OpenAI

Author

Hamish Suter

Overview

Founded
December 2015

Headquarters
San Francisco, California, USA

Founders
Sam Altman, Elon Musk, Ilya Sutskever, Greg Brockman, Trevor Blackwell, Vicki Cheung, Andrej Karpathy, Durk Kingma, John Schulman, Pamela Vagata, and Wojciech Zaremba

CEO
Sam Altman

Model Line
GPT-3.5, GPT-4, GPT-4o

Key Products
ChatGPT, DALL·E, Codex, Whisper

Valuation
$300 billion

Revenue (Jun 9 2025)
$10 billion

Key Partner
Microsoft

Structure
Capped-profit (OpenAI LP)

Valuation and Revenue from (CNBC¹)

This report analyzes OpenAI, the definitive market leader in the generative AI space. While the company faces significant operational costs, intense competition, and regulatory headwinds, its first-mover advantage, superior model performance, and deep integration with Microsoft’s enterprise ecosystem present a powerful, category-defining investment opportunity. The following analysis will break down OpenAI’s value chain, competitive moat, and the core risks and opportunities that will shape its future valuation.

OpenAI is a large language model and artificial intelligence development firm. It works to assist individuals and businesses by automating tasks in the hopes to increase overall productivity through natural language processing. These tasks range from simple math to advanced data analysis. OpenAI captivated the world and turbocharged the “AI race” when it released its LLM chat bot “ChatGpt” in 2022.

Value Chain

graph LR
  %% Define subgraphs
  subgraph "Input"
    A[Hardware]
    B[Energy]
    C[Data]
    style Input fill:#E0E0E0,stroke:#333,stroke-width:2px,color:#000,font-size:24px;
  end

  subgraph "OpenAI"
    D[Model Training and Fine Tuning]
    E[Model Deployment]
    style OpenAI fill:#E0E0E0,stroke:#333,stroke-width:2px,color:#000,font-size:24px;
  end

  subgraph "Output"
    F[Distribution Channels]
    G[End Users]
    style Output fill:#E0E0E0,stroke:#333,stroke-width:2px,color:#000,font-size:24px;
  end

  %% Connections
  A --> D
  B --> D
  C --> D
  D --> E
  E --> F
  F --> G

  %% Style classes
  classDef inputNode fill:#3498DB,stroke:#2C3E50,stroke-width:2px,rx:10,ry:10,color:#fff,font-weight:bold;
  classDef processNode fill:#18BC9C,stroke:#2C3E50,stroke-width:2px,rx:10,ry:10,color:#fff,font-weight:bold;
  classDef outputNode fill:#2C3E50,stroke:#18BC9C,stroke-width:2px,rx:10,ry:10,color:#ECF0F1,font-weight:bold;

  %% Assign classes to nodes
  class A,B,C inputNode;
  class D,E processNode;
  class F,G outputNode;

  %% Edge styling (this depends on Mermaid renderer support)
  linkStyle default stroke:#2C3E50,stroke-width:2px,stroke-linejoin:round;

Inputs

Hardware

Historically, OpenAI has primarily sourced its GPUs from NVIDIA and other GPU sellers. Currently, OpenAI is developing its own custom AI chip, aiming to finalize the design by late 2025 to reduce dependence on suppliers like NVIDIA. TSMC will manufacture these chips using their advanced proprietary 3-nanometer process technology. This move reflects its strategy to gain more control over its AI infrastructure and manage growing computational demands.
(Reuters²)

Energy

In January 2025, OpenAI introduced its Stargate Project. It is a $500 billion plan aimed at developing the world’s most advanced AI infrastructure. The Stargate project will require massive energy resources, leading OpenAI to pursue long-term energy procurement agreements. Its energy needs are projected to rival those of major data centers, underscoring the intensity of AI workloads. Due to the construction of ten new data centers, the project’s energy demands are expected to rise to 15 gigawatts. That’s roughly the same amount of electricity used by a small country like Cuba.
(Certrec³)

Data

OpenAI has formed partnerships with data-rich organizations like Shutterstock, Reddit, and Axel Springer to access diverse and proprietary datasets. These agreements are key to training robust and high-performing AI models. These companies continually bring in new data, allowing AI models to improve performance and evolve to current human trends.
(Foundation Inc.⁴)

