Here’s a preview of what we’ll cover this week: 

Macro: Bank Earnings Highlights; The Consumer Is Still Spending; Don’t Trust Business Surveys; Capital Markets Took a Breath; AI Capex: The Bankers Have Seen the Order Book

Markets: Tech Took the Hit; The Flight to Quality; Anthropic’s $500 Billion Haircut; Compute Demand Isn’t Going Anywhere 

Lumida Curations: Jensen Huang on Why Compute Demand is Compounding; Elon Musk on Tesla’s Entry Into Chip Fabs; Mark Zuckerberg Says Meta Is Not an App Company

For months, we’ve been discussing how wealth management (and financial services more generally) will be ripe for agentic disruption.

Today, we are pleased to unveil the world’s first agentic wealth advisor.

You can meet her.

This is Elena Navarro. She’s the world’s first agentic wealth advisor built by Lumida.

She joins you live on a Zoom call.

You can talk to her and ask her questions the way you would with a human advisor, except she is available on weekends, after office hours, and she doesn’t miss a beat.

We’ve also rolled out Sara. She’s an Agentic Advisor that focuses on tax mitigation opportunities. Both are still in beta.

You can try Sara here at vdr.lumida.com

If you’re a VC, and want to meet Elena, click the Investor Sign-In button on the top right.

(We need to add concurrency to Elena, so we need to limit access for now).

Meet Sara now.

(Disclosure: Don’t rely on either for tax or investing advice, and usage of these agentic services does not establish a service relationship between Lumida and you. We are showing you a glimpse of the future.)

Well-designed AI fits like a glove.

Unlike static SaaS, it isn’t a tool you have to pick up and learn.

AI travels. It moves to wherever the customer already is, on their terms, in their channel of choice.

As a result, we’re going to see AI rugpull SaaS and services.

The last great business model was SaaS. For 20 years, it sold you a tool, and you did the work.

You logged in, clicked around, and reviewed dashboards. The software was the product. Your time was the cost.

AI flips that. It gives everyone access to expertise.

The question is no longer “do you know the answer?” It’s “did you tell me before I had to ask?”

What’s left to compete on is the experience.

That’s the next great business model: Experience as a Service.

It’s possible now because of attention.

Great service has always required human attention, and attention doesn’t scale. So every service industry rationed it.

The top 1% got a dedicated advisor. Everyone else got a portal login and a holiday gift basket.

AI removes that limit.

Attention becomes nearly free, and every customer can get what used to be reserved for the few.

Business models like Iconiq, which serves Mark Zuckerberg’s family office, will be opening up to the many.

VCs need to rethink the concept of ‘Product Market Fit’ in the age of AI. The new model is ‘Experience Market Fit’.

The product is no longer the software.

It’s the SMS that tells you what to do before you know to ask. It’s the email with your K-1s that arrives before you ask for it.

It’s the app that shows where you stand right now, not once a year in a year-end summary.

You don’t buy software. You buy the feeling that someone competent is on it, all the time.

Take the CPA.

Everyone hates their CPA.

Nobody hates their CPA’s work.

The return gets filed, and the numbers are (usually) right.

What people hate is everything around it: the email that disappears for three weeks, being told “we’ll look at that after tax season,” and learning in October about a move you needed to make last December.

The CPA’s product was never the problem. The experience was.

Now picture the Experience-as-a-Service version:

“Your Q3 estimate is due Friday. We’ve gathered the K-1s and 1099s. Here’s the amount to set aside.”

“You sold the rental. Here’s what that means, and two moves to make before Dec 31 to put those losses to work.”

The legacy CPA firm gets squeezed from both sides. Software handles the preparation, and experience-first firms take the client.

That squeeze is also where the labor displacement shows up.

Chasing documents, sending reminders, and answering “where do I stand?” used to take a floor of associates. That work is exactly what the agent now does.

The same pattern is coming for wealth management, legal, insurance, healthcare admin, and any business where the work was fine but the experience was the complaint.

