We’re past the heaviest two weeks of earnings, but we have tons more reports to come. The cloud giants, Amazon, Microsoft, and Alphabet, have signaled continued spending even as their backlog swells, providing support for the AI trade. Software companies have turned in a very mixed performance, with cyber-security being the one bright spot. This week features the traditionally most expensive stocks in tech such as Palantir and Astera Labs, but also the tools makers that are riding the AI wave such as Datadog and Digital Ocean. And, special mention of AXT, whose wafer sales are a remarkable new chapter in its forty-year history.

Thursday evening’s earnings have prompted a jump in late trading of 11% in shares of Amazon and a drop of 8% for Apple, which is not surprising as I see Amazon one of the best positioned Mega Caps for AI along with Alphabet’s Google, while Apple has a mess on its hands with the rising cost of memory chips affecting iPhone and Mac costs and pricing, an issue that I expect to linger.

The evening also contained some interesting surprises that tend to reinforce the AI trade.

The biggest gainer after-hours is forty-year-old semiconductor company AXT of Fremont, California, up 37% in early Friday trading at $65.25

AXT makes the wafers that are used to make chips.

Tonight’s revenue upside of 39%, $48 million against a Street forecast for $34 million, is the biggest upside in at least five years, according to FactSet

The New York Times’s Rob Copeland on Thursday reported that hedge fund Situational Awareness has suffered a massive collapse in its AI-linked investments and had to be rescued by hedge fund Citadel, which swooped in to buy $10 billion of Situational’s stock holdings at “a substantial discount,” citing multiple unnamed parties.

The move was apparently prompted by gigantic declines of stocks held by Situational. The fund had enormous leverage, Copeland relates, so that the stock drops were amplified, prompting margin calls the firm couldn’t meet.

I never actually looked into Aschenbrenner’s investments. These are not, in my mind, the most promising firms to own in AI. They are mostly in the commodity business of renting GPUs to all comers.

On Wednesday evening’s conference call with Microsoft CEO Satya Nadella, we learned that Nadella has been reading one of the noteworthy business books of the year, “1873,” by Liaquat Ahamed. The book is about the speculative railroad building of the 1800s, and it’s very timely as it talks about excesses, credit bubbles, and what happens when it all comes crashing down — the very issues that hang over the artificial intelligence boom in which Microsoft is participating along with other tech giants.

Nadella brought up the book in response to a question by Bernstein analyst Mark Moerdler, who wanted to know “How does Microsoft protect itself if there really is overcapacity and overbuilding of data centers or overbuilding of chips?”

Said Nadella, “All of us are reading this 1873 as the book to be read,” said Nadella. What he’s learned, however, from the book, seems to miss Ahamed’s point, and didn’t really answer Moerdler’s question. “You’ve got to run an efficient railroad,” said Nadella, which is not really the point of the book. The point is the entire world inefficiently allocated capital for a period of 25 years. “I think you've got to get the product shape right,” added Nadella, which is what a product manager says when they have no answer for a really big question.

Tuesday evening was another very mixed bag for the AI trade, with shares of chip equipment maker KLA down 6% in late trading, and data center power supplier Vertiv down 12% in early Wednesday trading. But two other TL20 stocks to consider did very nicely, Seagate and Bloom Energy, their shares surging after-hours.

If I had to sum it up, I would say that the very focused product portfolios of Bloom and Seagate — a fuel cell, a hard drive — make for very clean reports, while the diversity of the businesses at KLA and Vertiv mean there are more “moving parts” every quarter, if you will, which makes it harder for the results to always satisfy the Street.

Bloom, maker of fuel cells that convert hydrogen and natural gas into energy to power data centers, continues to be a runaway train of sorts. The company’s reported revenue for the quarter came in 29% above consensus, at $1.07 billion,  the seventh quarter in a row of higher-than-expected sales.

• We continue to be in the unwind of the artificial intelligence trade. There’s no fundamental reason other than the stocks had “gone parabolic” from April through June, and it was time for them to come back to earth.

• The TL20 group of stocks to consider is a good barometer of that AI trade with names such as Vertiv and Micron Technology.

• The most interesting question coming out of last week’s reports is, Are investors rotating in a meaningful way from chips into software stocks.

The first batch of earnings are in for the week, and the result is punishment for stocks that had risen sharply this year and have failed to meet high expectations, including glass and fiber-optics giant Corning, down 16% at $116.32, and chip maker Amkor, down 24% at $46.35.

What you’re seeing with today’s losers is that revenue expectations had gotten ahead of reality based on exuberance about the current business environment, and now that’s being corrected. Business is still fundamentally healthy for these companies, but not always as big as some would like.

