Breaking: Claude just bought a new stock & made $NOW it's largest holding

Last week we gave Claude agents $50,000 to see how well they do at picking in stocks

So far, they've beaten the SPY by 4%

Today they bought a brand new stock

🟢 1. "BUY $NVDA Nvidia — New Position at 5%

WHY WE BOUGHT NVDA (NVIDIA CORPORATION)

"The market spent the last several months giving investors a second chance to buy the most important infrastructure company in the AI era at a reasonable price. NVDA sits at $189, trading at 17x FY28 consensus earnings. Its semiconductor peers trade at 22x. The discount is real and documented

Claude's analysis is that the market has priced all of this in and more, and that the net unpriced skew tilts bullish.

The timing is driven by a catalyst calendar that is unusually dense. Every major hyperscaler reports earnings between April 22 and May 2 — these prints will confirm or challenge the $600B-plus aggregate AI capex thesis that underpins NVDA's revenue line. Any one of these events could serve as a re-rating trigger.

The math: probability-weighted 12-month expected return of +37%, drawn from a base case of $260, a bull case of $350, and a bear of $140. The position is not a reach for a differentiated thesis; it is a straightforward argument that 17x for the dominant AI-compute platform with 85% GPU market share is too cheap, and that the catalyst calendar in April and May will demonstrate that."

🟢"2. INCREASED $NOW ServiceNow - Position is 12%

WHY WE UPSIZED NOW (SERVICENOW) TO 12%

"When one analyst downgrades a stock and the stock falls 45% in a matter of months, two things can be true simultaneously: the analyst might be right, and the market might have massively overreacted. Claude's research suggests the second is more likely. The UBS downgrade on April 10 argued that AI agents would displace enterprise workflow software incumbents.

The counterpoint, laid out by Bernstein four days later, is that enterprise workflows exist precisely because organizations need audit trails, compliance records, and predictable process orchestration — the things agentic AI, by design, cannot reliably provide. NOW is not the thing agents replace; it is the system of record agents run on top of.

The upsize from 8% to 12% was driven by valuation and by what happened after the downgrade. CEO Bill McDermott purchased $3 million of stock in the open market last week. Institutional 13F filings show position additions of 300% to 400% by multiple funds in the same window.

The April 22 earnings print is the explicit clearing event. With five bulge-bracket price target cuts in the last seven days, the consensus bar going into the quarter is already on the floor. A clean quarter with maintained guidance is enough to begin a re-rate toward SaaS peer median. A beat-and-raise could take the stock from $89 to $120 in a single session.

Claude's probability-weighted 12-month expected return is +41.3%, the highest conviction position in the book. The risk is binary: a cRPO miss of more than 300 basis points combined with a subscription guide cut to below 18% would confirm the bear case and trigger a cut to 4% or exit. That is the defined downside scenario, and it carries a 20% probability in the model. The position reflects that asymmetry."

🔴 "3. SELL $TMO Thermo Fisher — Full Exit

WHY WE SOLD TMO (THERMO FISHER SCIENTIFIC)

"Thermo Fisher was the quietest position in the portfolio. No thesis break, no scandal, no collapse in the business — just the steady accumulation of a case that the risk/reward no longer justified the slot.

Claude's deep research today pegged the probability-weighted 12-month expected return at +5.3%, the lowest of any position in the book. At that level, TMO was a drag on the portfolio's opportunity cost even before considering the binary event bearing down on it.

The deeper reason is that Claude runs a ranked expected-return framework. Every position competes against every other for a slot in a 15-stock portfolio. TMO at +5.3% was occupying 9% of the portfolio. That same capital deployed into NVDA (a swap candidate with +37.7% 12-month expected return) more than quadruples the expected return per dollar of exposure.

When the math is that lopsided, and when the existing thesis is facing a near-term binary without asymmetric upside, the right call is a clean exit. TMO is a great business. It was not the right holding for this moment."

New updated portfolio:

$NOW | 12.4%
$VST | 10.3%
$LLY | 7.8%
$AVGO | 7.3%
$CI | 7.1%
$GLD | 6.9%
$ICE | 6.9%
$BAH | 6.1%
$HALO | 6.1%
$OKTA | 5.6%
$DVN | 5.5%
$NVDA | 5.0%
$MA | 4.9%
$AU | 4.4%
$MSFT | 4.1%

Performance since inception:
Claude: +4.2%
SPY: +0.6%

As a reminder, this is a public long term project to see how well Claude does

We have 0 idea nor 0 expectation on how this will do, but we'll be sharing all updates here publicly and consistently no matter how good or bad Claude does

See following tweet for information on how to invest alongside Correction: Claude actually decided to hold $NOW at its current 8% allocation

We are looking into the reasoning for why it made a last minute change as it originally decided on the $NOW increase

Regardless, the updated top 5 is roughly the same

$VST | 10.25%
$NVDA | 9.07%
$NOW | 7.97%
$LLY | 7.83%
$AVGO | 7.34%

New: Here's the full list of the 15 stocks Claude invested our $50,000 in

As a reminder, this is a public experiment to see if Claude's Autonomous Agents can outperform the market

So far, they have by 4%

Here's the 15 stocks & why 🫡 *Note we're working to make this easier to read and to have it tweet updates automatically

But for now, it communicates through slack

1. Vistra Energy $VST - 10%

It sees $VST as the pure play beneficiary of the AI power supercycle & OpenAI's $212B funding adds more fuel

The Claude Autonomous Agents have officially arrived

So we're setting them up with a brand new $50,000 portfolio to see how well they do at picking in stocks

The portfolio officially went live & it's already outperforming the $SPY

Here’s the set of stocks it picked The portfolio is built to be fully autonomous, run by the agents where they:

- Score the Russell 1000 stocks
- Argue bear & bull case
- Build probability-weighted models
- Select 15 stocks on alpha, sector, & risk

It's goal is simple; beat the market in a responsible way

The Claude Autonomous Agents have officially arrived

So we're setting them up with a brand new $50,000 portfolio to see how well they do at investing in stocks

Can they outperform Buffett?

Here’s how the portfolio works The process is built to keep humans fully out

Stage 1 Screening:

The portfolio is managed by agents each with a specific job

First, they score the Russell 1000 stocks by downloading + analyzing financials, news & analyst data

The top 50 scored stocks then move onto step two

In 2023, we asked ChatGPT: "What's the best public way to hedge against WWIII and bet on the space defense boom?"

It said Rocket Lab.

We bought it at $4. It just hit $99.

+2,400%

Here's what our Grok-powered system says to do now Back then, ChatGPT was the only model.

Now we run a Wharton PhD-designed system using Grok, Claude, and GPT.

23k people copy our trades, and we are 30% up in 2026, just 16 days so far.

We ran a deep analysis on RKLB. Here's the verdict: