outbound shipments of the materials used in electric vehicles, weapons and high-tech manufacturing dropped to 6,173 tons, the lowest level since June, according to customs data released on Tuesday. This category is typically dominated by rare-earth magnets, the industrial components that played a pivotal role for China in facing down America’s trade offensive.

The US and China are still fleshing out details of a trade truce clinched by Presidents Xi Jinping and Donald Trump in Seoul in late October. The two sides have given their negotiators until end-November to agree on supply terms for US-bound rare earths, according to people familiar with the matter. (…)

• #China rare earths deal will ‘hopefully’ be done by Thanksgiving, Bessent says – Reuters

China’s Grip on American Medicine Cabinets Grows More Entrenched

Roughly one-in-four generic drugs taken by Americans rely on key ingredients from China, according to a report released Tuesday by the US-China Economic and Security Review Commission. The often low-cost staples account for 90% of the medicines used by Americans. Some of the ingredients — found in blood thinners, antibiotics and cancer treatments — are produced only in China.

With China’s recent restrictions on rare earth minerals top of mind, the commission said that similar moves involving drug ingredients “could have drastic consequences for the US healthcare system, causing supply shocks that would result in loss of lives and force hospitals to make tough choices in allocating insufficient supply.” (…)

Much of the government’s understanding of China’s reach is an estimate because the Food and Drug Administration doesn’t collect data on where the basic building blocks of medicine are made. The group is recommending Congress prepare legislation that would require companies to disclose that information to the FDA.

“We’re really, really far away from figuring this out,” Miller said.

Even as its control over generic drugs draws criticism, China is working to replicate that success in the production of more innovative treatments, according to the report. Economic incentives and a more lax regulatory landscape have made China an important development partner for brand-name pharmaceutical companies around the world, particularly for conducting “cheap, fast early-stage exploration,” the report said.

A survey last year by the Biotechnology Innovation Organization, an industry trade group, found 79% of 124 biopharmaceutical companies had China-based development and manufacturing partners. Most biotech companies don’t have the money needed to make drugs in the US, a key priority for President Donald Trump.

FDA Commissioner Martin Makary floated the idea last month of lowering the multimillion-dollar fees companies are charged for reviewing new medications if early-stage studies are done in the US, rather than in China. Makary told a gathering of pharmaceutical supply chain experts in Washington that the agency is eyeing upcoming user fee negotiations with the industry that occur every five years to negotiate the potentially lower prices. (…)

China isn’t done yet. According to the report, it is leading in so-called “synthetic biology,” or the artificial creation of biological organisms.

Dominance in that scientific field puts China in an indispensable position on a number of medical fronts, from making amino acids crucial to insulin and antibiotics to developing mRNA technologies and genetically engineered cells. Importantly, it also entrenches the country in every aspect of pharmaceutical production.

“The Chinese synthetic biology industry, for the foreseeable future, will have access to the innovations and know how of global competitors,” the report said. (…)

China isn’t the only country the US relies on for its drug supply. India also plays a large role, producing the bulk of the country’s generic drugs in finished form. While India makes many of the key pharmaceutical ingredients itself, a large share of the necessary materials come from China, according to the report. Also affected are brand-name drugs from Europe, where companies get more than half of their key ingredients from China, the report said.

In the end, fixing supply chain vulnerabilities will require a wholesale approach, take years and be difficult to pull off, the authors of the report said. It will require “significant modifications to US and global economic statecraft, tools, and approaches,” including efforts to bolster domestic manufacturing, they said.

While the Trump administration has secured commitments from some large drugmakers to open manufacturing plants in the US, they don’t include generic companies that can’t afford it. Recently imposed restrictions and reductions in research funding at US universities and other institutions also could limit America’s chances at extracting itself from China’s grasp.

AI CORNER

AliQianwen APP Beta Test, Competes with ChatGPT in Full Scope

On November 17, Alibaba officially announced the “Qwen” project, fully entering the AI to C market. On the same day, the public beta version of Qwen APP was launched. Based on the world’s top-performing open-source model Qwen3, it competes comprehensively with ChatGPT by being free and integrating with various life scenarios. The core management of Alibaba regards the “Qwen” project as the “battle for the future in the AI era.” (…)

The international version of Qwen APP for the global market will be launched soon, directly competing with ChatGPT for overseas users by leveraging the overseas influence of the Qwen model.

In February this year, Alibaba announced an investment of 38 billion yuan for AI infrastructure construction and set a long-term goal of expanding the energy consumption scale of cloud data centers tenfold by 2032. Since fully open-sourcing in 2023, Alibaba’s Qwen has surpassed models like Llama and Deepseek, becoming the most powerful and widely used open-source large model globally.

To date, the global download count of the Qwen series model has exceeded 600 million, accumulating significant industry reputation. Recently, Alibaba released the flagship model Qwen3-Max, whose performance surpasses international competitors such as GPT5 and Claude Opus4, placing it among the top three globally.

Qwen is rapidly capturing the Silicon Valley market. Airbnb CEO Brian Chesky publicly stated that the company is “heavily relying on Qwen,” as it is faster and better than openAI models. NVIDIA CEO Jensen Huang said that Qwen has captured a large share of the global open-source model market and continues to expand its share. Alibaba’s open-source model Qwen is becoming the foundation of Silicon Valley.

Alibaba’s management believes that the development of AI will go through three stages: “learning from people,” “assisting people,” and “exceeding people.” Currently, the capabilities of large models have entered the Agentic AI stage of “assisting people,” and the timing for Alibaba to heavily enter the consumer market is now mature.