You’re also seeing the continued unwind of the AI trade given substantial investor anxiety about how much is sustainable in massive data center build-outs.

I have been skeptical of artificial intelligence being successful in the enterprise because of the enormous complexity of using AI to engineer something that is of acceptable quality. On Monday, analyst Rishi Jaluria of RBC Capital provided some excellent data that reinforces my skepticism.

It’s a delightful piece of work, one of the most enjoyable I’ve come across in a while. I don’t know if it will change investors’ minds, but it’s worth pondering.

Jaluria offers a “model,” or, as he puts it, a “formal framework, which investors can scrutinize and battle-test,” which incorporates the typical cost of commercial software and the hypothetical cost to build stuff with AI to equal or replace the commercial stuff.

Jaluria compared what it costs to build an AI app versus buying one from Hubspot, Microsoft and other established software vendors. His conclusion is that the “total cost of ownership,” TCO, of making something from AI is not worth it.

The week ahead is ferociously busy for earnings with Apple, Meta, Amazon and Microsoft and other tech giants, but also interesting for a host of lesser-known companies.  

The table at the bottom of the post features ninety-one companies — ninety-one! — reporting earnings this week. Hey, if they ever change the rules for quarterly disclosure, this might be one of your last times to pig out on such a buffet. 

I expect to hear bad news from Apple, Amazon, Microsoft and Meta. Apple is grappling with the rising cost of DRAM, with its attempts to secure the U.S.’s approval of purchase of DRAM chips from China’s CXMT still very much in the headlines. Microsoft and Meta will likely, similar to Alphabet last week, be in the penalty box for their skyrocketing capital spending plans.

For others, such as Reddit and Coursera, the story will be a mixed bag of whether they can survive the erosion of AI of their content business. Reddit reports Thursday evening.

Consider instead several of the more interesting names, most under a trillion dollars in market cap.

This first full week of earnings reports showed what looks like a turn away from the semiconductor beneficiaries of the artificial intelligence trade and toward the beleaguered software group. 

Shares of TE Connectivity, Texas Instruments, MaxLinear, Intel, and BE Semiconductor, all of which had very favorable reports, sold off hard. All three had been up by a huge amount this year prior to their reports, as you can see from the table below. 

Google also sold off hard despite robust results reported Wednesday evening. 

Conversely, software makers SAP, RingCentral, VeriSign and Wex all had significant gains after trailing the market this year. 

Thursday was a banner day for followers of the microprocessor trade, with Intel reporting results after market close and Advanced Micro Devices holding what has become an annual affair in San Francisco, its update on its artificial intelligence chips.

Intel initially spiked double-digits in late trading, but gave up some gains but is up just 2% in early Friday trading, at $102.87. AMD shares traded down two points during Thursday’s regular session.

The key for Intel still is the “foundry” business, bringing in outside customers to use Intel’s factories.

The song remains the same at Intel, in a sense: Going back in time to the Gelsinger days, Intel has always had interested parties, and never found a large customer. In years past, the PDK simply didn’t deliver the goods. Hopefully, with Tan’s rigor, and the sense of opportunity, this 14A will be the real deal. We may know in October.

With turmoil over Chinese AI models, worries about AI infrastructure’s private credit market exposure, and persistent questions about return-on-investment of massive spending, it was no surprise the Street spent all of Wednesday evening peppering Alphabet’s Google’s CEO, Sundar Pichai, and CFO, Anat Ashkenazi, about all those things.

It sounded to me like no one got a straight answer about any of it. As a result, Alphabet shares sold off by five percent in early trading Thursday morning.

The report itself was favorable. Google’s cloud revenue surged 85%, year over year, to $24.8 billion, an astounding rate at such scale and higher than expected. The cloud business now has a remarkable backlog of business of $514 billion. Google CEO Sundar Pichai noted that the use of “tokens,” the unit of currency for AI, rose by 40%, quarter over quarter, to 22 billion tokens per minute. And five hundred Google Cloud customers have now exceeded a trillion tokens in usage apiece in the past twelve months.

With shares of Micron Technology up 235%, year to date, at a recent $955.60, including more than doubling in the past ninety days, the question of how long a very strong DRAM and NAND flash market can last is one of the most interesting questions in the very hot AI chip market.

The question implicates shares of competitors Samsung Electronics, which have more than doubled this year, and SK Hynix, whose newly minted American Depository Receipts are down two percent from their first-day close at a recent $166.65; and NAND maker Sandisk, which is up an astounding 470%.

