The following table offers an update on where the TL20 group of stocks to consider stand as far as valuations, as of close of trading on May 1st.

Most but not all of them have seen their P/E multiples and their price-to-sales multiples surge with the 36% gain of the group this year.

There are a few discounts, which I’ve marked in red in the table. The most amazing item is Nvidia at twelve times projected sales, down from eighteen times six months ago. Nvidia is also one of the weaker performers this year, up only 6%, below the Nasdaq’s 8% rise.

A very big week of earnings reports drove stocks higher, the Nasdaq Composite Index and the Standard & Poor’s 500 Index closing up 1% for the week, and the TL20 group of stocks to consider up 3% for the week. 

This week’s gain brings the TL20 to a gain of 428% since inception in July of 2022.

The biggest item from the earnings reports was the confluence of four reports on Wednesday evening from Alphabet, Amazon, Meta and Microsoft. 

As I had predicted when I went on The Schwab Network on Tuesday, Alphabet and Amazon showed up the best of the four that night, while Meta and Microsoft sold off. 

I think Alphabet’s Google and Amazon are simply better prepared for the epoch we are in of tight chip supply because they both began their AI chip efforts a decade ago or more, while Meta and Microsoft were wasting time on things such as “The Metaverse.”

The artificial intelligence stuff that everyone is so crazy about, the stuff they are spending hundreds of billions on, has a number of problems. Some are very widely publicized, such as the notorious “hallucinations,” which are more properly called “confabulations.” And the newer versions of AI, called “agents,” can go rogue, causing tremendous damage to computer systems.

But the biggest unreported risk is the potentially skyrocketing cost for every task such agents perform. That is a potential disaster not too far down the road. I expect lots of frustration and disappointment in enterprises as that unfolds. I also expect some interesting opportunities for public stock investors which I’ll come to in a moment.

The artificial intelligence stuff that everyone is so crazy about, the stuff they are spending hundreds of billions on, has a number of problems. Some are very widely publicized, such as the notorious “hallucinations,” which are more properly called “confabulations.” And the newer versions of AI, called “agents,” can go rogue, causing tremendous damage to computer systems.

But the biggest unreported risk is the potentially skyrocketing cost for every task such agents perform. That is a potential disaster not too far down the road. I expect lots of frustration and disappointment in enterprises as that unfolds. I also expect some interesting opportunities for public stock investors which I’ll come to in a moment.

THE LOOMING TOKEN NIGHTMARE

Agents are intended to automate some of the work of humans by doing things such as looking up information in a customer database or processing an inbound invoice.

A new report from the University of Michigan and collaborators, titled, “How do AI agents spend your money,”  makes clear that the cost can quickly get out of control, and that the cost bears no relation to the quality of work done, and that there is no reliable way to anticipate those costs.

Shares of Twilio are up 18% in early trading Friday morning at $174.96, after the company topped expectations Thursday evening with profit and revenue and forecast higher as well. It was one of the best reports in a big night for software that included Atlassian, Reddit, and Roblox.

Atlassian shares are up 18%, Reddit is up 9%, and Roblox is down 24%.

I had a chance to speak with Twilio CEO Khozema Shipchandler Thursday about the results and his take on the change in Twilio’s profile that he has been engineering for the past eighteen months or so.

“We had one of the strongest quarters in the hisotry of the company,” said Shipchandler. “Fundamentally, we’ve transformed the company.”

Readers who’ve followed the five prior interviews I’ve done with Shipchandler (see box at right) know that Shipchandler has been working to lift the company’s level of sales growth and to expand profit margins.

Wednesday night’s reports from Alphabet and Amazon had a more positive effect for those two companies’ stocks than did the reports of Meta Platforms and Microsoft, which I would say was consistent with my preview early in the day outlining better prospects for the former two.

And most interesting to me was that we got more detail on both Alphabet and Amazon as competitors to Microsoft. 

The stock movements this morning show the disparity: Alphabet shares are higher by 8% in early trading Thursday morning at $376.36; Amazon stock is up 3% $269.70; Meta shares are down 9% at $606.36; and Microsoft is down 2% at $416.20.

Heading into tonight’s Big Four set of reports from Alphabet, Amazon, Meta and Microsoft, one fact stands out more than any other: preparation.

Alphabet’s Google and Amazon’s AWS have been preparing for the AI epoch for a decade or more; Meta and Microsoft have not.

I cannot tell you if that means Alphabet and Amazon stocks will pop and Meta and Microsoft shares will sag tonight. But preparation makes the former two the better general AI bets on any given day than the latter two.

