Friends, on the surface, Oracle's earnings report shows strong AI cloud orders.
But the market's reaction is like a boss who sees orders booming and suddenly realizes the kitchen needs to be rebuilt, staff doubled, electricity bills explode, and cash flow bleeds first.
Do you think Wall Street gets excited seeing AI orders?
Of course it does.
But after three seconds of excitement, it whips out a calculator and asks coldly:
How much will it cost to fulfill these orders?
That's the most thrilling part after Oracle's earnings.
AI cloud demand is still there.
Orders are not weak either.
But the stock price is still under pressure.
Why?
Because the market is beginning to realize a very practical problem:
AI is not free.
The bigger the cloud orders, the more money data centers need to burn.
The stronger the customer demand, the scarier the capital expenditure.
The prettier the revenue story, the more profit margins and cash flow need to be re-examined.
Suddenly, the logic for AI software stocks has changed.
Before, everyone only asked:
Who can eat AI demand?
Now they ask:
Who can eat AI demand without burning themselves to ashes first?
So today, the most noteworthy thing is not just Oracle.
It's that Oracle has sounded an alarm for the entire AI software market:
In the second half of AI, you can't just look at how big the orders are.
You also need to look at who can charge lighter, steadier, and more sustainably.
That's why institutions are now focusing on the infrastructure software Fab Five named by BofA:
They are not necessarily the companies that make headlines best.
But their positions are very practical.
When AI agents need to work and need data, who charges?
When AI systems run and need monitoring, who charges?
When AI applications need to be developed and need databases, who charges?
When AI customer service needs to reach customers and needs communication APIs, who charges?
When AI software needs continuous deployment and updates and needs supply chain tools, who charges?
That's the core of today's issue.
So it's not that AI software stocks are ebbing.
It's that institutions are rotating from the most crowded big stories to second-tier toll stations along the AI implementation chain.
Today we only break down the most practical question:
After Oracle's earnings made the market see the cost pressure of AI cloud, why are institutions starting to dig into second-tier AI software toll stations?
Among these Fab Five, who is most likely to take the baton?
Before reading, please like and bookmark this. Today's issue is not urging you to rush in and buy tickets, but to help you see clearly where institutional money is moving.
First, why Oracle's earnings became a signal for AI software stock rotation.
Oracle's problem is not the lack of AI demand.
On the contrary, AI cloud orders and cloud infrastructure demand are still very strong.
What the market truly fears is the cost of fulfilling this demand is too high.
AI cloud is not a light-asset business.
It's not that you write a few lines of code today and profits pour in tomorrow.
You have to build data centers.
Buy hardware.
Solve power supply.
Do cooling.
Raise financing.
Bear depreciation.
And face long-term capital expenditure pressure.
It's like a restaurant suddenly overwhelmed with orders.
Of course the owner is happy when people are queuing around the corner.
But the next second, he finds the kitchen is too small, the fridge is insufficient, staff is inadequate, rent is rising, utilities are exploding, and equipment needs loans.
Then the question arises:
The business looks great, but is the owner making money, or exhausting himself just to take orders?
Oracle now has that flavor.
AI orders are huge, which is certainly good.
But the market now starts asking:
Are these orders profit opportunities or capital expenditure black holes?
So after Oracle's earnings, the market didn't simply cheer.
It shifted from "order imagination" to "ledger judgment."
This sentence is important:
Oracle told the market that AI demand is real, but so are AI bills.
In the most comfortable days of the AI rally, everyone only looked at demand.
Strong demand = stock up.
Big orders = stock up.
Many customers = stock up.
But now entering the second half, the market won't be so naive.
It will ask:
How much will it cost to meet this demand?
How much profit can these orders bring?
Will this customer explode your capital expenditure?
Is this AI story a cash flow engine or a cash flow shredder?
That's the biggest change in AI software stocks now.
Demand is not the problem.
The delivery method is the problem.
So money starts thinking:
Are there some companies that can capture AI implementation demand without having to madly build data centers and burn capital expenditure like AI cloud?
Are there some companies that are stuck in the AI implementation chain, charging continuously through software, data, monitoring, APIs, and toolchains?
This leads to the second point:
Why institutions are shifting from big stories to second-tier toll stations.
AI cloud is responsible for burning money to expand capacity; infrastructure software is responsible for collecting tolls on the road to implementation.
You must remember this sentence.
In the second half of AI, it's not just about who can burn the most money, but also who can charge tolls on the road where others are burning money.
BofA's Fab Five—Snowflake, Datadog, MongoDB, Twilio, JFrog—are exactly stuck in these positions.