Three brilliant technology analysts dug into the details of what might continue to make Micron stock work. You don’t have to go through them because I’ve extrapolated and summarize the high points for you.

Once again, the tech trade is rebounding on Monday morning. Shares of many of the TL20 stocks to consider, all of which, just about, are linked to a AI, are on the march higher. Fiber-optics component maker Lumentum is one of the strongest, up 8% at $790.74. Sandisk is also doing well, up 6%. I don’t think there’s any special significance here, as I indicated in this morning's podcast. There is simply a reversal of the trade, a continued search for direction…

UBS MAKES THE CALL TO GO BOTH WAYS!

Among the many who are searching for direction in Street research this morning are UBS strategists John Talbott and Jonathan Carson, who offer two different stock screens. One screen is a guide to look for bargains in beaten-down names, the other one consists of suggestions for how to reposition oneself, assuming the recent tech sell-offs are some kind of lasting “reset” of stocks.

Sigh. Perhaps slightly more certain is life on LHS 1140b, a small planet forty-nine light years from Earth that has been found to have a helium-based atmosphere, and therefore might support life, as reported today in Science magazine by researchers at Harvard University…

SO MANY AI GIGAWATTS, SO MANY WAFERS

Meanwhile, back on earth, makers of semiconductor capital equipment, such as Applied Materials, Lam Research, and KLA are now actually in “correction territory,” according to Bernstein analysts Stacy Rasgon, David Dai and colleagues in a note this morning. Despite the correction, “The group is still up >80% year-to-date,” they note, “with heavily expanded valuations,” they observe.

The week ending July 17th was one of the worst weekly returns ever for the AI-heavy TL20 group of stocks to consider, down 10%, though still up 47% for the year. 

This is not the beginning of the end, simply a cooling-off of what had been a parabolic move upward.

There are a lot of explanations. The folks at Citrini Research contend that investors just don’t get how important AI is.

Bloomberg is talking about how the latest breakthrough Chinese AI model, “Kimi K3” from Chinese startup Moonshot AI, is the source of market panic. 

Fact is, there will steadily be a commoditization of technology that has been touted as miraculous by OpenAI, Anthropic, SpaceX, etc., but can be replicated fairly easily.

Shares of Taiwan Semiconductor are trading down in Thursday pre-market trading by 4% at $402.22 despite a very robust outlook from the company overnight.

As I wrote on Monday, the revenue for the June quarter was already known, so the issue at hand was the forecast for the rest of this year and beyond.

My sense is the stock is down because of another big increase in spending, which is a good thing and a bad thing.

Shares of the entire AI trade are falling hard today, including a 9% drop in Micron Technology stock, and one possible explanation is a report late Tuesday by Reuters’s Max Cherney stating that CoreWeave, the neo-cloud computing provider, has been talking with bankers about ways to hedge against a decline in memory-chip prices, citing unnamed sources.

“AI cloud computing company CoreWeave is exploring the use of financial derivatives as a potential hedge against a future drop in memory and storage chip prices,” writes Cherney.

Cherney relates that CoreWeave is locked into pricing agreements with Micron — something I don’t believe either company has actually disclosed — and that CoreWeave wants to make sure it has a way not to overpay if memory prices drop. For that reason, relates Cherney, CoreWeave is considering, among other things, use of puts, contracts that would allow CoreWeave to sell its contracted chips if memory prices drop below the contract price down the road.

Shares of IBM are down 23% in Tuesday morning trading at $224.91, after the company preannounced the June quarter’s revenue and profit below consensus as a result of mainframe sales being hit by a shift in customer spending to servers, storage and memory chips — all the things that are being consumed in massive quantities to fund the AI buildout.

Regarding ASML’s report tomorrow morning of its June-quarter revenue and profit, Jefferies & Co.’s Jenardan Menon relates to clients today that “it feels as if there is an awful lot riding on tomorrow,” observing that “if ASML cannot increase its shipments, then TSMC [Taiwan Semi] cannot raise capex [capital spending], and Applied materials and WFE [semi equipment] expectations will be difficult to achieve, etc., etc.”

Shares of data center operator Terawulf are down 7% Tuesday at $19.22, and other AI data center names are being hit hard as well, after New York State governor Kathy Hochul Tuesday signed an executive order pausing for a year any construction of data centers in New York of fifty megawatts or larger while the potential effects of the data centers can be studied.

We’re heading toward earnings reports this week for ASML and Taiwan Semiconductor, two of the most consequential names of the AI trade, and that happens amidst a tug of war between deepening skepticism over AI investments and continued positive trends for companies such as Micron Technology and SanDisk that serve the AI trade.