Alibaba stated that the Qwen APP released this time is a preliminary version, which will use the most advanced model to create a “smart personal AI assistant that can chat and handle tasks.” In addition to being smart in chatting, “being able to handle tasks” will be a key focus area for the Qwen APP. The strategic goal of Qwen APP is to become the AI lifestyle entrance of the future.

Currently, Qwen has already shown some ability to handle tasks. For example, one instruction can allow the Qwen APP to complete a research report in a few seconds and turn it into a beautiful PowerPoint with dozens of pages. Not long ago, Qwen won the championship in a real-time investment competition against global top models such as ChatGPT, Gemini, and Grok.

It has been revealed that Alibaba is planning to integrate various life scenarios such as maps, food delivery, ticket booking, office work, learning, shopping, and health into the Qwen APP, enabling Qwen to have stronger task-handling capabilities.

Samsung hikes memory chip prices by up to 60% as shortage worsens, sources say

Samsung Electronics this month raised prices of certain memory chips – now in short supply due to the global race to build AI data centres – by as much as 60% compared to September, two people with knowledge of the hikes said.

Shares of Samsung, SK Hynix and U.S. chipmakers rallied sharply on the news which underlines how the boom in artificial intelligence has stoked intense demand for chip units specifically designed for AI tasks as well as the memory chips used in those units.

Soaring prices for these memory chips, which are mainly used in servers, are likely to add to stress for big companies building out data infrastructure. They also threaten to increase the costs of other products like smartphones and computers in which they are also used.

Many of the largest server makers and data center builders are “now accepting that they won’t get nearly enough product. The price premiums being paid are extreme,” Tobey Gonnerman, president of semiconductor distributor Fusion Worldwide, told Reuters.

The South Korean firm’s contract prices for 32 gigabyte(GB) DDR5 memory chip modules jumped to $239 in November, up from $149 in September, he said.

DDR memory chips are used in servers, computers and other devices, assisting with computing performance by temporarily storing data and managing rapid data transfer and retrievals.

Samsung also lifted prices of 16GB DDR5 and 128GB DDR5 chips by about 50% to $135 and $1,194 respectively. Prices of 64GB DDR5 and 96GB DDR5 have gone up by more than 30%, Gonnerman said.

Samsung declined to comment. It separately announced on Sunday that it will build a new chip production line at its plant in South Korea, as it expects AI will drive demand for the mid- and long-term.

The chip crunch has been so severe that it has spurred panic buying by some customers, according to industry executives and analysts.

China’s top contract chipmaker SMIC said on Friday that the memory chip shortage has meant that customers are holding back orders for other types of chips that are also used in their products.

Xiaomi, a Chinese smartphone, electronics and auto manufacturer, also warned last month that the surging prices have raised the cost of making phones.

The Debate About the Quality of AI Earnings

Ed Yardeni

Michael Burry, the man behind the “Big Short” during the Great Financial Crisis, is shorting the AI trade because he notes that hyperscalers have been depreciating their GPU chip investments over more than 3 years. He thinks that they should be doing it for under three years. (…)

Hyperscalers are stretching GPU depreciation schedules, a move that lowers expenses and boosts reported earnings. Critics argue that this is aggressive accounting since GPUs often become obsolete faster.

Many major hyperscalers publicly use an estimated useful life for their AI server equipment, including GPUs, of five to six years. This is an extension from their historical depreciation schedules for general-purpose servers, which were often around three years. Companies like Microsoft and Oracle have been cited as using or factoring in a useful life of up to six years for their new AI chips/servers. Cloud GPU rental company CoreWeave also extended its GPU depreciation period to six years, from four years, in 2023.

Amazon (AWS) uses shorter schedules closer to four years, while Meta has pushed to extreme lengths of 11–12 years. Microsoft, Google, and Oracle generally fall in the four- to five-year range.

Critics, including some prominent investors, argue that the true economic lifespan is much shorter, perhaps one to three years. Nvidia is now releasing new, significantly more powerful and energy-efficient AI chips (like the Blackwell and Rubin generations) on a one-year product cycle. This rapid innovation can make older chips economically obsolete for high-end AI training workloads much faster than a five- to six-year schedule suggests. High utilization rates (60%-70%) in demanding AI workloads also contribute to faster physical degradation.

Hyperscalers justify the longer depreciation schedule by arguing for a value cascade model. They contend that older generation GPUs, once replaced in top-tier training jobs, are simply cascaded down to power less computationally intense but high-volume inference (running the model) or other tasks, where they can still generate significant economic value for years. They also cite continuous software and data center operational improvements that extend the hardware’s life and efficiency.

If depreciation schedules don’t align with real-world replacement cycles, companies may be overstating their profitability and underestimating the capital-intensive nature of AI infrastructure. That would increase the chances that the AI boom is turning into an AI bubble that may be about to burst.

We side with the hyperscalers rather than Michael Burry in the depreciation debate. Data Centers existed before AI caught on in late 2022, when ChatGPT was first introduced. During 2021, there were as many as 4,000 of them in the US as a result of the rapidly increasing demand for cloud computing. Many are still operating with their original chips. The revenues and earnings of the hyperscalers continue to rise rapidly.

I agree with Ed. Performance needs will increasingly vary as AI moves from frontier AI training to less demanding applications such as inference, fine-tuning, edge deployments requiring moderate performance needs, and batch processing. Cloud providers are developing sophisticated scheduling and lifecycle management systems to dynamically allocate GPUs according to workload intensity.

And it’s not like we are about to have excess GPU production. Demand still far exceeds production capacity.