Alphabet began over a decade ago developing the world’s first large-scale project for a custom AI processor, what was eventually introduced in 2016 as the “Tensor Processing Unit,” or TPU. That chip was specialized for AI work inside  Google’s cloud, a distinction from the “general purpose” GPU chips that Nvidia sells to the world.

The entire AI trade is weaker on Tuesday, in no small part because of the article this morning by Berber Jin in The Wall Street Journal, OpenAI misses key revenue, user targets in high-stakes sprint toward IPO.

Jin, citing multiple unnamed sources, writes that Sarah Friar has told management that the company “might not be able to pay for future computing contracts if revenue doesn’t grow fast enough,” and that there is an effort to control costs at the company.

Jin cites a “joint statement” from Friar and CEO Sam Altman stating, “We are totally aligned on buying as much compute as we can and working hard on it together every day.”

It’s an important story because it’s a preview of what happens when OpenAI comes public. Everything changes once a company is public.

Earnings reports are seeing a lot of red Tuesday morning, as of the companies reporting overnight are trading down in early trading, including two TL20 names, Corning and Celestica, down 10% and 14%, respectively, and also chip-design software maker Cadence Design Systems, down two points.

Surprisingly weak are shares of semiconductor packaging firm Amkor following a strong report given that chip packaging is one of the hot areas these days.

I attribute the sell-off of Celestica, Corning and Amkor to elevated expectations and elevated stock prices. The results were fine, however, the shares of Corning were up 92% heading into this report, and the same for Amkor, and Celestica is up 43% since the start of the year.

There’s been some discussion about press reports last week that Elon Musk’s SpaceX has an option to purchase AI coding startup Cursor for $60 billion. Cursor makes a tool that lets software programmers use the AI from Anthropic, OpenAI and others to automatically write software instructions.

Bloomberg’s Edward Ludlow Monday morning writes that if Musk gets control of Cursor, it could be a big problem for Nvidia CEO Jensen Huang, as his firm is also an investor in Cursor. Entrepreneur Chamath Palihapitiya wrote in his newsletter Sunday that a deal could give Cursor access to much more compute power from Musk’s xAI, and give xAI an entrée into enterprise computing, being a “win-win” for the two.

I think a lot of this is much ado about nothing. I suspect Cursor may end up being a footnote.

The week ending April 24th was another strong one for tech, the Nasdaq Composite Index up 2% and the TL20 group of stocks to consider up 5%. That brings the TL20 to a return of 32% since the start of the year, and 415% since inception in July of 2022.

The week’s returns were dominated by positive earnings reports from semiconductors companies Texas Instruments, STMicroelectronics, BE Semiconductor, Lam Research, and MaxLinear.

They're all benefiting from what I have called the semiconductor renaissance, which is the continued importance of the cutting edge of computer chip development, and how it is powering all of what we're seeing with the AI trade.

Kudos to HSBC’s Frank Lee, who on Monday raised his rating on Intel shares to Buy from Hold just two days before the quarterly report, which turns out to have been “stellar,” as one analyst put it. Revenue beat expectations by nine percent, the most in years, and profit of 29 cents crushed the consensus of two cents.

The stock is up 22% on Friday at $81.71, for a gain this year of 123%.

The stock has gotten three upgrades this morning, from Citigroup, Roth Capital, and Evercore ISI.

The main takeaway from the report and the call with CEO Lip-Bu Tan is that data center chip sales — the Xeon server business — is going to end up much better than expected, as Lee predicted.

Things that happened Thursday: Shares of chip equipment giant Lam Research sold off by three points to close at $258.56 despite a very strong report Wednesday evening that drove price targets as high as $385, and even the bulls on the stock concede that the valuation has gotten “rich,” which for the moment is deterring additional buyers (though, as I’ve written, some stocks are pricey for good reason); in the latest sign that the private market has gotten lost its discipline with respect to AI valuations, Reuters’s Juby Babu relates that Elon Musk’s SpaceX has said it has an option to purchase AI coding startup Cursor for $60 billion, an incredibly high valuation for a startup that probably doesn’t yet have a billion dollars in annual revenue (and which I expect may end up being obsoleted by Anthropic); shares of BE Semiconductor rose almost four percent following better than expected results and outlook, and management said all three big memory-producers, Micron, Samsung, and SK Hynix, are evaluating the company’s “hybrid bonding” tools for making “high-bandwidth memory,” HBM, somewhat relieving worries that popped up in March, and prompting Needham & Co.’s Charles Shi to declare that “the debate on ‘if’ HBM is happening or not should be settled.”

