As Epistrophy Capital Research grows, the newsletter takes a different look starting this week (thanks to the great Scott Galloway for the direction). This weekly missive is still focused on helping TV bookers track our work at Epistrophy, I’m hoping the few of you on this list get some story ideas, actionable insight perhaps you’ll even enjoy it. If you know of any other journalists that might find this useful, feel free to make an introduction!

This week earnings come raring back! We have sixteen freaking companies reporting in over a four day stretch, so there’s work to done. See the list of all the events we’re following at the end of this note (and look, a clickable table of contents!)a

As always, I’m focused on three things:
1) Technology-driven change;
2) the latest in innovation and startup trends, and;
3) stock fraud.

Companies Discussed

In This Note:

• Companies Discussed

• Stargate: So Little, And Yet So Much

  • AI's Black Hole Budget

  • Larry Ellison: Art of the Deal

  • Trump Goes Nuclear, This Time With Tiny Reactors

  • Full House at the Digital Inn

• TI Finds A Bottom, Keeps On Digging

• Tweet O’The Week: Netflix

• Epistrophy In The News

• Upcoming Events We’ll Be Following

• Availability This Week

Stargate: So Little, And Yet So Much

AI's Black Hole Budget

When OpenAI's Sam Altman, Oracle's (ORCL:NYSE) Larry Ellison, and SoftBank's Masayoshi Son joined President Donald Trump at the White House to unveil "Stargate" —a grandiose plan to pour $500 billion into U.S. artificial intelligence infrastructure—it sounded like science fiction for a reason. Nonetheless, it was a resounding declaration of confidence in America's technological future, the promise of 100,000 new jobs, and an effusive endorsement of Trump's leadership in the dawning age of AI on his first full day back on the job. It also might show that Ellison, not Trump, mastered the art of this deal.

But first: the numbers behind Stargate quickly crumble under scrutiny. For the triumvirate of SoftBank, Oracle, and OpenAI to reach the vaunted $500 billion investment mark, they would need to invest an average of over $40 billion per year, each, for the next four years. To put that figure into perspective, it dwarfs the combined annual R&D budgets of tech titans Apple (AAPL:NASDAQ), Alphabet (GOOGL:NASDAQ), and Microsoft (MSFT:NASDAQ).

SoftBank, the Japanese conglomerate leading the investment charge, claims it will deploy $100 billion immediately and scale up to $500 billion within just four years. Did any of the reporters in the room even ask if Softbank has the money? It doesn’t. It’s much-touted money losing Vision Fund 2 has only secured $60.8 billion in total commitments globally. The notion that SoftBank could somehow conjure an additional $40 billion in annual funding — specifically for Stargate — strains credulity.

And I love me some Oracle (indeed, I own a chunk of ORCL shares). But Oracle's total tangible assets currently sit at $56 billion. Oracle just posted a record year of capital expenditures at $11 billion. As impressive as that figure might be a 4x leap for part of one moonshot project seems unlikely.

And then there's the white-hot AI darling of the moment, OpenAI. The ChatGPT creator has attracted a staggering amount of funding in recent years—$17.9 billion in total, including a record-breaking $6.6 billion round just recently. But even as the cash pours in, reports OpenAI is burning through a staggering $5 billion per year. Another “partner” unable to pony up a new $40 billion or even $10 billion-a-year.

The White House said the first Stargate construction site in Abilene, Texas, and with that the project's initial scope became clearer—and far less impressive than the "immediate $100 billion deployment" trumpeted at the White House. Ten data center buildings are currently rising from the Texas dirt, each spanning 500,000 square feet. And Stargate is a new name, but Oracle has been working on this since at least mid-2024 (see our Dec. 8, 2024 research report “Oracle’s AI Flex”.) While undoubtedly a substantial undertaking, rough calculations based on comparable projects suggest a total investment closer to $3-5 billion, not the $100 billion claimed.  

The big bogey here is MGX. It’s the most enigmatic player in the Stargate saga. MGX launched early last year, with the United Arab Emirates sovereign wealth fund Mubadala Investment Co. and AI firm G42 (a firm that recently shed its China ties to appease the Biden Administration) as founding partners. Ahmed Yahia Al Idrissi, Mubadala’s direct investments platform CEO is also CEO of MGX. It was established with the goal of deploying over $100 billion into artificial intelligence and semiconductor projects, MGX has positioned itself as a key backer of Stargate – but part of $100 billion doesn’t get to $500 billion. Details about MGX's exact financial commitment to the venture remain murky. But with the Trump Administration’s blessing these data centers will surely have an easier time raising money than it would have a week ago.

