AI, Software & Cloud Terms

AI that creates new output such as text, images, code and audio. As companies roll out more generative AI services, it is driving earnings at semiconductor, cloud and software companies.

e.g. Chatbots that write answers to your questions and tools that draw pictures from a sentence are typical examples.

An AI model trained on huge amounts of text so it can understand and produce language like a person. The bigger the model, the more computing power it needs for training and operation, which lifts demand for AI chips and data centers.

Training is the process of feeding data to an AI model to make it smarter; inference is using the finished model to produce answers. Training means large one-off spending, while inference creates steady demand as users grow, so each shapes revenue differently for related companies.

e.g. Building the chatbot is training; answering each question people ask it is inference.

The small chunks into which an AI model splits text for processing. AI services are often priced by token count, so more tokens processed can lift revenue for both AI and cloud companies.

How much companies want to buy dedicated chips, such as GPUs, that run AI workloads quickly. When cloud companies raise their AI spending plans, earnings expectations for chip, server and power-equipment makers rise too.

Renting servers or software over the internet instead of buying them. IaaS rents servers and storage, PaaS rents development environments and SaaS rents finished applications; a company's profit structure depends on which layer it sells.

e.g. Paying a monthly fee for company email and document software is SaaS.

A facility that houses many servers and runs them around the clock with power and cooling systems. As AI demand drives more data center construction, the benefit can reach power, cooling and real estate companies as well as chipmakers.

A giant cloud company that runs enormous data centers around the world. Their capital spending plans drive sales at AI chip and server makers, so the market watches them closely at earnings time.

e.g. Amazon (AWS), Microsoft (Azure) and Google Cloud are the main examples.

Revenue from customers who pay monthly or yearly to keep using a service. It is easier to forecast than one-time sales, so companies with a high subscription share often earn higher valuations in the stock market.

The revenue that current subscription contracts would generate over a year. It is a key gauge of a software company's growth speed, so investors always check its growth rate at earnings time.

e.g. 10 customers paying $1,000 a month gives ARR of $1,000 × 12 months × 10 customers = $120,000.

The share of customers or revenue lost over a period because subscriptions were canceled. A lower churn rate means customers are satisfied and revenue is steadier, so it matters a lot for subscription companies.

Compares what last year's customers spend this year with what they spent last year. Above 100% means existing customers are spending more, so revenue grows even without new customers, making it a quality measure for software companies.

e.g. If last year's customers spent $100M and the same customers spend $120M this year, NRR is 120%.

CAC is the marketing and sales cost of winning one customer; LTV is the profit that customer brings before leaving. LTV needs to be well above CAC for growth to actually make money.

Seat-based pricing charges per employee using the software; usage-based pricing charges for the data or computing actually used. If AI replaces human work, seat counts can shrink, so a company's pricing model affects its growth outlook.

The amount already contracted but not yet delivered, so it has not yet been recognized as revenue. It previews future revenue, so fast-growing RPO is read as evidence of future growth.

The total value of orders received but not yet delivered or provided. For cloud and AI infrastructure companies, backlog shows revenue visibility for the next several years and is often cited at earnings time.

A benchmark that treats a software company as healthy if revenue growth plus profit margin (usually operating or free cash flow margin) exceeds 40%. It lets you compare fast-growing but loss-making companies with slow but profitable ones using a single number.

e.g. Revenue growth of 30% plus a 15% margin totals 45%, which clears the bar.

Using services from several cloud providers at the same time. As more companies try to avoid being locked into one vendor, software companies that manage multiple clouds in one place can benefit.

Software that gathers a company's scattered data in one place to store, organize and analyze it. AI only works well on well-organized data, so these platforms draw attention as AI spending grows.

The industry that protects company systems from hacking and data leaks. Zero trust is a security approach that never trusts access automatically, even from inside the company, and verifies every time. Security is hard to cut even in a downturn, so security company revenue tends to be relatively stable.

Processing data right on nearby devices or small servers instead of sending it to a distant data center. It is used where fast response matters, such as autonomous driving and factory automation, and is a growth theme for related chip and network companies.

e.g. On-device AI, where AI features run directly inside a smartphone, follows a similar trend.

An AI model released so anyone can download, use and modify it. As strong open models spread, price competition can intensify for paid-model companies, while demand for chips and cloud capacity to run the models can grow.

AI that, given a goal, plans on its own and uses multiple programs to get the work done without being told each step. Software companies are promoting it as a new revenue source, so related announcements can move stock prices.

e.g. An AI that takes a business trip request and handles flight search, booking and calendar entry by itself is an example.

Total data center power divided by the power the servers actually use. The closer to 1, the less electricity is wasted on cooling and other overhead, so it shows operating efficiency in power-hungry AI data centers.

e.g. If servers use 100 and the whole facility uses 150, PUE is 1.5.

The rise and fall of big tech spending on AI data centers, servers and chips. That spending becomes revenue for semiconductor, power and networking companies, so news of higher or lower spending plans moves related stocks together.

Spreading the cost of expensive GPU servers over several years as an expense. Heavy AI investment raises expenses for years and can weigh on margins, and the number of years chosen (useful life) changes how profits look.

A newer cloud company focused on renting out GPU servers for AI. Growth is fast, but many borrow heavily to buy GPUs and may depend on a few large customers, so risk needs to be weighed too.

e.g. CoreWeave is an example of this category.

A service that lets customers rent GPUs by the hour instead of buying them. Rental prices and GPU utilization drive profit, so whether new chips cause older chips' rental rates to fall is a key thing to check.

The push by countries to build and run their own AI using domestic data and infrastructure. As governments increase spending on data centers and AI models, chip and cloud companies are gaining a new customer group.

Technology that uses qubits, which can be 0 and 1 at the same time, to solve certain problems much faster than conventional computers. It is still early in commercialization, so related stocks often swing on expectations rather than revenue.