They are not the same type of company.
They are not those AI star stocks that make retail investors' blood boil just by hearing the name.
But their commonality is clear:
The more AI is implemented, the more important their positions become.
Because AI truly entering enterprises doesn't end with buying a GPU.
It needs data.
Needs monitoring.
Needs databases.
Needs communication interfaces.
Needs software delivery.
Needs security governance.
Needs continuous operations.
These links are not optional decorations.
They are the water, electricity, and gas that enterprises cannot bypass when AI truly runs.
You may not like these companies.
But you cannot deny these links.
That's why institutions are now re-studying second-tier software stocks.
Not because they suddenly became sexy.
But because after AI implementation, whether it's sexy or not doesn't matter; whether it can charge is what matters.
First company, Snowflake.
Snowflake's position is not simply "I also have AI features."
It represents the data granary behind AI agents.
If an AI agent wants to make decisions for an enterprise, it can't rely on guessing.
It needs to read customer data.
Read sales data.
Read inventory data.
Read financial data.
Read product data.
And read this data under security, permissions, and compliance frameworks.
The problem is, most enterprises' data is not neatly placed on the table waiting for AI to eat.
Reality is often:
One system for sales.
One system for finance.
One system for customer service.
One system for supply chain.
Permission rules are another set.
Data is scattered like melon seed shells after a family reunion dinner, everywhere on the floor.
At this point, you invite an AI agent in. Of course it can talk.
But if it can't get clean, reliable, governable data, it's like a consultant who talks well but hasn't seen the books.
Sharp mouth, dangerous advice.
What enterprise AI fears most is not that AI can't talk.
What it fears most is that AI talks confidently but the underlying data is wrong, messy, and incomplete.
That's not improving efficiency.
That's automating disaster.
So Snowflake's opportunity is:
The more enterprise AI moves toward production environments, the more it needs a platform where data can be safely called, governed, and consumed.
If AI agents need to eat, data is the granary.
And Snowflake wants to be that granary.
But we can't oversell here.
Snowflake can't just rely on the big word "data foundation" to support its valuation.
The market ultimately wants to see:
Is customer consumption increasing?
Is Cortex AI driving higher usage?
Can partnerships with large model companies translate into real revenue?
If the presentations are beautiful and the partner list is luxurious, but customer consumption hasn't significantly risen, the market won't keep giving patience.
For Snowflake, the key is not whether there's a story.
The key is whether customer usage truly grows.
Second company, Datadog.
For Datadog, I prefer to call it:
The operations tax of AI production environments.
It sounds unsexy, but businesses that are very profitable often don't rely on sexiness.
Once an enterprise really puts AI agents into production, it discovers a problem:
Deploying AI is just the beginning.
Managing AI is the trouble.
Are agents running wild?
Is the call chain broken?
Is inference cost exploding?
Are APIs abnormal?
Are user requests stuck?
Have security incidents been detected?
Has the model output suddenly gone crazy?
The boss can't solve these problems by burning incense every day.
You can't throw a bunch of AI agents into the enterprise system and then clasp your hands and say:
Guys, please don't mess up today.
That's not management.
That's superstition.
So enterprises need monitoring, logs, tracing, alerting, and observability.
Datadog's opportunity is not doing AI itself.
It sells the operations tax after AI runs.
The more complex the AI system, the less enterprises dare to go bare.
Previously, enterprises monitored servers, databases, APIs, and application performance.
In the future, they will also monitor model calls, agent behavior, inference cost, latency, abnormal requests, and security events.
That's why institutions see Datadog.
More AI deployments lead to more complex systems.
More complex systems make problems harder to debug.
Harder to debug makes enterprises more willing to pay for observability.
So Datadog is not selling a cool AI feature.
It is selling the peace of mind after AI enters production environments.
But the risk is also clear.
Datadog is no longer an unnoticed cold ticket.
The market already recognizes the value of AI monitoring.
So it can't just keep living on keywords like AI observability.
It must deliver real growth.
Are customers expanding deployments?
Is revenue continuing to accelerate?
Are profit margins stable?
Are AI-related products making a real contribution?
These are the keys.
If the conferences are lively and PPTs are beautiful, but no changes in earnings reports, the market will turn hostile.
Datadog's line is good, but it has entered a phase of "strong expectations, strong verification."
Third company, MongoDB.
MongoDB is different from Snowflake.
Snowflake leans toward enterprise data analytics and data platform.
MongoDB leans more toward application development and modern databases.
Its position can be understood as:
The memory system for AI applications.
Why is that?
AI applications don't end after one model call.
A real AI application needs to save user context.
Save task state.