Outputs

Distribution Channels

OpenAI distributes its AI products largely through Microsoft, which integrates its models into Azure, Microsoft 365, and GitHub Copilot. This alliance gives OpenAI broad enterprise access and accelerates market penetration through existing Microsoft user bases. OpenAI also has a growing in-house Go-To-Market team to sell and customize its enterprise contracts.
(Carilu⁵)

End Users

Currently, the primary end users of OpenAI are retail consumers, but large enterprises are starting to adopt the advanced technology. For example, PwC will become OpenAI’s largest enterprise customer, deploying ChatGPT to ~100,000 employees for tasks like auditing, consulting, and client communication.
(A. Age⁶)

Product Offerings

GPT-4 / GPT-4o / GPT-3.5

These are large language models designed for processing and generating human-like text. Their applications include conversational AI, assisting with code generation, text summarization, and various natural language processing tasks. They operate by predicting the next word in a sequence based on patterns learned from vast datasets.

GPT-3.5 is a fast and efficient language model, best for everyday tasks like chatting, summarizing, or basic writing.
GPT-4 is a more advanced model with stronger reasoning, better understanding of complex instructions, and has the ability for image input.
GPT-4o is the most powerful and versatile model, capable of real-time text, image, and audio processing, combining speed, accuracy, and multimodal capabilities, making it ChatGPT’s flagship model.

DALL·E

DALL·E is a generative artificial intelligence model focused on image creation. It produces visual content from descriptive text prompts. Its function is to translate linguistic descriptions into corresponding visual representations, a process also known as text-to-image generation.

Whisper

Whisper is an automatic speech recognition (ASR) system. Its primary function is to convert spoken audio into written text. It is designed to operate across multiple languages and aims for high accuracy in transcription a.k.a speech-to-text.

Embeddings API

This API provides numerical vector representations of text. These embeddings capture semantic relationships between words and phrases, making them useful for tasks such as facilitating search operations, organizing data into clusters, and powering recommendation systems based on content similarity.

Moderation API

The Moderation API is a tool for content filtering. It is designed to automatically identify and flag text content that may be harmful or inappropriate, aiming to support safer digital interactions by filtering user-generated text.

Sora

Sora is a multimodal generative model capable of creating and editing video content. It operates by interpreting text prompts to generate dynamic visual sequences, extending OpenAI’s generative capabilities beyond static images into text into video generation.

Example Use Case - Embeddings API

A global financial services firm has transformed its wealth management mobile app by integrating OpenAI’s API to power an intelligent virtual financial assistant, now used by thousands of high-net-worth clients.

Previously, clients relied on static dashboards, delayed analyst commentary, and manual advisor outreach to interpret market changes and portfolio impacts. With GPT-4 integrated via OpenAI’s API, they can now ask sophisticated, personalized questions in natural language such as, “What impact did last week’s Fed rate hike have on my bond holdings?” or “What are the key risks in my tech sector exposure this quarter?”

The assistant supports dynamic follow-up interactions. When a client asks, “Should I rebalance my portfolio?”, GPT-4 offers a risk analysis and prompts a conversation with their advisor, linking directly to a scheduling tool. It can also generate visualizations, earnings summaries, or ESG insights by pulling data from internal systems and formatting responses through natural language generation.

While GPT-4 enhances wealth management apps with fast, personalized insights, human financial advisors still play a crucial role. They offer emotional intelligence, strategic guidance, and a deep understanding of clients’ goals, especially during life transitions or market volatility, where empathy and trust are essential.

Advisors also help with long-term planning, behavioral coaching, and navigating complex decisions that go beyond data. Rather than replacing advisors, GPT-4 frees them to focus on deeper, higher-value conversations, making their role even more important in a tech-augmented experience.

This integration shifts the experience from a passive client dashboard to a proactive, always-available advisory service. Clients stay better informed and engaged in real time, without waiting for scheduled reviews, while advisors can focus on more strategic, high-value discussions.

Revenue Source Breakdown

Percentage of total revenue from individual consumers vs. businesses collected by (Bloomberg⁷)

Individual Consumers:

ChatGPT Plus

A subscription plan offering enhanced access to GPT-4, faster response times, and priority availability during high traffic.

Usage-Based Fees

Pay-as-you-go pricing for API and tool usage, suitable for users who prefer flexibility without a monthly commitment.