CEREBRAS FILES FOR IPO‍ ‍

One of my favorite private chip companies, Cerebras Systems, on Thursday filed its S-1 prospectus for an initial public offering after having filed back in 2024 only to withdraw the filing.

I wrote in 2024 that memory chips, DRAM and NAND flash, in particular, are key to artificial intelligence because of the enormous amount of bits and bytes that need to be processed. That structural change in computing keeps showing up again and again.

The latest piece of evidence is the report early Thursday morning of SK Hynix, the South Korean manufacturer of DRAM and NAND that competes with Micron Technology and Samsung Electronics, forming the Big Three of memory technologies.

SK Hynix shares rose fractionally in Seoul trading while Micron stock is down a point in early trading in the U.S.

Hynix’s revenue and profit easily topped consensus, but what was most interesting were remarks by Kim Woohyun regarding the “spot” price of memory chips. The spot price is the price on the open market, what someone pays when they spontaneously go to buy the chips rather than buy at a price established via contract.

Another quarter, another win for semiconductors, and another big disappointment for software.

Shares of Texas Instruments are up eleven percent in late trading Wednesday evening at $262.99 after the biggest quarterly upside in four years, and fellow analog chip makers are rallying sharply, with ON Semiconductor and Microchip Technology up five percent, and Analog Devices up almost as much.

Shares of ServiceNow, however, are down thirteen percent  at $88.90 despite the company raising its outlook for revenue this year linked to artificial intelligence. Just one metric, “current portion of remaining performance obligation,” or, cRPO, came in a tad light. That just shows how little patience software investors have these days! Price targets are dropping sharply on ServiceNow, as low as $85.

Things going on this morning: Shares of energy technology giant Vertiv are down 3% Wednesday after the company’s earnings report in the morning offered a quarterly revenue outlook below consensus (stock is up almost 90% since the start of the year); Seagate shares are up a couple points at $573.31 as Barclays analyst Tom O’Malley upgraded the stock to Overweight from Equal Weight just a day before Seagate’s earnings call, with a $625 price target, declaring “We think drives are the most interesting play in the storage/ memory hierarchy” (he also likes Micron Technology, which is cheap at a P/E of four times) — note Seagate stock is up 109% this year;

GOOD IS NOT GOOD ENOUGH FOR SOFTWARE

Software maker Twilio continues a rising tide of approval as Merrill Lynch’s Koji Ikeda on Wednesday becomes the latest fan, raising his rating on the shares to Buy from Underperform as part of a broad review of “infrastructure” software.

The week that ended April 17th saw the indices and also the TL20 group of stocks to consider “go parabolic,” meaning, a sharp upward movement in a very compressed period of time.

‍That’s nice, but also a point of concern given that some stocks are at elevated valuations, and given the fragility of the “AI trade.”

A few weeks back, the TL20 had erased all its gains for the year, but then it came roaring back since the start of April. The TL20 is now up 26% for the year and up 391% from inception through April 17th.

Things this morning on the Street: Software maker Dynatrace’s shares were initiated at a Buy by Goodman Sachs software analyst Gabriela Borges with a $45 price target, emphasizing the importance of “observability” to AI deployments, although Dynatrace shares are not responding, remaining flat today at $35.42; UBS analyst Manav Gupta hiked his price target on shares of Bloom Energy to $251 from $170 based on an expectation for higher-power GPU racks in data centers, boosting the stock 4% to $228.20; Citigroup analyst Vikram Bagri also raised his price target on Bloom, to $229 from $162; and UBS’s Karl Keirstead cut his price target on Snowflake stock to $210 from $235, writing that the stock still cannot get out from under the “AI destroys everything” worries for the software industry, with Snowflake stock rising fractionally today to $151.51 for a 31% decline this year.

INTEL DOUBLE BOOST

Elsewhere, shares of Intel are up fractionally at $65.77 following a double upgrade Tuesday morning.

I will remember Tim Cook’s fifteen-year tenure at Apple mostly for the financial overhaul of the company. Activists Carl Icahn and David Einhorn began agitating for shareholder returns shortly after Cook took over the company  in 2011 from the late Steve Jobs, who showed little interest in dividends and buybacks.

Cook instituted massive programs for both, driving dividends to fifteen billion dollars annually and share repurchases to over ninety billion dollars annually. The dividend, now $1.04 per share annually, arguably drove a new era of capital returns focus among U.S. equities. In mollifying two of the Street’s loudest voices, Cook lead the way.

Cook and former CFO Luca Maestri (now head of corporate services) also arguably changed the approach of big tech toward balance sheets, enforcing a philosophy of “zero net cash,” which Apple has steadily worked to achieve.