The winners won’t have a better product. They’ll have a better experience.

For founders and investors, that’s the shift. Stop building the product. Start building the experience.

You can experience it firsthand by talking to Elena.

I am in Las Vegas this week, attending the Money20/20 conference (18th-21st Oct). 

Exciting event to meet with entrepreneurs and VCs, and see what’s new in the AI world. I have been attending it for about a decade now.

If you’re there and would like to meet, schedule 30 minutes here.

(I can take meetings Monday morning after 5pm PST.)

I will then be in SF and Palo Alto from Wed thru Friday afternoon.

We’re meeting with the world’s leading VCs, and it would be great to connect.

This Wednesday, we hosted a private Lumida investor evening in New York for family offices, founders and investors.

Thank you to everyone who joined us. We’re glad you could make it.

I sat down with Dan Madden, Principal at Squadra Ventures.

Dan is a Marine veteran who served in the Pentagon and on the House Armed Services Committee before moving into defense venture capital.

The feedback was great. 

The full conversation premieres on our YouTube channel on October 20. Watch it here, and set a reminder so you catch it when it goes live.

We have also been posting some curations from the event on our X page; follow it here to watch the best bits. 

We have also curated a defense tech primer for investors interested in defense tech. You can download it here. It discusses how AI is rebuilding defense, ten private companies at the forefront of the revolution, and the upside and the risks for investors.

I also joined Sebastian Purcell on Real Vision’s The AI Trade alongside Max Gokhman, Head of AI and Digital Assets at Franklin Templeton.

We also talked about the risks building in private credit and data-center financing, and weighed them against the enormous capital still pouring into AI infrastructure.

It is well worth a watch if you’re positioning around the AI cycle. Watch on Real Vision’s website.

Nvidia’s Circular Financing

I explain why this is an overblown issue and what people are getting wrong here.

Earnings season started this week, with big banks rolling out their results.

Bank earnings are my cheat code for understanding the health of the economy.

Banks track every paycheck, card swipe, loan draw and repayment. 

Then, they get on a call and spend an hour answering the toughest analysts on the street.

And, unlike government data, the numbers they report can’t be revised. 

Last week, we said the banks would have strong earnings.

This will be true next quarter and the quarter after that.

Goldman posted $18.7 billion in revenue, its second-highest ever. Citi delivered its best third-quarter revenue in a decade. JPMorgan earned $15.3 billion with a 20% return on capital.

Many of these names have sold off as rates moved higher despite earnings going to record levels. That creates an opportunity as rates come back down.

Here’s a chart of the 10-year showing rates coming back down…

Start with the numbers.

At JPMorgan, debit and credit card spend grew 10% y/y.

At Wells Fargo, debit spend rose 8%, accelerating from 5% a year ago.

That’s cash consumers already have on hand.

Credit card spend rose 12%. 

(That’s very brisk spending…too brisk, and why the Fed is raising rates and should never have cut last year.)

Jeremy Barnum (CFO, JPMorgan): “The consumer is currently very healthy, primarily because of a healthy labor market.”

The more interesting news came from the bottom-end.

All year, the bear case has been that the K-shaped economy cracks from the lower end. 

Brian Moynihan (Chair and CEO, BofA) is watching the direct deposits hitting his customers’ accounts, and he says paychecks are growing in every income band, and “we see the low having picked up.”

That’s a non-consensus insight. The bottom half of the K-shaped consumer is doing OK too. (I’m not diminishing genuine hardships and real issues out there - I’m simply saying credit is not falling apart, and consumers can get jobs.)

On credit…

BofA’s card loss rate improved to 3.22%, and delinquencies have fallen year-on-year for six straight quarters.

Wells Fargo said early-stage delinquencies are at historically low levels.

Moynihan (BofA): “in terms of seeing any stress on the consumer, absolutely not in terms of credit quality.” 

