AI Is Killing Software, But API Is Eating the World
Everyone is panicking about AI destroying SaaS. They're looking at the wrong thing. The UI layer is dying. The API layer is exploding. The real winners aren't building AI. They're feeding it data and selling APIs.
The Year Salesforce Crashed, Twilio Was Up 52%
Over the past two weeks, Salesforce crashed. Adobe crashed. Monday.com crashed. ServiceNow crashed.
Every headline screams the same thing: AI is eating software. AI is killing SaaS. Software is dead.
Gartner projects enterprise software spending will rise 40% over the next three years, but all of it flowing to AI. The SaaS per-seat model is collapsing because AI agents don't need seats.
But in the same time frame, Twilio's stock gained 52% in 2024 full year (Nasdaq). Stripe's valuation hit $106.7 billion (Sacra, September 2025). Plaid's 2025 ARR reached $430 million with a valuation rebound to $6.1 billion (Sacra, 2025).
These three companies share one trait. Their core product isn't an app. It isn't an interface.
They only sell APIs.
While everyone panics about "software is dead," pure API companies are hitting all-time highs.
This is not a coincidence.
71% of Internet Traffic Is API Calls
In 2024, API calls accounted for 71% of all internet traffic (The New Stack).
Behind every webpage, every app, every ChatGPT conversation, APIs are doing the work.
API call volume grew 60% year-over-year (Visma, 2024). AI-driven API traffic grew 73% (The New Stack, 2024).
Gartner predicts that by 2026, over 30% of new API demand growth will come from AI and LLMs.
The number of active APIs globally has reached 200 million, projected to surpass 1.7 billion by 2030 (The New Stack).
Postman's 2025 report: 82% of enterprises now identify as "API-first," up from 66% in 2023. 65% of enterprises now generate revenue directly through APIs.
The AI API market was $3.31 billion in 2024 and is projected to reach $30.89 billion by 2032, a CAGR of 32.2% (Congruence Market Insights).
This isn't "software dying" data. This is software evolving.
Direction of evolution: shed the UI, keep the API.
Stripe's Core Product Is the API, Not the Dashboard
Stripe has a dashboard. It has a mobile app. It has a full management interface. As a seller, you can see every transaction, manage customers, and process refunds right in the backend.
But the interface isn't what Stripe sells. Stripe sells APIs. A developer writes a few lines of code and payment processing is live in any product. The dashboard is a management tool. The API is the core of the business model.
In 2025, Stripe's total revenue reached approximately $19.4 billion (Backlinko, Coinlaw). It processed over $1 trillion in payment volume (SQ Magazine). Valuation: $106.7 billion by September 2025 (Sacra).
Business model: 2.9% + $0.30 per successful transaction (Stripe website). No subscription fees. No per-seat pricing. More transactions, more revenue.
An AI agent won't invent its own payment system. It will call Stripe's API. More AI agents, more transactions, more Stripe revenue. The dashboard won't be called by AI, but the API will.
Jensen Huang said it clearly: AI won't replace software. AI will become software's biggest user.
Stripe is the proof. It proves one thing: in the API era, interfaces are cost centers. APIs are profit centers.
Why Twilio Gained While SaaS Collapsed
Twilio builds communication APIs. SMS, voice, video, email. It's not an app for people. It's a component for other software.
2024 revenue: $4.46 billion (Twilio earnings). Stock: up 52% for full year 2024 (Nasdaq), from a 52-week low of $57.51 to a high of $142.
While everyone eulogized SaaS, this pure API company surged.
The reason is simple: AI agents need to send texts, make calls, send emails. They won't build their own communication stack. They'll call Twilio's API.
By Q3 2025, Twilio revenue reached $1.3 billion, up 15% YoY, accelerating (Twilio earnings).
SaaS is falling. APIs are rising. Same industry. Two completely different outcomes.
Financial Data API Calls Grew 427% in One Year
Plaid is the data bridge between banks and fintech apps. When you connect your bank account on Venmo or Robinhood, Plaid's API is doing the work.
2025 ARR: $430 million, valuation rebounded to $6.1 billion (Sacra). Operating at positive margins.
In the Open Banking space: API calls surpassed 22 billion in the 12 months ending July 2025, up 36% YoY (Open Banking UK). Juniper Research projects Open Banking API calls will grow 427% from 137 billion in 2025.
