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?”

Shares of Broadcom are up almost six percent Tuesday at $332.77 after the company announced in a federal filing after market close on Monday that it inked an agreement with Alphabet’s Google to supply the company with Google’s “Tensor Processing Unit,” or, TPU, its custom AI chip, and related technology through 2031.

Broadcom and Google have collaborated for years to produce the TPU, which Google designs and Broadcom then helps to shepherd through the manufacturing process with Taiwan Semiconductor Manufacturing. So, this is an extension of an existing arrangement.

In the same filing, Broadcom confirmed a deal first announced in December with AI startup Anthropic, saying it will supply 3.5 gigawatts of computing capacity to Anthropic, in the form of complete rack systems using the TPU, through 2027.

The week that ended April 3rd, interrupted on Friday by the Good Friday markets holiday in the U.S., saw stocks surge, with the Nasdaq Composite Index up 4%, the Standard & Poor’s 500 up 3%, and the TL20 group of stocks to consider up 6%, with most names seeing net gains for the week. 

The headlines were dominated by continued debates over the memory-chip trade concerning Micron Technology, SanDisk, and Seagate, by new worries for ASML and chip-equipment makers with respect to China, and by the announcement of an unprecedented $122 billion private round of funding for OpenAI.

The memory-chip debate continues as the Street wonders how long extraordinarily high profit margin can continue for Micron from surging DRAM and NAND prices. I covered this a bit last week, the worries over “spot pricing.”

Oracle’s bizarre transformation from a software titan to an industrial company gained a new component on Monday as the company announced it has appointed Hilary Maxson, CFO of industrial giant Schneider Electric for six years, as its new CFO, replacing current CFO Doug Kehring immediately. 

Oracle stock is down fractionally at $145.57 on Monday, but the analysts covering the company seem pretty pleased. They have to reckon, however, with what I think a proper stock valuation is for Oracle, which I would argue is much less than the current valuation.

In the press release announcing Maxson, Oracle lauds her role at Schneider transforming the company “from an electrical equipment supplier into a digital energy technology partner for key segments, like utilities and datacenters.” Prior to Schneider, Maxson was an executive for a dozen years at American utility AES. (Kehring is staying on at Oracle as head of operations.)

The week that ended March 27th saw the TL20 group of stocks to consider lose all of its gains, and last week, the abbreviated week ending with the Good Friday holiday, saw all that come rocketing back.

The TL20 is up now six percent for the year, vastly outperforming the major indices, which remain in negative territory. Most names rose last week, as you can see in the table below.

The artificial intelligence trade has been dominated by wild amounts of money being spoken of, including more than a trillion dollars that OpenAI has committed to spending on Nvidia GPU chips and other infrastructure. I believe all that money talk is about to come to its denouement, and not in a good way.

OpenAI on Wednesday announced its latest investment round, $122 billion in funding lead by returning investors Nvidia, Amazon, and SoftBank Group, along with a gaggle of other investors including the venture firm a16z (Marc Andreessen and team.)

The round values OpenAI at $852 billion, the company said. That allows us to put a price on all the spending the company has pledged and what it may amount to. OpenAI noted that its revenue is coming in at $2 billion a month, implying revenue this year will reach or exceed $24 billion.

One of the more unusual arguments in favor of Microsoft stock that I’ve seen in a while is contained in the report Wednesday by Yi Fu Lee of The Benchmark Company, who initiated coverage on the shares with a Buy rating and a $450 price target, which would be a 26% gain from the closing price of 369.37. (The shares sold off slightly on Wednesday.)

Lee values Microsoft at 8.8 times fiscal 2027’s likely revenue, a modest multiple for a powerful software outfit.

Lee’s argument is that investors have ignored the value of all the data of Microsoft customers contained in the many applications that the company sells, including the Office 365 productivity suite, the Dynamics back-office software, and LinkedIn.

There continues to be in the popular imagination a seriously misleading presumption about quantum computing, which is that we will build quantum computers very soon that are hugely powerful. 

As I’ve written at length, quantum has a “scaling” challenge: no lab or company has demonstrated that they can reliably build computers with the hundreds of thousands of quantum “gates” necessary to do real work. 

Because the public keeps missing that obvious fact, we’re getting doomsday articles such as the piece in Barron’s this week by Mackenzie Tatananni that claims that Google has just told the world that conventional data encryption will be cracked by 2029, years earlier than previously expected. 

Shares of chip maker Marvell Technology were soaring thirteen percent on Tuesday to $99.19 as it became the latest beneficiary of Nvidia’s largesse, the two announcing they will work on “silicon photonics” together — the integration of data communications into the processor itself — and that Nvidia will take a $2 billion stake in Marvell (three percent of Marvell’s $77 billion market cap.)