B2B:

ChatGPT Enterprise

This version offers strengthened security, high performance, unlimited GPT-4 access, data analysis capabilities, administrative tools, single sign-on (SSO), and enterprise-level privacy and support.

API Subscriptions

Scalable APIs that allow businesses to integrate GPT models into their applications and workflows. These subscriptions offer flexible pricing, fine-tuning options, and enterprise features such as usage tracking and service level agreements (SLAs).

Custom Solutions

AI services tailored to specific business requirements. These can include model customization, integration of proprietary data, and the development of AI copilots.

Regional Usage

The above graph represents the percentage of people surveyed in BCG’s CCI Global Consumer Sentiment survey that use ChatGPT in 22 countries (Boston Consulting Group⁸).

ChatGPT usage varies widely by region. Asia shows the broadest range, with India leading globally (45%) and countries like Thailand much lower (14%). Africa has high engagement, especially in Morocco and South Africa. South America also sees strong usage in Brazil and Argentina (both 32%). In contrast, Europe tends to be more cautious, with lower adoption across countries like France and Germany. North America and Oceania sit in the mid-range (22–23%), while the Middle East shows mixed results—UAE is high (34%), but others like Saudi Arabia are much lower.

This fragmented adoption map highlights both a risk (the need for complex, localized go-to-market strategies) and a significant opportunity for growth in currently underserved, high-potential regions.

Competitors

Claude

Claude competes with OpenAI’s GPT-4 by prioritizing safety, alignment, and controllability. Developed by former OpenAI researchers at Anthropic, Claude is positioned as a more predictable and steerable alternative to GPT, particularly appealing to enterprises concerned about AI behavior. Techniques like Constitutional AI give it a differentiation point in the growing demand for trustworthy AI.

Deepseek

Deepseek challenges OpenAI in the domain of intelligent search and information retrieval. While OpenAI integrates its models into tools like ChatGPT and Microsoft’s Copilot, Deepseek focuses on tailored, context-aware search experiences. It competes by offering highly relevant results and enterprise-ready tools that use LLMs to enhance discovery across unstructured data—an area OpenAI is expanding into with features like advanced data browsing.

Google DeepMind

DeepMind competes directly with OpenAI at the cutting edge of AI research, particularly in areas like reinforcement learning, multimodal AI, and scientific applications. Its Gemini models aim to rival or surpass OpenAI’s GPT series by integrating advanced reasoning and tool use. DeepMind’s access to Google’s massive compute and data gives it a significant infrastructure advantage, making it a formidable competitor in both foundational research and applied AI.

Grok

Grok-style assistants compete with OpenAI’s conversational AI by focusing on domain-specific guidance and problem-solving, particularly in educational or support contexts. While OpenAI emphasizes general-purpose assistance via ChatGPT, Grok-like systems often target narrower use cases with tailored knowledge, making them competitive in focused verticals such as tutoring or internal enterprise support.

Hugging Face

Hugging Face competes with OpenAI by fostering an open-source alternative to closed model ecosystems. Its platform allows developers to access, fine-tune, and deploy a wide range of models, including some that rival GPT-class capabilities. While OpenAI focuses on proprietary systems, Hugging Face emphasizes transparency, openness, and customizability—appealing to researchers, startups, and enterprises looking for more control.

Microsoft

While Microsoft is a strategic partner and investor in OpenAI, it also competes indirectly by embedding OpenAI’s technology into its own ecosystem. Through Azure, Microsoft has the infrastructure to train and serve models at scale, and it could potentially develop independent models or platforms in parallel. Its deep integration of GPT into tools like Copilot also means it competes in user experience and enterprise AI deployment.

Perplexity

Perplexity AI directly competes with OpenAI by reimagining search with LLMs. Unlike traditional search engines or ChatGPT’s browsing capabilities, Perplexity delivers concise, cited answers pulled from multiple sources. It positions itself as a hybrid between search and chat, providing real-time, transparent answers—something OpenAI users currently rely on through plugins or custom browsing.

Poe

Poe competes with OpenAI not through model development but through aggregation and interface. By offering access to multiple LLMs, including OpenAI’s, Poe positions itself as a neutral platform focused on user experience and versatility. It reduces vendor lock-in and gives users a comparative view across AI systems, indirectly challenging OpenAI’s dominance as a single-model provider.