Here’s Apple’s history of capital returns:

I have to appreciate the candor of Josh Claman, chief executive officer of data center technology startup Accelsius. Shortly into an interview with me last week, he remarked with a sigh, “this kind of conversation really has no upside for me,” something he repeated several times during our talk.

Accelsius is privately held, so Claman has nothing to gain from answering investor-style questions. But forty-three percent of Accelsius belongs to public conglomerate Innventure, about which I’ve written at length. That makes everything that Accelsius does of some interest to Innventure investors. So, Claman’s handlers offered him for a chat via Zoom, and he was a good sport about it.

Accelsius makes a fascinating technology to cool computer chips in the AI data center. I wanted to know how fast sales of those chips will ramp because we only get very broad outlines from Innventure management.

It turns out that the question of when is a complex question. There is a certain inevitability to what Accelsius is building that seems like it will lead to a tipping point. And yet, there are a lot of hurdles in the market that making timing difficult to gauge.

Shares of Taiwan Semiconductor Manufacturing are down a couple points at the open, at $366.36 for the American Depository Receipts on the New York Stock Exchange, despite a strong report this morning, and an increased outlook for this year. 

The proximate cause of the stock weakness, I would say, is the company merely reiterating its outlook from January to spend $52 billion to $56 billion in capital on plant and equipment this year, with CFO Wendell Huang saying the actual spending may be “towards the high end.” 

Even though CEO C.C. Wei emphasized the positive take, that “supply is tight and demand continues to increase,” nevertheless, not increasing the actual dollar forecast of spend leaves some investors without what they would consider a sign of even greater confidence. 

“This may have fallen short of more aggressive investor expectations” for capital spending, writes Krish Shankar of TD Cowen, a bull on the stock.

Shares of chip equipment giant ASML are down five percent Wednesday at $1,437.18 for the Nasdaq-listed American Depository Receipts, mainly reacting to a lower-than-expected sales outlook for this quarter (but an increased outlook for the full year 2026.)

What adds a little salt to the wounds today is that this is the first quarter that the company ceased reporting the number of orders it received for equipment, something they warned us a year ago they would do.. 

The argument a year ago from ASML’s CEO, Christophe Fouquet, and CFO, Roger Dassen, was that order rates can be volatile, quarter to quarter, so, it’s best not to obsess over them quarter to quarter. But, when a given quarter’s forecast doesn’t totally delight, as this one didn’t, order data would have been a little bit of a comfort. And, to add even more salt, the January report, the last quarter in which they did disclose, had been a blow-out report at the time, with order numbers coming in way above expectations

Shares of Bloom Energy, makers alternative energy products in the form of a solid-oxide fuel cells used to power data centers, soared 24% on Tuesday to close at $219.03, after the company late Monday announced an expansion of its relationship with Oracle as a customer, which prompted at least one bear on the stock to ease up on his negative view.

The Oracle deal has the potential to deliver 20% to 50% upside to the Street’s revenue estimates, writes Jefferies & Co. analyst Dushyant Ailani, who raised his rating on Bloom shares Tuesday to Hold from Underperform, with a $187 price target, up from $97.

Ailani is “nearly on the bandwagon” with the stock, he writes, but what is holding him back is valuation. “We view valuation as stretched,” he writes, adding, “Bloom epitomizes retails infatuation with 'new' technology and data centers with the valuation divorced from fundamentals.”

One of the newer chapters of the frantic artificial intelligence trade has been the positive view on the operators of so-called content distribution networks, or, CDNs, principally Fastly and Akamai. Those stocks were two of the only bright spots in software and services, rising as much as 229% through early April in Fastly’s case, and 37% in Akamai’s case.

Last week those gains began to unravel fast as investors realized they blew the notion of AI “agents” out of proportion.

The bullish argument in favor of Fastly and Akamai and others such as Cloudflare, and privately held Vercel, is that lots of AI programs in the form of agents would run around the enterprise, creating a surge in traffic across networks, which would be a direct boost to the companies selling the ability to manage network bandwidth.

There is a scarcity of advanced chip-manufacturing capacity, notes Northland Securities’s Gus Richard on Monday, raising his price target on Intel shares to $92 from $54. That would represent a return of about forty percent from Monday’s close of $65.18.

Richard reiterates an outperform rating on Intel stock.

Richard emphasizes that there just isn’t that much sophisticated manufacturing capacity extant, and so he believes Intel should trade at three times the value of its manufacturing property, plant and equipment (PP&E).

“Intel now has 3 fabs capable of 3nm and below,” he notes.

“3nm capacity is expected to tighten in CY27, increasing the value of 3nm fabs.”