Larry Ellison: Art of the Deal

To understand how this likely plays out, think of what the players in this scenario do. Softbank is the money finder, Oracle is the builder, OpenAI is the user. Let’s chart it:

Larry Ellison seems to have orchestrated a fantastic deal for Oracle here. The most likely scenario that I see is that this will be some kind of new entity owned by Softbank and Oracle, with OpenAI as the anchor tenant. But no matter the outcome, Oracle suddenly emerges as the default standard for OpenAI as the ChatGPT maker, for the first time, moves away from it’s long time backer and partner Microsoft.

Importantly, Altman and Mustafa Suleyman, co-founder of DeepMind and current CEO of Inflection AI — now CEO of Microsoft’s AI division after acquiring Inflection for $650 M — — are at polar ends in the AI industry. “I think the big news is that, you know, OpenAI is moving off of Microsoft,” Salesforce CEO Marc Benioff said in little noticed comments to Bloomberg BusinessWeek editor Brad Stone at the World Economic Forum in Davos. "I think it's extremely important that OpenAI gets to other platforms because Microsoft is building their own AI...I don't think Microsoft will use OpenAI in the future."

Trump Goes Nuclear, This Time With Tiny Reactors

I don’t know if anyone else noticed this Trump aside — there are so many — but he said something that could signify a significant shift for U.S. energy and AI policy. At that Stargate press conference Trump suggested a new approach to energy access for AI infrastructure. "I'm going to help a lot through emergency declarations,” he said. “We'll make it possible for them to get that production done very easily, at their own plants if they want, where they will build, at the AI plant, energy generation." Is this a new Federal policy that would enable AI companies to build unprecedented off-grid power generation facilities?

Maybe you were focused elsewhere on Nov. 4, but there was big, bad news for AI. The Federal Energy Regulatory Commission (FERC) rejected an amended interconnection service agreement (ISA) that would’ve let an Amazon Web Services (AMZN:NASDAQ) connect directly to a Talon Energy (TLN:NYSE) nuclear plant. Talon’s Susquehanna’s 2,520 MW nuclear generating facility (located in Luzerne County, Pennsylvania, with two 1,260 MW units underneath two 50-story cooling towers), would directly connect to an Amazon data center. But FERC found that they plan would jack energy prices to consumers. “Energy prices would increase significantly as low cost nuclear energy is displaced by higher cost energy on the overall supply curve,” their ruling read. “Capacity prices would increase as the supply of capacity to the market is reduced. Emissions would also be expected to increase as thermal resources that are next in the supply curve are dispatched to meet load to replace the nuclear energy.”

The day before the press conference Trump declared a

Trumps statement suggests a FERC reversal, potentially AI infrastructure development through both nuclear and fossil fuel options (regular consumer energy prices be damned.)

Full House at the Digital Inn

On data centers: Counting the number of data centers is dumb — Oracle hs over 160 public and private data centers, but they range from single customer data centers at 25 kW to one 64,000 times larger — a planned 1.6 gW data center powered by three small nuclear reactors. So we keep a model of counts megawatts of “absorbed” capacity — data centers that are fully built, operational, and occupied by tenants. (We read press releases, but we don’t count them.)

Here’s the shocker: in the last year, data center absorption has dramatically slowed. Yes, AI is driving so much demand that there are hardly any vacancies to fill. The construction is slower than ever. It’s still subject to delays in terms of equipment, real estate acquisition, power infrastructure and regulatory approvals. These hurdles can slow down the actual absorption rate, even as AI demand remains high.

Large-scale data centers, especially those designed for AI workloads, take time to build (one of the reasons I own shares of heating, cooling and electrical installer Comfort Systems (FIX: NYSE)) Construction of these facilities is a multi-year process. Absorption is slowing to these delays, but the long-term trend remains upward. And for every megawatt of absorbed capacity, over 13 mW was pre-leased last year, according to real estate advisory Datacenterhawk.

Stargate will likely never reach $100 billion or $500 billion in spending. But it might help accelerate the revolutionary potential and critical importance of artificial intelligence. The U.S. needs to cement its leadership in this world-altering domain.

This Stargate announcement highlights what was already true and it greenlights what will be true — tens of billions were already going to be spent building data centers this year. And now that this is President Trump’s goal, not Sam Altman’s, this announcement could greenlight more data center construction.