The bankers, however, also flagged what can break the consumer.

In other words, the consumer only weakens if unemployment rises.

That hasn’t happened.

Moynihan noted that initial claims over the last four weeks “averaged 190 something, which is lower nominally than prior to the pandemic.”

Productivity is why workers keep their jobs. Jobs are why consumers keep borrowing and spending.

Don’t Trust Business Surveys 

Business sentiment has been gloomy all year. Business behavior is not gloomy.

Every CEO thinks the recession is coming for the other guy.

The important sentiment signal is not what they are saying, it is what these businesses are doing.

At BofA, commercial loans rose 8% y/y.

Moynihan said they “grew across every single segment from small business to business banking, to global commercial banking to global corporate investment banking.” 

At Wells Fargo, average loans and deposits each grew 11%.

Companies are borrowing because they are investing.

They are also borrowing responsibly. Mike Santomassimo (CFO, Wells Fargo) said his commercial clients “have been pretty prudent now for a while for years in terms of how much leverage they’re putting on.”

Barnum acknowledged that input costs and “the diesel narrative” are squeezing middle-market companies, and that it creates some caution. But balance sheets are strong enough to absorb it.

The risk to watch is the refinancing calendar.

Jane Fraser (Chair and CEO, Citi) flagged that “you’ve got an upcoming 3-year maturity wall. And that is going to occur in a much different rate spread and liquidity environment.”

Companies that borrowed at 2021 rates will refinance at 2026 rates, and some will feel it.

The banks are already getting ahead of that.

Dimon is pulling back from the loosest corners of credit, with “a little too much PIK, a little too weak covenants,” and said, “We just don’t feel you have to chase that kind of stuff anymore.”

Solomon (Chairman and CEO, Goldman Sachs) opened his call with this: “The U.S. data center build-out could require north of $10 trillion in investment through 2032, averaging roughly 3.6% of U.S. GDP annually.

In our conversations with clients, demand for compute significantly exceeds supply at the moment.”

Take a step back. What does that mean for semiconductor demand thru 2032?

Fraser (Chair and CEO, Citi) was more direct about whether higher rates will slow the spending: “I don’t think that higher rates are going to trump the imperative to invest there.”

We agree. Hyperscalers are not rate sensitive buyers. They can finance thru cashflow and return on invested capital.

Her reason: “There’s simply not enough compute. Energy remains a bottleneck, requiring more investment.”

Is anyone worried?

Yes.

Scharf (CEO, Wells Fargo) said the AI cycle “increasingly resembles prior infrastructure booms, creating both significant opportunities but also risks given the significant leverage used to finance the build-out.”

But, Scharf’s worry got an answer in GS’s earnings call.

Denis Coleman (CFO, Goldman Sachs): “this cycle is different from prior cycles in that the largest, biggest companies driving the cycle have significant capacity to use free cash flow to finance this.”

The 1999 fiber boom was built on debt by companies with little or no cash flow. This buildout is being funded mostly from the cash flows of the most profitable companies in the world.

GS’s Solomon pointed to exactly where: “There’s a big difference from something that’s really generated to serve a hyperscaler versus some of the neo clouds.”

That’s the right question for credit investors. A data center leased to Microsoft and one leased to a neocloud are two very different bets.

So how should you read all this?

Every banker on these calls is stress-testing for an AI slowdown. None of them is seeing one. 

Scharf’s conclusion is the one that counts: the positive forces from these investments “have thus far outweighed the potential downside.”

The time to worry is when bankers stop worrying. They haven’t yet.

So, the takeaway: 

Spending up. Lending up. Card losses down.

The economy is hot, hot and hot. (Inflation is a real issue as we noted back here.)

Consumer confidence is low - because of midterm election anxiety, gas prices, geopolitics, and rates. 

These are all transient issues. 

After midterms, historically what we see is a broadening out in the year ahead rather than concentrated rally (which is most of this year). 

I also did an FSD talking about bank earnings. Watch here for that.