Shares of the heavily AI-influence TL20 group of stocks to consider rose 4% for the week ended Friday, July 10th, a recovery from two weeks of selling prior to that.

For the month of July, the TL20 is up 2% through Friday’s close, and up 64% for the year — way better than the 3% decline this month of the Nasdaq Composite Index and the 13% gain of the Nasdaq thus far this year.

Ahead of Taiwan Semiconductor’s quarterly report on Thursday, the key number has, as usual, already been disclosed in TSM’s monthly sales report overnight. As usual, it’s above expectations.

Taiwan Semi’s June sales rose 68%, putting the quarterly total, $1,270 billion, in New Taiwan dollars, ahead of Street consensus for $1,254 billion.

Remember that April’s report, though very strong, saw the stock weak because management didn’t raise TSM’s outlook for its capital spending this year, causing some concern the business might not be as white hot as it has been.

So, the issue is really not the quarter but outlook for the second half of this year, as Wedbush analyst Matt Bryson points out in a note Monday: “The question in our view is how much can TSMC potentially exceed current 2H expectations (with any shortfall unlikely in light of the strong demand backdrop),” he writes.

Some of the parties that have been software’s strongest defenders amidst the threat of AI destroying software are starting to throw in the towel. 

Jackson Ader of KeyBanc on Wednesday cut his rating on shares of Salesforce to Sector Weight from Overweight, and he reflects back on the vigorous defense of software he’s been carrying on since August. 

Ader had previously argued the biggest software firms such as Salesforce were too big to be replaced by new stuff cobbled together with AI.

But, he relates, customer feedback is not good for Salesforce. “We attend more Salesforce partner and customer events than any other company in our coverage, and feedback from those customers has been consistent in two ways,” writes Ader. “1) customers' data is not in order to do meaningful AI work; and 2) Agentforce, as a product, just isn't there.” He’s referring to Salesforces offering for “agentic AI” to automate tasks such as sales prospecting.

I’m not surprised this is turning out to be a dud.

If you read through the 398-page prospectus for the expected offering Friday of American Depository Shares of SK Hynix, the Korean maker of DRAM and NAND, you will be struck by the lack of important information.

Hynix is the second-largest memory-chip company after Samsung Electronics, also based in Korea, and ahead of Kioxia (a spin-out of Japan’s Toshiba), Micron Technology and SanDisk. Hynix is also the top maker by revenue of “high-bandwidth memory,” HBM, currently the cutting edge of DRAM that is crucial for AI applications.

The company already has common shares listed on South Korea’s Kospi exchange. This $28 billion offering in the U.S. is meant to raise funds for the company’s factory expansion in Korea and China by tapping into the recent fervor for memory-chip stocks. Hynix’s ordinary shares are up 236% this year, slightly ahead of Micron’s 232% rise.

Given that Hynix is funding manufacturing expansion, what you would like to know from its prospectus, what’s crucial, is the shape of the global market for DRAM and NAND. That’s key because in a commodity industry, profit and peril rest on the control of supply to prevent over-supply situations.

It’s been a distressed market for software companies for a long time now, and that has proven fortuitous for forty-five-year-old Progress Software, which has built a billion-dollar business rolling up companies that might just have gone away otherwise.

Those privately held companies that Progress buys have promising technology, maybe $70 million to $100 million in annual sales, and their valuations are often at rock bottom. They are not distressed assets, but their small size, modest growth and minimal profit means they probably wouldn’t make it on their own, says Progress CEO Yogesh Gupta.

They might not want to sell at low valuations, but when Gupta meets with them, “I give them a reality check,” Gupta says, in a chat we had via video last week.

Tech is falling out of bed again today, with the TL20 group of stocks to consider down over 5%, the heaviest-hit being Bloom Energy and SanDisk.

The proximate cause is what appears at first blush a very positive preliminary profit report for the June quarter from Samsung Electronics overnight, which, nevertheless, sent shares down over 7%.

A WHOLE LOTTA RUMSFELD IN SPACEX

Even a raft of bullish notes on SpaceX cannot lift the mood, with SpaceX stock selling off by 5% at $152.84 — still below the first-day close on June 12th of $160.95.

The coverage is almost universally bullish, and some analysts are projecting prices as high as $900 — over time, mind you. The average target of all analysts at the moment is around $250.

Among the most bullish, Citigroup’s John Godyn starts coverage with a Buy rating and a $200 price target, but, “we believe this price target is a milestone along the path to $900.”