In “The Social Network” the Sean Parker character, played by Justin Timberlake, says: “You know what's cooler than a million dollars? A billion dollars.” 

$500 billion? Even cooler – but not to be believed. 

*(For that all that is sacred in this world, do not waste 1 hour and 56 minutes watching the 1994 movie Stargate. It is time lost for eternity.)

TI Finds A Bottom, Keeps On Digging

Texas Instruments (TXN: NASDAQ) Q4 results provide a stark reminder: in the sprawling semiconductor sector, not all companies march to the same beat. As many chipmakers enjoy a booming demand for AI accelerators, TI finds itself grappling with a more challenging reality in its core analog and embedded processing markets.

To understand TI's predicament, let's walk through the financials. Q4 revenue declined 2% year-over-year to $4.01 billion, while net income tumbled 12% to $1.21 billion. Nearly 70% of TI's sales are tied to industrial and automotive markets. CEO Haviv Ilan candidly acknowledged that some of these crucial sectors "still haven't found the bottom." Exemplary honesty from a CEO, whose company is built to last through a long drawn out cycle like this.

And yet the challenges persist. Leverage works both ways. TI's Embedded Processing saw a 25% sales decline which caused operating margins to plummet from 30% to 9% in a year. Exacerbating this pain was the fixed cost of a new 300mm fabrication plant. While this state-of-the-art facility should improve efficiency over time, in the near term it acts as a drag on profitability when sales falter.

Another concerning development is the sharp rise in TI's inventories, which ballooned to $4.5 billion, or 252 days' worth of supply. Some extra stockpiles are warranted to avoid the kinds of shortages that have recently plagued the industry. But if demand doesn't bounce back, TI could be forced to write down the value of that inventory. It's a precarious tightrope to walk.

Despite these near-term headwinds, TI's management remains resolute in its pursuit of a bold capacity expansion plan. The company is betting that by investing heavily to secure ample manufacturing capacity at competitive costs, it can cement a long-term strategic advantage.

Looking ahead to Q1, the company expects revenue to dip another 3% sequentially. But perhaps even more jarring is the anticipated drop in gross margins. For years, TI has enjoyed margins north of 40%, thanks to its manufacturing prowess and high-value product mix. But a combination of lower sales, higher depreciation expenses from its new fabs, and reduced factory utilization could push margins down into the mid-30s. While likely temporary, such a precipitous decline would be psychologically bruising.

Yet even amid these challenges, there are reasons for optimism about TI's longer-term prospects. The company has a long history of successfully navigating the semiconductor industry's notorious boom-bust cycles. Its vast scale and diverse customer base provide some insulation against market volatility. And its commitment to innovation and manufacturing excellence has historically served it well. That said, the current downturn looks to be an especially difficult one, and TI will likely face several quarters of subdued performance.

Many of its analog peers, such as Analog Devices (ADI: NASDAQ) and Microchip Technology (MCHP: NASDAQ), are also contending with soft industrial demand. Even more diversified semiconductor giants like Broadcom (AVGO: NASDAQ) have seen pockets of weakness.

TI's Q4 results underscore the stark dichotomy currently unfolding in the semiconductor world. While some chipmakers are riding high on the AI wave, others like TI are navigating a more trying environment in their core markets. TI's long-term strategy and proven resilience offer some reassurance. But the road ahead looks challenging, and the company will likely have to weather several more quarters of lackluster performance before its end markets recover. In the meantime, it may find itself somewhat overshadowed by its more glamorous, AI-focused peers.

Tweet O’The Week: Netflix

Epistrophy In The News

Appear on Bloomberg TV to talk about Netflix? Right in my wheel house. And what a pivotal quarter for this company — they blew out numbers. I expected to come on right when the numbers crossed, but actually had time enough to update my model, really understand the great subscriber numbers and then… my voice died. On live TV! A lingering cough decided to graduate to laryngitis right while I was on the air (painful to watch). Ugh!

Paul Bagnell was gentlemanly enough not to mention it!e-close/2025/01/21/netflix-posts-blowout-fourth-quarter

Upcoming Events We’ll Be Following

Availability This Week

I’m available all week and would love to expand on the thoughts above, though I’ll be busy now that earnings reports are back (as are our Drill Down Earnings podcasts). Written reports are available to clients, with video summaries on YouTube and, of course our popular summaries of the summaries (yes, the concavity indicator) on Instagram and Tiktok (speaking of Tiktok, we’re still alive!).

I hope these notes are helpful to you. I’d love to discuss them further and, as always, comments, questions and ideas are appreciated.