Tech had its worst week in months. The Nasdaq 100 fell 5.5%, while the S&P 500 slipped about 2%.

The trigger was a new Gartner report.

It estimates AI data center operators will need $8.5 trillion in revenue through 2030 to cover their costs and earn a healthy return. Projected revenue over the same period is just $3.6 trillion.

That gap spooked the credit market first.

CoreWeave’s bonds fell below 80 cents on the dollar, and the debt financing of major data center projects sold off with it.

Then, the fear spread to tech stocks, with XLK declining ~1% on the news.

Investors went looking for safety, and they found it in Treasuries, which topped this week. 

We think the Gartner math speaks more to Anthropic and OpenAI hitting revenue growth — not the underlying demand for compute. That’s the distinction.

Should Anthropic and OpenAI fail to achieve the revenue required to fund their $1 Tn obligations, others will step in and own the compute (hyperscalers, governments, etc.).

That’s our lens.

(Gartner has been losing relevance and they are looking for a moment of fear to get some attention maybe?)

Rental prices for Nvidia’s B200 hit an all-time high last month, up 68% from a year ago.

Chips usually get cheaper to rent as they age and newer generations arrive. These are getting more expensive.

Even older generations of Nvidia chips are still in use and still earning.

Dylan Patel explained why in this curation. Token prices are falling more slowly than the number of tokens each chip can produce is rising. So a GPU generates more revenue as models get more efficient, not less.

Lenders see it the same way. Apollo is leading $40 billion of financing for SpaceX’s Nvidia chips, with the GPUs themselves as collateral. Nobody lends against an asset they expect to be worthless in three years.

Gartner’s report also has a timing problem. It compares costs through 2030 with revenue through 2030. But a buildout’s spending happens up front, while its revenue comes over the life of the asset. If you judge a power plant on its first four years of revenue, you’d never build one.

And revenue forecasts for new technology are almost always too low.

At the start of this year, analysts expected the hyperscalers to spend $630 billion in 2027. Goldman now expects $1.1 trillion.

The sell-off, however, had a silver lining. It made Nasdaq cheaper.

Look at the chart below.

The Nasdaq 100 now trades at about 21x forward earnings, its lowest multiple since the AI cycle began in 2023.

Each time it has fallen to this level over the last three years, it has bounced.

The timing helps, too.

Earnings season gives the “E” a chance to do some work, and corporate buyback windows reopen around October 26. (Oct 26th is also when markets rallied after the 2023 interest rate scare.)

History is on the side of the bounce.

Bespoke studied every day since 1995 when the Nasdaq 100, semiconductors and software all fell more than 2% in the same session.

Over the following week, the Nasdaq 100 averaged nearly 3x its normal weekly return, and semis more than 3x. Days like this one have usually brought a quick rebound in the following week.

The flight to quality wasn’t just happening in bonds this week. It has been happening inside the stock market since August.

Goldman’s Quality index is up 20% over the last three months, hitting a 12-month high with very little volatility along the way.

Investors are no longer paying for growth stocks at any price. They’re paying for the quality of earnings.

The longer chart shows how big this shift could be.

Goldman measures the performance of a composite that goes long quality and short momentum. It fell almost 60% from its 2023 peak as the market chased story stocks. It’s up about 55% this year and has only recovered part of that decline.

Why now?

Earnings growth is expected to slow over the next 12 months, in both AI and non-AI companies. 

Slowing from 30% still leaves very strong growth. 

But when growth peaks, investors start paying up for consistency.

Look at the following chart. The blue line measures percentage of earnings revision, while the white line is return on quality stocks.

Quality has tended to outperform exactly when earnings revisions roll over.

This is why we think this rotation has further to run.

Growth had a crazy run, and things are cooling off. We expect value to beat growth from here.

However, semis and memory are the exception, because their growth and quality overlap. 

Nvidia trades cheaper than the S&P, and Micron trades under 7x forward earnings with most of next year’s supply already committed.

