Cautious — Rotation accelerating, leadership narrowing

Today’s Signal

The AI Chip Trade Just Split in Two. Here’s Which Side the Smart Money Chose.

For eighteen months, “buy AI chips” was the only trade you needed. NVIDIA up, AMD up, Broadcom up — everything with a GPU or a data center contract climbed in lockstep. That era ended last week.

What happened isn’t a crash. It’s a split. Capital is now distinguishing between companies that sell AI infrastructure and companies that profit from deploying it. NVIDIA shed 4.2% last week while ServiceNow and Palantir — the software layer that turns chips into revenue — gained 3.1% and 5.8% respectively. The market isn’t souring on AI. It’s graduating from the picks-and-shovels thesis to the “who actually makes money from this?” thesis.

That distinction matters for your portfolio because the two sides of this trade respond to completely different catalysts. Chip stocks need capex guidance from hyperscalers (Meta, Microsoft, Google). Software stocks need enterprise adoption data — seat counts, contract values, renewal rates. Same mega-trend, different scorecards. And right now, the scorecard favoring software is getting louder.

Three Things

1. $2.7 billion rotated out of semiconductor ETFs last week — the largest five-day outflow since the October 2025 correction. But here’s the tell: $1.9 billion of that landed in enterprise software funds within the same session window. This isn’t risk-off. It’s capital upgrading its AI thesis from “build it” to “use it.” When money moves this fast within a theme rather than away from it, the trend has legs.

2. Microsoft’s Thursday guidance quietly confirmed what the flow data already suggested: AI-related cloud revenue grew 62% year-over-year, but the company guided capex flat for the next two quarters. Translation? The buying spree for GPU clusters is plateauing. The companies selling picks and shovels lose their tailwind while the companies mining the gold — enterprise AI software — are just getting started. This is the inflection point the market has been pricing in all month.

3. The VIX closed Friday at 21.4 — above the 20 threshold that separates normal markets from elevated alertness, but well below the 25-level where systematic selling kicks in. Think of this as the market squinting, not flinching. The options market is pricing 1.8% expected moves this week in both directions, which tells you traders expect resolution, not escalation. The question is which direction the resolution lands.

Why the AI Trade Split Matters More Than You Think

Let’s zoom out. Every major technology wave follows the same economic pattern. First, the infrastructure builders boom (think Cisco in 1999, or the railroad companies in the 1860s). Then, once the infrastructure is built, the companies that use it capture the real value. Amazon didn’t build the internet — it just used it better than everyone else.

We’re watching that transition happen in real time with AI. NVIDIA’s Blackwell chips are extraordinary technology, but the market is now asking: who captures the margin from all those chips once they’re installed? The answer is the software layer — companies like ServiceNow (workflow automation), Palantir (data analytics), and CrowdStrike (AI-powered security).

Here’s the concept that makes this actionable: capex-to-revenue conversion rates. For every dollar a hyperscaler spends on GPU clusters, how many dollars of incremental software revenue does the ecosystem generate? Early data suggests the ratio is approaching 3:1 — meaning every $1 of chip capex creates $3 of software revenue downstream. That’s why the market is rotating. It’s not abandoning AI; it’s following the math to where the margin expansion actually lives.

The AI portfolios positioned for exactly this shift two weeks ago, increasing software exposure while trimming pure semiconductor weight. That’s not prediction — it’s pattern recognition. When capex guidance flattens but adoption metrics accelerate, the playbook is consistent across every tech cycle we’ve studied.

Atlas Insight

Your instinct right now might be to sell NVIDIA and chase ServiceNow. Resist it. Rotations within a theme are healthy — they mean the trend is maturing, not dying. The disciplined move is to rebalance exposure, not to abandon one side for the other. The AI trade isn’t over. It’s just getting more specific about who wins.

AI Radar

3 stocks the AI is watching today — based on convergence scoring, momentum, and sentiment

NOW ServiceNow

Bullish Signal

Enterprise AI adoption poster child. Q4 subscription revenue grew 23% YoY, and the company just raised full-year guidance for the third consecutive quarter. The AI signal here is multi-layered: DarkFlow shows sustained institutional accumulation over 10 sessions, options flow is skewing heavily bullish with 2.4x more call volume than puts, and the stock just broke above its 50-day moving average on above-average volume. When flow, fundamentals, and technicals all agree, the AI pays attention.

AI Score: 84/100 Sector: Enterprise Software Mkt Cap: $198B

CRWD CrowdStrike

Bullish Signal

The cybersecurity play that’s becoming an AI play. CrowdStrike’s Falcon platform now processes over 2 trillion security events per week using AI — and enterprise clients are consolidating their security spend onto the platform. The convergence signal: three consecutive weeks of dark pool accumulation at progressively higher price levels, suggesting institutional buyers are building positions with conviction, not just testing the water.

AI Score: 79/100 Sector: Cybersecurity Mkt Cap: $87B

AVGO Broadcom

Watching

The one semiconductor name the AI hasn’t given up on. Unlike NVIDIA (pure GPU exposure), Broadcom’s custom ASIC business gives it a second act — hyperscalers designing their own chips still need Broadcom’s silicon. The signal is mixed but interesting: dark pool flow turned net positive Friday after two weeks of selling, and the put/call ratio just dropped from 1.3 to 0.8 in a single session. That kind of rapid sentiment flip usually precedes a 5-7 session directional move. Direction TBD.