Save conversation history.
Save tool call results.
Save documents and unstructured data.
And support rapid iteration and developers building new applications.
Especially for agentic AI applications.
Today the user asks the agent to look up information, tomorrow to place an order, the day after to contact a customer, and then to generate a report.
In this process, there are many states, memories, task chains, and data structures to save.
This cannot all be solved with traditional simple tables.
Developers need flexible data storage.
Need modern databases.
Need foundational capabilities that support complex application scenarios.
That's MongoDB's position.
It is not the stage lighting of AI applications.
It is more like the memory system of AI applications.
Stage lighting is dazzling, but if the memory system breaks, the whole application is like an amnesiac patient.
It forgets everything said a minute ago.
Enterprises cannot accept something like that in core processes.
So the more agentic AI develops, the more complex application states become, the more developers need flexible data storage and database capabilities.
That's where MongoDB may benefit.
But risks must also be stated clearly.
MongoDB needs to prove that the heat of AI application development can translate into real revenue growth, not just stay in the developer community being lively.
Developers liking it is one thing.
Enterprises paying is another.
Many stars on GitHub do not equal many cash flows for the company.
So for MongoDB, what to watch next:
Are enterprise customers growing?
Is Atlas usage increasing?
Is AI application development driving database consumption?
If these metrics don't keep up, the story of AI application memory system may just be a nice narrative.
Fourth company, Twilio.
Twilio is not the most typical AI software stock.
But it is stuck in a very practical position:
The communication pipe between enterprises and customers.
In the future, AI agents won't just sit in the back analyzing data.
They will send SMS.
Make calls.
Send WhatsApp messages.
Do voice interactions.
Make customer notifications.
Handle customer service inquiries.
Interact with sales leads.
All these actions need communication APIs.
AI customer service doesn't operate in thin air.
AI sales doesn't sell in their minds.
They ultimately need to reach customers.
Send messages, make calls, confirm appointments, send verification codes, make post-sale reminders, handle order progress.
All these rely on communication infrastructure.
Twilio's opportunity is here.
If enterprises really start using AI agents for customer service, sales, marketing, and customer operations, communication API demand may be reactivated.
We can compare with Meta.
Meta uses WhatsApp, Instagram, Messenger to grab entrances.
It is the platform entrance.
Twilio provides programmable communication pipes to enterprises.
It is the underlying water pipe.
One stands at the customer entrance front desk.
The other hides in the enterprise system back office.
But the back office water pipe is also indispensable.
For an AI agent to talk to customers, it needs telephone lines, SMS channels, and message interfaces.
Otherwise, no matter how smart the agent is, it can only sit inside the company talking to itself.
But Twilio's risk is also realistic.
Its past growth has been volatile, and the market won't suddenly change its view just because of the word AI.
Twilio must prove:
AI customer service, AI sales, and automated communication can really drive API usage and revenue recovery.
Otherwise it's just briefly illuminated by the AI concept, not truly revalued.
So Twilio's line is very practical, but also needs earnings to prove it.
Fifth company, JFrog.
JFrog is not as good at telling stories as Palantir, nor as sexy as Cloudflare.
But its position is very foundational.
You can think of it as:
The supply chain security guard behind AI software.
AI software doesn't end when code is written.
Once it truly enters enterprises, it needs continuous updates.
Continuous deployment.
Continuous testing.
Continuous scanning for security risks.
Continuous management of dependency packages.
Continuous checking of supply chain vulnerabilities.
In the AI era, the software supply chain will only be more complex.
Because there are more codes.
More models.
More automated deployments.
More dependency packages.
More open source components.
More security risks as well.
Do you think enterprise AI development is just a group of engineers dancing around a large model?
No.
Behind it is a complete software engineering system.
Where does code come from?
Are there vulnerabilities in package dependencies?
How are versions managed?
How are deployments tracked?
How are models and applications updated?
Has security scanning been done?
Can you roll back if something goes wrong?
These questions are very mundane.
But what enterprises fear most are precisely these mundane issues.
Because what can really break a company is not necessarily a grand AI strategic mistake.
Sometimes it's a supply chain vulnerability, a dependency package contamination, a deployment error, an undetected security risk.
JFrog's position is helping enterprises manage these software delivery and supply chain issues.
It may not stand in the spotlight.
But if enterprises are truly developing a lot of AI applications, software delivery, security scanning, package management, and DevOps processes will all become more important.
In one sentence:
The more AI applications, the less the underlying software supply chain can be chaotic.
But the risk is that JFrog's story is too foundational; market imagination may not be as direct as Palantir and Cloudflare.