The deal follows investments in Intel last year — almost five percent of the stock — and Coreweave, of which it owns almost thirteen percent, according to 13-F filings compiled by FactSet Systems.

Nvidia’s public equity spending soared in 2025 from almost nothing in all the years prior. According to the 10-K filing for the year, out of a total of $63 billion of cash and equivalents on the books at the end of the January quarter, Nvidia’s investments in marketable securities, which totaled $52 billion, included $12.9 billion in equities (the rest being money markets and debt). That is up from just $381 million a year earlier.

Shares of conglomerate Innventure are up ten percent Tuesday morning at $3.90, after CEO Bill Haskell told analysts on the company’s earnings call Monday evening that his stock is priced too low relative to its assets.

Innventure is majority owner of Accelsius, a startup making “two-phase” liquid-cooling technology for data centers. That company is valued at $665 million, based on outside investments by industrial giant Johnson Controls and French building infrastructure giant Legrand. Innventure stock is currently trading for just $296 million, based on 82 million shares outstanding.

“We're not gonna complain about the market,” said Haskell, “but we are going to state facts. The facts suggest there is a significant gap between where our shares trade and what this platform is worth.”

Shares of conglomerate Innventure are up ten percent Tuesday morning at $3.90, after CEO Bill Haskell told analysts on the company’s earnings call Monday evening that his stock is priced too low relative to its assets.

Innventure is majority owner of Accelsius, a startup making “two-phase” liquid-cooling technology for data centers. That company is valued at $665 million, based on outside investments by industrial giant Johnson Controls and French building infrastructure giant Legrand. Innventure stock is currently trading for just $296 million, based on 82 million shares outstanding.

“We're not gonna complain about the market,” said Haskell, “but we are going to state facts. The facts suggest there is a significant gap between where our shares trade and what this platform is worth.”

The week that ended March 27th brought the biggest challenge to the group of the TL20 group of stocks to consider since the beginning of the year, pushing the TL20 down 4%, erasing all gains for the year. 

At least the TL20 is still in better shape than the 10% decline of the Nasdaq since the start of the year, and the 7% decline of the Standard & Poor’s 500. 

My post about the TL20 Sunday goes through the names with some comments about how they’re doing. ‍ ‍

Not all are negative. Amphenol and Analog Devices are holding up relatively well, with little news flow since their last earnings reports. 

The TL20 group of stocks to consider on Friday hit its lowest level since December 31st, elegantly erasing the year’s gains. The stock suffered worse than the Nasdaq Composite Index and the S&P 500 this week, with the 5.5% drop on Thursday being the group’s fifteenth-worst single-day decline. A summary of where they’re at.

Analysts at Jefferies & Co. on. Wednesday published one of the strangest reports I’ve read in a long time, which laid out the case that we are at an “inflection point” for humanoid robots, but actually managed to convince me we are nowhere close to such an event.

The title is cute: “From Aasimov to the Assembly Line, Humanoid Robots Begin to Clock In.”

Except that, they aren’t really. “Atlas is not expected to approach a meaningful cumulative volume of 30,000 units until close to 2030 with Hyundai,” the report notes of Boston Robotics’s flagship unit, Atlas. Tesla’s “Optimus rollout is likely to be dominated by internal use in its factories, given constraints such as harmonic reducer torque density and ongoing mechanical redesign,” they note.

The bulk of the report, which was put together by a whole team of Jefferies analysts, including semi analyst Blayne Curtis and colleagues from all different areas of equities, is filled with rich technical details that are thoughtful.

However, the top-level perspective that has been thrown onto the thing strains to convince you of some imminent financial development that will be useful to a trader or possibly an investor. It is tremendous “organ music,” as in the opening fugue of a great drama.

“Humanity appears to be on the precipice of wider adoption of humanoid automatons,” it boldly states. “Humanoid robots—machines structurally resembling humans and built to operate in environments designed for people—are transitioning from longstanding aspiration to commercial reality.”

Except that the 46-page piece itself, though, is like the fine-print disclaimer about what doesn’t yet work regarding human-like robots.

Morgan Stanley chip analyst Joseph Moore engages in a spirited defense of memory-chip stocks Micron Technology and SanDisk on Thursday, after a sharp drop in their shares this week.

We are not in normal times as his main message, as things that would mark a normal “cyclical peak” in their businesses are moot with the rise of “agentic AI” and the need for much, much more memory.