Richard notes Intel has come a long way:

Rumors are swirling that Nvidia could be considering buying Dell or HP or another PC-related company following a report Monday by Charlie Demerjian of the tech blog SemiAccurate.

Dell closed up 7% and HP shares closed up 5% Monday. Nvidia shares rose fractionally.

Citing what he says has been a year of his own research, Demerjian writes, “Nvidia is looking to make a huge purchase that will reshape the PC and server landscape like nothing else has done since the computer was invented. And let us say again, we are dead serious here.”

I don’t have any information beyond Demerjian’s speculation — and I give him the benefit of the doubt that he’s come up with real signals, real chatter, what have you.

My interest is in Nvidia’s enviable profit versus the PC makers. Nvidia’s gross profit margin at the moment is about 75%. Dell’s margin is 19%, and HP’s is similar.

Solid-state drive maker SanDIsk has been an amazing performer, its shares up 301% this year, and almost 3,000% since it spun out of Western Digital last year, at a recent $952.50.

As SanDisk stock keeps rising, there’s quite a bit of anxiety that the very healthy market for NAND-flash-based drives is bound to cool off, as prices have become a deterrent to further consumption.

NAND, and NAND-based devices such as solid-state drives, are traditionally a commodity, and that has traditionally lead to vicious cycles: sharp swings in the market, driving up prices and margins only to drive them down sharply again.

On Monday, Amit Daryanani agues things are different. The market is still cyclical, but profit will hold up better for SanDisk this time because the company is “structurally” more profitable. He starts the stock at Outperform with a $1,200 price target. That’s based on 12 times what he projects as $100 per share in earnings in the fiscal year ending June of 2027. (He’s slightly higher than consensus for about $99.)

The week that ended April 10th, 2026, was a solid one for stocks as the “AI trade” was reinvigorated by the PR stunt of startup Anthropic, even as plain-old software names Nutanix and ServiceNow got downgraded.

Shares of the TL20 group of stocks to consider soared a collective 12% for the week, lead by shares of Broadcom, Micron Technology, ASML and SanDisk, up 18%, 11%, 13% and 18%, respectively. The only thing doing really poorly in AI last week was Palantir, down 13% and now down 28% year-to-date.

The Nasdaq Composite Index was up 5% and the Standard & Poor’s 500 Index, up 4%.

Anthropic said its “Mythos” AI model is so powerful it will not release it generally to the public but instead share the model with a select group of firms including Google, Nvidia and Cisco. It will do so in conjunction with a cyber-security framework it calls Glasswing.

You would think one of the areas of software most vulnerable to being dis-intermediated by artificial intelligence would be that of “observability,” also sometimes called “DevOps.” 

That is software that monitors programs running inside companies to make sure they’re performing right. Those tools, of which Dynatrace and Datadog are the main providers, could conceivably be replaced some day by the kinds of automation emerging every single week from the AI giants, such as the “Glasswing” software that Anthropic promoted this week.

Every time those companies show their resilience, such as Datadog’s earnings report in February, their stocks sell off anyway. Datadog stock is down almost 20% this year, and up 17% in twelve months, vastly trailing the Nasdaq. Dynatrace has been even worse, down 22% this year and the same amount over the past twelve months.

But, the DevOps stuff will prove more resilient than expected, writes Howard Ma of Guggenheim Securities, who on Thursday raised his rating on Datadog to a Buy from Neutral, after concluding that there is plenty of stuff that AI won’t be able to do on its own.

It’s not every day that a bunch of incredibly powerful companies that are in bed together stage a PR stunt, but it does seem like it’s happening more and more.

This week, AI startup Anthropic, which is valued at $380 billion as of its latest $30 billion round of funding, announced that it a) has a new kind of AI large language model in the works, called “Mythos,” and b) that it is not releasing Mythos to the public because it’s too powerful, but that it will instead provide the model to various parties that are cooperating to address thousands of flaws that Mythos found in software. 

The cyber-security effort is called Project Glasswing, and, surprise, it brings together mostly Anthropic investors. Project participants including Alphabet’s Google, Microsoft, Nvidia, Amazon, Cisco Systems, and JPMorgan Chase, all of which have put money into Anthropic, according to FactSet Systems. 

In addition, Broadcom is part of the effort. As I mentioned Tuesday, Broadcom is selling racks of Google’s “TPU” AI chips to Anthropic, making both Broadcom and Google vendors to Anthropic with a financial interest.

“Did the agent double-check all the things that might be wrong? Did it make assumptions, and, inasmuch as you could even say that it's making an assumption, did it get confused somehow?”