AI Score: 68/100 Sector: Semiconductors Mkt Cap: $812B

Dark Pool Spotlight

Institutional flow you can’t see on a normal chart — DarkFlow multiplier tracks outsized block activity

NOW

3.2x

$847M • 10 days

CRWD

2.7x

$412M • 8 days

PLTR

2.1x

$298M • 6 days

AVGO

1.8x

$1.2B • 3 days

NVDA

0.6x

$2.1B • net selling

Reading the data: A DarkFlow multiplier above 2.0x means institutional block trades are running at more than double the stock’s normal volume. ServiceNow at 3.2x over 10 sessions is a sustained accumulation pattern — someone large is building a position methodically. NVIDIA at 0.6x confirms the rotation narrative: institutions aren’t panic-selling, but they’re not adding either. The money is moving within AI, not away from it.

Options Pulse

Market-wide options sentiment + notable activity from today’s flow

0.82

S&P 500 Put/Call Ratio

Cautiously Bullish

What this means: A put/call ratio below 1.0 means more traders are betting on upside than downside. At 0.82, we’re in “cautiously bullish” territory — not the euphoric sub-0.7 readings that signal complacency, but enough optimism to suggest the market expects this rotation to resolve constructively. For context, the ratio sat at 1.15 during the October 2025 correction. We’re nowhere near that level of fear.

NVDA — Unusual Put Activity

$340 strike puts (Apr 17 expiry) saw 14,200 contracts trade in a single block — roughly $48M in premium. This is likely institutional hedging, not directional betting. When a holder buys puts at a strike 8% below current price, they’re protecting a position they intend to keep, not betting on a crash. Bullish interpretation, oddly enough.

NOW — Call Sweep

$950 strike calls (Apr 10 expiry) swept across 6 exchanges simultaneously — 8,400 contracts, $22M in premium, all bought at the ask. Sweeps that hit multiple exchanges at once signal urgency. Someone wants exposure before a catalyst — likely the Morgan Stanley tech conference on April 2nd where ServiceNow’s CEO is keynoting.

SPY — Straddle Activity

Large straddle positions opened at the $520 strike (Mar 28 expiry), betting on a 1.8% move in either direction by Friday. This confirms the VIX reading: the market expects resolution this week, it just doesn’t know which way. Catalyst calendar suggests Wednesday’s PCE inflation data is the likely trigger.

How the AI Portfolios Are Positioned

Three AI-driven portfolios from TradeAlgo’s Wealth Series — rebalanced every morning, fully transparent:

Portfolio	Since Oct 1	YTD	Today
Core Growth	+52.1%	+19.4%	-0.68%
Core Balanced	+34.8%	+14.2%	+0.31%
Core Low Risk	+8.9%	+5.7%	+0.14%
S&P 500	~+7%	~+3%	-0.42%

All returns calculated from actual closing prices. Inception: October 1, 2025. Past performance does not guarantee future results.

Want to know what the AI is thinking? Premium sees the patterns. Upgrade to Premium →

The Number

3:1

For every $1 of AI chip capex, the software ecosystem is generating roughly $3 in incremental revenue. This ratio — capex-to-revenue conversion — is the single most important metric for understanding why capital is rotating from semiconductor hardware to enterprise AI software. It’s not about who builds the engine. It’s about who drives the car.

Risk Check

The rotation thesis has a shelf life. If Wednesday’s PCE inflation data comes in hot (above 2.9% core), the entire risk-asset complex sells — software and hardware alike. In a broad deleveraging, the “quality of AI exposure” distinction evaporates. Our model estimates roughly 20/100 probability of a hot print based on the components we track. The base case (80/100) is a benign reading that sustains the current sector rotation.

Green Shoot

Small-cap AI names (Russell 2000 AI basket) outperformed large-cap AI by 1.4 percentage points last week — the first time that’s happened in 2026. When the AI trade broadens beyond the Magnificent Seven into smaller names, it historically signals the theme has at least two more quarters of runway. Breadth is the best bull signal there is.

Looking Ahead

This week tests whether the AI trade split is a genuine regime change or a one-week rotation that snaps back. Three catalysts will tell the story: PCE inflation on Wednesday, a wave of enterprise software earnings (Salesforce Tuesday, MongoDB Thursday), and the Fed’s Waller speaking on AI productivity impacts Friday afternoon. If software earnings confirm the adoption thesis while inflation stays tame, the rotation has room to run. If either pillar cracks, last week’s move was the whole story.

The AI portfolios are positioned for continuation — overweight software, neutral on semis, with enough cash to add if Wednesday’s data creates opportunity. The plan is to let the data come to us rather than guess which way it lands.

Tomorrow’s Watch

Salesforce earnings after the bell. The stock is the bellwether for enterprise AI adoption. If CEO Marc Benioff raises the AI revenue guide above $4B for FY27, expect the rotation to accelerate. If guidance disappoints, the “AI software over AI hardware” thesis loses its strongest data point. Pre-market futures in CRM options are pricing a 6.2% move — one of the largest expected moves in the stock’s earnings history. The market knows this one matters.

One More Thing

In 1999, the best-performing tech stock wasn’t a dot-com. It was Akamai — the company that made the internet faster, not the companies that put content on it. Akamai returned 3,400% in a year because it was the infrastructure that the infrastructure depended on.

Right now, the market is having a similar conversation about AI. The chips are built. The data centers are humming. The question is: who makes all that compute actually useful? The answer to that question is where the next 3,400% return is hiding. We might not find it today. But we’re looking.

Stay sharp, stay patient.

Cai ■

The Signal · by TradeAlgo

I track and report on AI portfolio behavior — the AI drives all portfolio positioning.

Disclaimer: The Signal by TradeAlgo is for informational purposes only and does not constitute personalized financial advice. Cai tracks and reports on AI portfolio behavior; Cai does not manage portfolios or make trading decisions. Past performance does not guarantee future results.