“Looking for sell signals from prior cycles misses the point,” writes Moore. “Memory is not just constrained by AI demand — memory is increasingly THE primary constraint on AI demand.”

Moore has an Overweight rating on Micron shares, and a $520 price target — almost 50% upside from a recent $357 — and also Overweight on SanDisk stock with a $690 price target, about 13% from today’s $611.

I think Moore is right on the money on all points. Both Micron and SanDisk are in the TL20 group of stocks to consider, and I see good times continuing for both, as I wrote at the beginning of the year.  

Both stocks are down sharply today despite his endorsement.

Moore outlines his sense of the factors weighing on shares:

Shares of ARM Holdings are up nineteen percent on Wednesday at $159.99 after the company said Tuesday that it will go into competition for the first time ever with its customers, making its own microprocessor for the data centers, the “AGI CPU.”

At an event at the Fort Mason Center in San Francisco, ARM CEO Rene Haas talked with analysts about the virtues of the AGI CPU to relieve the exhausted x86 CPUs of Intel and AMD. (Interestingly, both AMD and Intel shares surged 7% on the news — we’ll come back to that.)

The question now becomes, having earned the most expensive stock multiple of all semiconductor stocks based on a business of one-hundred-percent profit, or nearly, should ARM stock still trade for twenty-four times next year’s sales as it transitions to a business with fifty-percent profit margin?

Probably not, but let’s go through the numbers.

Merrill Lynch software analyst Tal Liani on Tuesday initiated coverage of CoreWeave, the debt-laden operator of artificial intelligence clouds that is under contract to OpenAI and Microsoft and Meta. It was one of the more interesting initiations in a while because it is positive, with a Buy rating, and a $100 target, but behind the lines, the subtext is quite negative to my mind.

Liani doesn’t write it, but I think the conclusion is that at some point, CoreWeave becomes just a software vendor, shedding the money-losing data center operations.

Liani’s critique boils down to three questions that he calls the “central debates” of investors: 1) Will the company lose its edge as Microsoft and the others build their own AI data centers, and as the game shifts from AI “training,” the making of new AI models, to “inference,” the making of predictions for the masses; 2) are they going to be able to transition to a broader market after having worked for a captive handful of customers for years; and 3) will they be able to keep financing operations that depend on quickly rushing equipment into service in order to pay their enormous debt service.

Deutsche Bank semiconductor analyst Ross Seymore this morning offered up a “chart book” of where semiconductor valuations sit. He notes that the chip equipment companies, including KLA and Lam Research, are the most expensive group: their forward price-to-earnings multiple is well above what the median had been for their stocks.

By contrast, Nvidia and Advanced Micro Devices are well below their median. Seymore sees the disparity between processor companies like Nvidia and equipment companies like Lam to be “puzzling dislocations in relative valuation.”

Seymore offers a handy chart that shows which stocks are expensive and which are cheap based on where their P/E is now relative to the historical median:

I wrote a week ago or so that software stocks are trading below their average price-to-sales multiple and could be acquisition bait.

That has not prevented software from continuing to trade lower in price and lower in reputation.

The latest casualties on Monday are nine software stocks that Jason Ader of William Blair cut from either Buy or Hold to either hold or Sell after concluding that, as he puts it, “AI has injected unprecedented uncertainty into the software sector — every company will need to rethink their product, their pricing, and their GTM strategy.”

I’ve laid out the nine for you in a handy table:

It might seem pointless to be worrying about stock allocations with a war in the Middle East, but my philosophy tends to be in good times and bad, one needs to focus on the quality of stock investment — if you're going to be putting money into the market at all. 

Last week, for the seventh rebalancing of the TL20group of stocks to consider, I swapped out five names and swapped in five new ones.

Based on a reference date of March 13th, the TL20 is down a third of a percent in the past week, holding up better than a two-percent fall in the broader indices. 

Year to date, the TL20 is up 5%, better than down 7% for the Nasdaq Composite, and down 5% for the S&P 500.

Since inception, the TL20 is up 309%, more than triple the Nasdaq and more than quadruple the S&P. 

It has also done much better than benchmarks such as Cathie Wood’s ARK Innovation ETF, the ARKK, up 57%; the IGV, the iShares Tech-Software ETF, up 52%; and the Philadelphia Semiconductor Index, up 185%.

Even though the stock market was fairly terrible this past week, they were partying in Los Angeles.

That was where fiber-optic vendors such as Lumentum and Coherent were showing off their wares on the show floor of the OFC conference, one of the most closely watched trade shows for the fiber-optic industry. The outlook was extremely positive. The show was sold out, with over 16,000 attendees.