We think both do very well over the next year.

Quality, however, doesn’t mean paying any price.

Look at Apple in its danger territory.

I feel like Bernie Sanders. Once again, Apple is at an insane valuation.

Apple trades at 37.2x forward earnings, above its 2020 peak of 36.6x. That’s the highest level since 2017.

Apple is one of the barometers I watch, because it tells you how much investors will pay to feel safe. Right now, the answer is a record amount.

We’ve seen this before.

After the 2020 peak, the multiple spent two years compressing toward the mid-20s. When Apple got close to 35x again in late 2024, it was back in the mid-20s by spring.

When Apple reaches these levels, it tends to give some back. At best, it treads water.

We think the rest of the Mag 7 could lag from here, too. A handful of megacaps carried the index for most of this year. 

The next leg should be broader, led by the average stock catching up and by the rate-sensitive names that are only starting to recover.

Earnings are strong, sentiment has reset, and the Nasdaq is trading at a multiple where it has bounced before. There are plenty of opportunities left.

You just won’t find the best ones in the stocks everyone already owns.

Anthropic gave investors their first look at its Q3 numbers this week, and the market didn’t love what it saw.

Anthropic’s third-quarter revenue rose 45% from the prior quarter to $16.8 billion. A 45% sequential increase sounds great, until you remember Q2.

Revenue more than doubled that quarter, to $11.6 billion. In one quarter, Anthropic’s growth rate went from nearly 150% to 45%.

The run rate tells the same story. Reports suggest Anthropic expects to reach a $110 billion annual revenue run rate by year-end. 

The current figure is at about $74 billion, up from roughly $65 billion in late July. Funny how markets criticize OpenAI, and they are at almost the same revenue run-rate.

Reuters is now saying Anthropic’s IPO valuation expectations have slipped to about $1.5 trillion, down from $2 trillion.

Why is Anthropic’s growth slowing?

Two reasons:

First, the easy market is getting saturated. Claude Code drove the Q2 surge, and by now nearly every serious developer already uses an AI coding tool. Growth has to come from somewhere new, and Anthropic doesn’t have the new yet. 

Second, and the bigger reason, is competition. Anthropic is losing share to cheaper Chinese models.

Beijing has chosen a side: cheap, open-source models over expensive closed ones. Think of it as the BYD playbook applied to intelligence.

China doesn’t need the best model. It needs a good-enough model it can manufacture, deliver and deploy at a lower cost than anyone else.

China isn’t trying to brew the best coffee. It’s trying to sell the most pods. And, it is driving customers away from the Black Ivory Coffee Dario brews.

Energy = Compute

Here’s another way to look at Anthropic’s numbers.

Goldman estimates that one gigawatt of AI compute needs about $11.6 billion in revenue a year to earn a 15% return. That covers everything: chips, buildings, networking and power. 

Anthropic ran about 1.4 gigawatts at the end of 2025 and plans to reach about 5 gigawatts by the end of 2026. That puts it at roughly 3.5 gigawatts on average last quarter.

Now, here’s some math: annualize $16.8 billion a quarter and you get about $67 billion. Divide by 3.5 gigawatts, and Anthropic is earning roughly $19 billion per gigawatt.

That’s 1.7x Goldman’s bar, and Anthropic’s earning it today, which can then travel across the flywheel.

Jensen Interview with BG2 / Altimeter

Anthropic’s bear case is that revenue growth is slowing. This is a big topic, and we can focus on this next week.

The amount of compute each customer needs is also rising fast.

Jensen Huang explained why in his recent interview with Brad Gerstner (Altimeter). You can also watch the best insights from this interview on the Lumida Invest App. Download now.

You can now view Lumida Curations in the Lumida Invest App. Curated insights delivered on time. Download the app here.

Jensen Huang argues that AI compute demand is compounding because reasoning models explore, test and prune multiple paths before answering, using far more inference than simple recall, and because better output means they get hired for more jobs.