The financial system is being reconstructed at the API layer.
Chainlink Solved a Problem for Blockchain. API Is Solving the Same Problem for AI
Blockchain has a fundamental flaw: it's a closed system. It doesn't know today's gold price. It doesn't know the weather. It doesn't know if a flight is delayed.
Chainlink built a decentralized "oracle" network that sources real-world data from multiple providers, reaches consensus, and feeds it to smart contracts. Without Chainlink, blockchain cannot interact with the real world.
Chainlink (LINK) reached a market cap near $20 billion. Because everyone realized: no matter which blockchain wins, they all need external data. Chainlink is non-negotiable.
AI large language models have the exact same flaw. Training data has a cutoff date. No access to live market data. Can't reach your bank account. Can't operate your calendar.
APIs solve this.
MCP (Model Context Protocol), released by Anthropic in late 2024, now adopted by OpenAI and Google (Wikipedia, 2025). Called the "USB-C port for AI." At its core: a standardized API calling framework. AI agents connect to external APIs through MCP, fetch data, execute operations.
In the blockchain world, if you understood "Chainlink is indispensable to DeFi" early in 2020, you made tens of multiples.
In the AI world, the same logic is replaying. No matter whether GPT, Claude, Gemini, or open-source models win, they all need APIs to connect to the real world.
API is AI's Chainlink.
The Only Difference Between What's Dying and What's Thriving
The distinction isn't "are you a software company." The distinction is "are you selling a UI or a capability."
Dying: companies that sell buttons. Core value is a nice interface helping users click through tasks. Project management tools, simple CRMs, basic accounting software. AI agents can execute those tasks directly. They don't need buttons.
Thriving: companies that sell capabilities. They sell irreplaceable underlying capabilities. Payment processing, identity verification, communication infrastructure, financial data pipelines, geolocation services. AI agents won't invent these. They'll call them via API.
| Dying | Thriving |
|---|---|
| Per-seat pricing | Per-API-call / per-transaction pricing |
| Core value is the UI | Core value is data and capabilities |
| Users are humans | Users are AI agents |
| Replaceable by AI | Amplified by AI |
The overall SaaS market is still growing, approximately $316 billion in 2025 and projected to reach $376 billion in 2026 (Fortune Business Insights). But nearly all growth is flowing to AI-native and API-first companies. Traditional "button-selling" SaaS is being squeezed.
One interesting exception: Vertical SaaS (industry-specific software). They have deep industry data and highly customized workflows. AI doesn't kill them. AI needs their data to train and execute industry tasks. AI makes Vertical SaaS more valuable.
Data Pipelines Are the Scarcest Asset of the AI Era
AI model capabilities are no longer the bottleneck. Models can be replicated, open-sourced. GPT has alternatives. Claude has alternatives. Open-source models are everywhere.
Data is the bottleneck. Whoever can feed high-quality, real-time, structured real-world data to AI owns the strongest moat.
Everyone chases AI model companies. NVIDIA, OpenAI, Anthropic. Very few watch the companies that feed data to AI.
Scale AI extracts real-world data from images, text, voice, and video to provide training data for AI models. It labeled millions of images for Waymo and Cruise (Scale AI website). It doesn't build AI. It makes AI's raw material.
Palantir's core value isn't the AI model. It's integrating scattered organizational data into a unified data platform. Its AI Platform provides anonymized, regulatory-grade structured data to AI models (Palantir earnings). Palantir's stock surged in 2024. The market understood: data pipelines are scarcer than AI models.
Databricks, Matillion, Rivery are growing rapidly. Enterprises discovered a fact: in the AI era, data engineering matters more than AI engineering.
Software Has Gone Through Four Evolutions
1990s: You go to a store. Call to order food. Visit a bank teller.
2000s: You shop on websites. Email to communicate. Use online banking.
2010s: You shop on apps. Slack to communicate. Stripe to get paid.
2020s: AI agents buy things via API. AI communicates via API. AI processes payments via API.
Each evolution sheds the previous layer's UI and preserves and amplifies the underlying capability.
The app UI layer was simplified by SaaS in the 2010s. The SaaS seat model is being replaced by APIs in the 2020s. APIs have no UI. APIs are pure capability.
API is the final evolutionary form of software.
Marc Andreessen's 2011 thesis "Software is eating the world" hasn't been overturned. It's been upgraded.