Wow. I remember when a price target was just a price target, not a “milestone.”

This morning, I joined a video call with Merrill Lynch analysts Savita Subramanian and colleagues in which they offered their mid-year outlook. That included a negative slant on the AI trade, no surprise. Not only are valuations elevated, but the group are “starting to become a little clunkier,” as Subramanian put it.

What she meant is that the hyper-scalers, such as Alphabet’s Google, Microsoft, Amazon, Oracle, Meta, are more capital-intensive than energy exploration and production companies such as Exxonmobile.

The hyper-scalers are “all as capital intensive as one another, and, in fact, some of the hyper-scalers are more capital intensive than Exxon, Chevron, and ConocoPhillips,” said Subramanian.

My former Barron’s colleague Shuli Ren, now a Bloomberg columnist, penned a nice column this weekend about how legendary trader Michael Burry has written on his Substack that he is short Micron Technology stock.

Burry is one of the protagonists of the 2015 movie The Big Short about the subprime mortgage collapse that prompted The Great Recession of 2008, based on the novel of the same name by Michael Lewis.

The temptation is obvious to make an analogy between the current artificial intelligence data center craze and the mortgage mess. Like the housing bubble eighteen years ago, a major bright spot in the economy currently is one industry, spending on data centers. And that spending is to support an obsession with AI “tokens” that is questionable.

It’s tough to be in an industry that represents a commodity because it is driven by supply, and the slightest whiff of extra supply becoming available throws into doubt the economic fortunes of the suppliers. (In non-commodity businesses, there’s only one or two suppliers, so supply shocks are less likely.)

That’s the situation faced by both DRAM maker Micron Technology and CoreWeave, the lessor of compute capacity that came public in March of last year. I have slightly more confidence in the survival of Micron. 

Wednesday, both stocks were rocked by news of new supply. 

Both Micron and CoreWeave are slightly higher Thursday morning.

I’m less concerned about Micron given that it takes years to build new factories for DRAM and NAND, which means that increasing aggregate supply for a DRAM and NAND industry that is strapped for supply cannot happen overnight.

In the case of cloud computing and AI data centers, it’s not easy, perhaps, but it is less complex to bring on supply of more GPUs, and that’s precisely what CoreWeave investors fear if Meta is going to rent out its own computers.

I’ve been negative of late on the over-valued AI companies such as SpaceX and OpenAI and Anthropic, and I wanted to clarify that I’m not entirely negative on tech.

We’ve seen real worries about the AI trade that have made for a volatile market of late, even though we just closed the best quarterly performance in the U.S. stock market in six years.

As an example of the volatility, last week, the TL20 group of stocks to consider was down 6%; this week it’s back up 5% as I record this.

Chip equipment prospects are going up again, in what seems an endless run of encouraging observations. 

Tuesday, Mehdi Hosseini of Susquehanna Financial raised his outlook for six names in his coverage, Applied Materials, ASML, Lam Research, Teradyne, KLA, and Advanced Energy, after concluding customers are paying a premium to get ahold of equipment.

The stocks are all rising sharply Tuesday, by over 7% in a couple cases.

Hosseini now models total chip equipment, “WFE,” or, the wafer fab equipment market, the catch-all term for the industry, speeding up its growth: Industry sales may total $142 billion this year, up 60%, with another hike of 61% to $196 billion in 2027, and a rise of 69% to $250 billion in 2028.

Quantum computing hopeful Quantinuum, which spun out of Honeywell in a $1.7 billion public offering on June 4th, is down 5% this afternoon at $71.95 despite most analysts today initiating coverage of the stock with a Buy rating or the equivalent, and price targets averaging $94, which would be a pretty terrific return from here.

The stock is still above the offer price of $60.

‍As is the case with so many startups, Quantinuum is being valued by bullish analysts based on potential returns way out in the future. Analysts are mostly using a multiple of sales or of profit based on 2030 projections, which means there’s a lot of hope baked into the price targets. 

I regard Quantinuum as a startup business that has to prove its technology is “scalable,” meaning, can be reliably improved generation after generation to reach a level that is commercially valuable. Every single one of Quantinuum’s supposed advantages is set against a backdrop of an industry that is immature. None of the players have proven the industry is ready for prime time, as I laid out in an October feature

Evercore ISI’s Mark Lipacis, starting the stock at Outperform with a $98 price target, lays out what’s at stake for each generation of machine: “Sol introduces the 2D sortable-array trap and integrated, scalable control electronics – which will prove to be the architectural foundation for everything after it,” he writes. “Thus a Sol slip would cascade.”