Retail investors tend to prefer stories that sound world-changing at first listen.
JFrog is like the rebar in a building.
No one praises it normally.
But when something goes wrong, everyone realizes how important the rebar is.
So JFrog needs to prove through earnings and customer growth that the AI development wave is indeed driving up software supply chain demand.
By now, you'll see that the Fab Five are worth looking at not because they have AI in their names.
Nor because institutions named them, they will definitely rise.
What really matters is:
Institutions are breaking down AI software stocks more finely.
Previously, the market bought AI software like buying a big basket.
Whoever touched AI had a chance.
Now institutions start to break down:
Who has the strongest data?
Who has the hardest monitoring?
Who benefits from application databases?
Who handles customer touchpoints via communication APIs?
Who controls enterprise delivery via software supply chain?
This shows AI software stocks have entered a more professional pricing phase.
Not every company that touches AI can rise.
But companies stuck at key links will be re-studied by the market.
So second-tier AI software opportunities are not a mindless catch-up logic.
It's a toll station logic.
This is very important.
What is catch-up logic?
Others have risen, you haven't, so you should also rise.
This logic is dangerous.
Because not rising may not be an opportunity; it could also be that the market simply doesn't want to buy.
Toll station logic is different.
It asks:
After AI truly lands, can the enterprise bypass you?
If it can't be bypassed, you have the possibility to charge tolls.
If it can be bypassed, no matter how good the story, it's useless.
So how should ordinary investors observe next?
First, see if capital flows continue to return to infrastructure software.
A one-day rebound doesn't mean much.
Continuous returns indicate institutions are truly rotating.
Second, see if these second-tier stocks can hold up during market volatility.
Strong stocks don't necessarily surge every day.
But whether they can fall less during downturns is important.
If the market fluctuates and they fall harder than anyone else, it means capital is just short-term hype, not true recognition.
Third, see if earnings reports show usage growth driven by AI.
Snowflake: consumption.
Datadog: customer expansion and product adoption.
MongoDB: developer demand and enterprise revenue.
Twilio: communication API usage.
JFrog: enterprise customers and software supply chain demand.
Fourth, see if valuations have been bid up too early.
Second-tier opportunities don't mean low risk.
If market expectations run ahead of earnings, they can also get crushed.
The most common mistake people make is:
The logic sounds right, so they forget about price.
But the stock market is not an essay contest.
It's not that whoever writes the best story will definitely rise.
If the price is too high, even a good company can become a bad trade.
Fifth, see if the company truly controls a necessary link in AI implementation.
Not every AI feature has value.
What is truly valuable is a position that enterprises cannot bypass.
Data, monitoring, application databases, communication APIs, software supply chain—why are these positions worth watching?
Because they are not the performance layer.
They are the foundation layer.
The performance layer is easily replaceable.
Once the foundation layer enters an enterprise system, customer switching costs are higher.
Next, there may be three scenarios.
First scenario: Institutions continue to return, second-tier infrastructure software takes the baton.
In this case, companies like Fab Five will continue to be mined by the market.
Capital will gradually spread from high-attention big stocks like Oracle, Palantir, Cloudflare to more niche infrastructure links.
Second scenario: Only a few companies emerge, with strong differentiation within the sector.
In this case, companies with clearer fundamentals and easier-to-explain positions like Datadog, Snowflake may be stronger.
Other companies need to prove themselves with earnings.
Third scenario: Tech stock risk appetite continues to decline, and high-valuation software stocks get re-rated downward.
In this case, even if the long-term logic isn't broken, short-term attention will be on valuation and profit delivery.
So this is not a blind bet.
It's a screening game.
You can't just because institutions named them, assume all five will rise.
You need to see whose toll station is hardest, whose revenue verification is clearest, whose valuation hasn't been bid to the moon yet.
Finally, a summary.
Oracle's earnings reminded the market:
AI cloud orders are huge, but AI bills are also huge.
When capital starts to realize the capital expenditure pressure behind first-tier AI cloud stories, it naturally seeks lighter, finer, more sustainable software links that can charge continuously.
The five companies mentioned above may not all skyrocket.
But together they illustrate a direction:
AI software stocks are not ebbing.
They are shifting from big stories to toll stations.
In the first phase, the market bought whoever could tell the best AI story.
In the second phase, institutions start looking for who can charge continuously during AI implementation.
That's the most noteworthy change in the second half of AI software stocks.
If you found this issue valuable, remember to like, bookmark, and subscribe.
In the next issue, we'll continue to analyze among the second-tier AI software list, who is most likely to be an underestimated dark horse that institutions overlook.
We'll continue next time.