Software was eating the world. Now, API is eating software. And AI is the mouth.
Three Layers of Investment Opportunity
API Infrastructure
The highest-certainty layer. No matter which AI wins, they all need these pipes.
| Company | What it does | Key data |
|---|---|---|
| Stripe | Payments API | $106.7B valuation, $19.4B revenue (2025) |
| Twilio | Communications API | $4.46B revenue, stock +52% full year |
| Plaid | Financial data API | $430M ARR, valued $6.1B |
| Cloudflare | Edge compute / API gateway | API management as core growth engine |
| Kong / Postman | API management platforms | API management market $8.86B (2025) |
Their business models naturally transition from per-seat to per-call/per-transaction pricing. This isn't a downgrade. It's an upgrade. Per-seat growth is capped by customer headcount. Per-API-call growth has no ceiling. Every AI agent is a new "user."
Data Pipelines
Companies that feed data to AI. AI's Chainlink.
| Company / Direction | What it does | Investment logic |
|---|---|---|
| Palantir | Enterprise data integration | Data integration is the new moat |
| Scale AI | Data labeling and synthetic data | Private, but represents direction |
| Databricks | Data engineering platform | Private, valued at $43B |
| Snowflake | Cloud data warehouse | Data sharing and AI feature engineering |
| Open Banking APIs | Financial data pipelines | API calls growing 427% (Juniper) |
AI models can be replicated and open-sourced. High-quality data pipelines cannot. Companies with unique data sources own the truly scarce resource of the AI era.
AI-Native API Companies
Next-generation companies packaging AI capabilities as APIs.
| Direction | Examples | Logic |
|---|---|---|
| AI Translation API | DeepL API | Developers won't train their own translation models |
| AI Voice API | ElevenLabs API | Same |
| AI Vision API | Clarifai, Roboflow | Same |
| AI Search API | Tavily, Perplexity | AI agents need search capabilities |
| AI Coding API | Various Code LLM APIs | Foundation of software automation |
The AI API market grows from $3.31B in 2024 to $30.89B by 2032, CAGR 32.2% (Congruence Market Insights). Every AI capability is becoming a callable API endpoint.
Where This Thesis Could Be Wrong
SaaS isn't actually dying. Global SaaS revenue is projected to reach $344 billion by 2027 and surpass $1 trillion by 2032 (Forbes). 85% of business applications are already SaaS. SaaS as a delivery model isn't dead. What's dying is a specific type: SaaS whose core value is just the UI.
UIs still have irreplaceable value. In heavily regulated industries (finance, healthcare, law), humans need interfaces, audit trails, operation logs. AI agents can't fully replace those. Vertical SaaS won't die. It'll become more valuable because its industry data is scarce.
Data quality is still a massive problem. AI agents frequently fail on dirty data. Human-in-the-loop remains essential. Full automation is far away. This actually strengthens the thesis: data pipeline companies (cleaning, labeling, structuring data) are more valuable, not less.
The API economy could be monopolized by platforms. If OpenAI or Google builds everything, will independent API companies survive? History says yes. Google built Google Maps but didn't eliminate all geolocation API companies. AWS built every cloud service, yet Stripe and Twilio still exist. Generic platforms can't serve every vertical.
Risk Warnings
API security risk. API traffic is 71% of the internet. The attack surface is massive. API security incidents are rising.
Valuation risk. Stripe at $106.7B, Twilio's P/S isn't cheap. "API is the new SaaS" could become the next bubble narrative.
Technology path risk. MCP may not be the final standard. Google's A2A (Agent-to-Agent) protocol is a competing standard. Standards wars could fragment the ecosystem.
Regulatory risk. Open Banking faces strict regulation in Europe. Data privacy laws may limit API data access scope.
One Line
Don't sell software. Don't sell apps. Don't sell buttons.
Sell pipes. Sell APIs. Sell data.
In 2011, Marc Andreessen said "software is eating the world."
In 2026, API is eating software. AI is the mouth.
Don't panic about the mouth. Become part of the supply chain.
Author
More Posts
Need a Custom Solution?
Still stuck or want someone to handle the heavy lifting? Send me a quick message. I reply to every inquiry within 24 hours—and yes, simple advice is always free.
Newsletter
Join the community
Subscribe to our newsletter for the latest news and updates