The big theme at the show, as it has been for years now, is fiber-optic lines replacing copper cabling in racks of computers in the data center. As speeds increase, and distances grow longer between computers, copper gives out and you need optics instead. 

That has always been true, but there is an added sense of momentum in the shift to optics because of Nvidia’s relentless takeover of the artificial intelligence data center.

Nvidia’s vision, as articulated by CEO Jensen Huang the day before at its own GTC conference in San Jose, is that multiple racks of computers will work together as one giant virtual supercomputer. In order for that to happen, the connections between racks of computers need to be bolstered with greater and greater bandwidth so that a chip in Rack A, say, will communicate directly with a chip in Rack Z, and all the racks in between will communicate, all at top speed.

I rebalanced the TL20 group of stocks to consider this week, with my usual laudatory words for DRAM and NAND chip maker  Micron Technology. 

Over a decade ago, I wrote for Barron’s that Micron could become one of the most valuable chip makers in the world as a result of the semiconductor “renaissance,” and Wednesday evening’s quarterly numbers from Micron affirmed my sense that Micron is becoming much more valuable. 

But, the stock is down three percent today at $447.02, even as stock price targets zoom to as high as $700 at some shops. This is similar to what we saw Wednesday with Nvidia stock selling off despite very positive commentary by Nvidia management on  Tuesday during a meeting with analysts. 

First, the numbers, which are, to my mind, more incredible even than those of Nvidia.

Analysts on Wednesday were mulling everything they heard from Nvidia CEO Jensen Huang and CFO Colette Kress on Tuesday during the session held for financial topics at the company’s GTC conference in San Jose, California.

The biggest technical point of Huang’s remarks was that the OpenClaw software that came out in January, which lets you run software agents (and can cause chaos), is the new operating system software for our age. It’s just like Windows or Mac for agents, he implied, because it manages resources and runs scheduling tasks, etc.

“Every single software company, every single company needs to have an OpenClaw strategy,” said Huang. “This is a very big deal.” Huang plans to help by having revealed this week Nvidia’s own version of the open-source software, which Huang calls “NemoClaw.” Oh, and, selling a lot of chips.

Whatever you think about speculative excess in AI, the fact is that we are at an historic moment in the history of computer chips, a semiconductor renaissance that has been gathering steam for years. That very profound change is sending ripples through the tech market. Nvidia is a big part of the story. Here are other ways an investor can capture the seismic shift.

In another era, Adobe’s financial report Thursday night would have been admirable. The company increased revenue faster than people expected, twelve percent, which is no small feat at $6.4 billion in revenue.

But CEO Shantanu Narayen is stepping down after eighteen years, and it’s not a victory lap. The threat of artificial intelligence has bedeviled the stock for a long time. The shares are down 38% in the past twelve months, and the multi-year, annualized return is mediocre. Since Narayen took the top spot in December of 2007, the shares are up 540%, which is well below the Nasdaq’s 900% in that time.

The search is on for his successor.

Asked what qualities his successor might have, Narayen Thursday evening offered some vague stuff about finding someone who will take advantage of the “opportunity that AI has.” He didn’t seem to have any real thoughts, or he kept them to himself.

I have some thoughts.

Here’s a fun game you can play at home: Which aging software giant will try to make up for a lack of innovation by acquiring a younger firm currently in the bargain bin? Which private equity shop will milk the cash flows of some struggling outfit under the pretense of “making it great again”? If you want to know what happens to software stocks, consider the prices at which firms have been taken out, including IBM’s $10 billion deal for Confluent and Salesforce’s $8 billion deal for Informatica.

“Thirty to forty percent of our code now is being generated with Claude Code.”

Oracle reported results after market close on Tuesday, and the stock jumped 9% in after-hours trading, and is up over 12% Wednesday morning, at $167.47.

The favorable response is mainly in response to the company’s impressive cloud computing sales growth of 44%.

However, nothing about the report changes my feeling this has become a bad proposition.

The company has gone massively into debt, they’re diluting shareholders, they’re eroding their margins, and, worst of all, we don’t know just how much the company is exposed to one giant customer, OpenAI.

It’s funny what is trending Tuesday morning: big cuts in price targets for software stocks even as tech companies are doubling-down on the artificial intelligence-generated code that is causing havoc.

ServiceNow, Salesforce, Hubspot and Zscaler are among names seeing price targets slashed at various brokers Tuesday morning. These price target cuts are mostly still accompanied by Buy ratings or the equivalent.

The argument is not that the companies are in trouble. Rather, the analysts are trying to find what the bottom is in stock valuations, so they are proactively cutting the valuation multiples they put on these stocks.