Watch Curation

Elon Musk argues that Tesla can approach chip fabs the way it approached cars and rockets: start with standard equipment, find the bottlenecks, then build custom machinery to push wafer throughput far beyond today’s process.

Watch Curation

Mark Zuckerberg argues that Meta is a deep technology company that builds all the way down to the hardware so it can control the full experience, and that the AI era will be shaped by a small handful of companies capable of building frontier superintelligence.

Watch Curation

Not Subscribed Yet? Don’t miss out on future insights—subscribe to the newsletter now!

For real-time updates, follow us on: 

X | Telegram | YouTube | TikTok | News | Ram’s X | Lumida Health | Lumida Tax | Lumida Memes

As Featured In

Disclaimer: Lumida Wealth Management LLC (“Lumida”) is located in New York, NY, and is an SEC registered investment adviser. SEC registration does not constitute an endorsement of the firm by the Commission nor does it indicate that the adviser has attained a particular level of skill or ability. Lumida only transacts business in those states in which it is registered, or qualifies for an exemption or exclusion from registration requirements. Any direct communication by Lumida with a prospective client will be conducted by a representative that is either registered or qualifies for an exemption or exclusion from registration in the state where the prospective client resides.

The information in this material has been obtained from sources believed to be reliable. While all reasonable care has been taken to ensure that the facts stated in this material are accurate and that the forecasts, opinions and expectations contained herein are fair and reasonable, Lumida, Inc. and Lumida Wealth Management LLC (collectively Lumida) make no representations or warranties whatsoever the completeness or accuracy of the material provided, except with respect to any disclosures relative to Lumida. Accordingly, no reliance should be placed on the accuracy, fairness or completeness of the information contained in this material. Any data discrepancies in this material could be the result of different calculations and/or adjustments. Lumida accepts no liability whatsoever for any loss arising from any use of this material or its contents, and neither Lumida nor any of its respective directors, officers or employees, shall be in any way responsible for the contents hereof, apart from the liabilities and responsibilities that may be imposed on them by the relevant regulatory authority in the jurisdiction in question, or the regulatory regime thereunder. Opinions, forecasts or projections contained in this material represent Lumida’s current opinions or judgment as of the day of the material only and are therefore subject to change without notice. Periodic updates may be provided on companies/industries based on company-specific developments or announcements, market conditions or any other publicly available information. There can be no assurance that future results or events will be consistent with any such opinions, forecasts or projections, which represent only one possible outcome. Furthermore, such opinions, forecasts or projections are subject to certain risks, uncertainties and assumptions that have not been verified, and future actual results or events could differ materially. The value of, or income from, any investments referred to in this material may fluctuate and/or be affected by changes in exchange rates. All pricing is indicative as of the close of market for the securities discussed, unless otherwise stated. Past performance is not indicative of future results. Accordingly, investors may receive back less than originally invested. This material is not intended as an offer or solicitation for the purchase or sale of any financial instrument. The opinions and recommendations herein do not take into account individual client circumstances, objectives, or needs and are not intended as recommendations of particular securities, financial instruments or strategies to particular clients. The recipients of this material must make their own independent decisions regarding any securities or financial instruments mentioned herein and should seek advice from such independent financial, legal, tax or other adviser as they deem necessary. Lumida may trade as a principal on the basis of its views and research, and it may also engage in transactions for its own account or for its clients’ accounts in a manner inconsistent with the views taken in this material, and Lumida is under no obligation to ensure that such other communication is brought to the attention of any recipient of this material. Others within Lumida may take views that are inconsistent with those taken in this material. Employees of Lumida not involved in the preparation of this material may have investments in the financial instruments or securities (or derivatives of such financial instruments or securities) mentioned in this material and may trade them in ways different from those discussed in this material. This material is not an advertisement for or marketing of any issuer, its products or services, or its securities in any jurisdiction.