Website traffic to the major AI players. The gap on the graph is due to the data not being tracked during the holidays. Numbers from (Similarweb⁹)

OpenAI dominates consistently, making up the vast majority of daily visits (over 150 million by May). Perplexity and Google hold a distant second but are steadily growing. Smaller but rising platforms include Claude, Microsoft, and Deepseek, all showing slight increases over time. Meta, Grok, and Huggingface have modest but stable usage. All other (smaller platforms combined) and newer entrants like Poe and Deepseek contribute only a small share. Overall, AI platform traffic is climbing steadily, passing 200 million daily average visits by mid-May 2025.

SWOT Analysis

Strengths

Leading foundational models
Large user base and brand recognition
Strong enterprise adoption
Integration with Microsoft Azure

OpenAI’s GPT-4o, GPT-4o mini, and GPT-3.5 turbo compared against Gemini Flash and Claude Haiku in evaluation benchmarks. These benchmarks measure each model’s ability to perform general tasks from math to biology. Scores compiled by (Writingmate¹⁰)

Weaknesses

High compute and energy costs
Dependency on chip suppliers (e.g., NVIDIA)
Limited regional compliance in some markets

Opportunities

Expansion into video (Sora)
Enterprise AI transformation
Global educational/training markets
Customizable AI agents and agents API
Move towards building an in-house, AI-powered device (T. Verge¹¹)

Threats

Highly competitive landscape
Regulatory scrutiny (AI safety, copyright, data use)
Dependency on infrastructure (data centers, power)
Open-source alternatives gaining traction
Controversies with Sam Altman, Elon Musk, and antitrust lawsuits with Microsoft (W. S. Journal¹²)
Risk of Commoditization: A fundamental threat is the potential for generative AI to become a low-margin, utility-like service

User Comparative Testing

OpenAI is currently the market leader in the artificial intelligence industry, boasting the highest-performing models, strongest brand recognition, and the most visited website in the space. Their market penetration has been both rapid and substantial.

While creating this report, I put ChatGPT up against its primary competing AI models. I tested ChatGPT for ideation and framework development, as well as to generate visually appealing graphs. However, I also tested Google Gemini and Anthropic Claude. Personally, I found Gemini the easiest to work with. It generally followed instructions well, and when it didn’t, it was straightforward to correct.

ChatGPT was my second choice. It’s capable of handling more complex tasks, like generating downloadable CSV files, but often overcomplicates simpler tasks and can be harder to reason with.

Claude ranked just behind ChatGPT. It didn’t stand out in any particular way and had similar limitations with complex tasks, though it was slightly more cooperative when refining output.

Looking Ahead

Looking ahead, the future of AI leadership is uncertain. It’s unclear whether there will be a singular winner or a fragmented field. I believe the AI race is trending toward a race to the bottom in terms of profitability. Competitors will continue eroding each other’s lead by pouring money into more data and fine-tuning, chasing marginal gains.

For AI companies to justify this level of unprofitability, they’ll likely need to build ancillary businesses around their models or position AI itself as the ancillary product. A fitting analogy is the train industry: if operators charged riders enough to turn a profit, demand would collapse as people switch to other transportation methods. The only way trains stay viable is through subsidies or external revenue streams. For instance, Hong Kong’s MTR turns a profit not from train fares but by owning the malls and apartments at its stations.

This analogy highlights a critical strategic imperative for OpenAI: to avoid commoditization, it must build a defensible economic moat beyond the models themselves. Its success will depend on creating a surrounding ecosystem, as evidenced by its strategic move into hardware with the Jony Ive collaboration. The key question for investors is whether this ecosystem can be built faster than competitors like Google and Meta can leverage their own vast, existing platforms.

OpenAI seems to understand this challenge. Its $6.4 billion acquisition of Jony Ive’s AI hardware startup signals a broader effort to build a surrounding business ecosystem (T. Verge¹³). However, it remains unclear whether OpenAI can build something large and defensible enough when compared to giants like Google and Amazon. It still has the lights on, but how long can that last once investors start expecting real returns.

References

CNBC (2025). Retrieved from https://www.cnbc.com/2025/06/09/openai-hits-10-billion-in-annualized-revenue-fueled-by-chatgpt-growth.html

Reuters (2025).OpenAI to develop custom AI chips to reduce dependence on NVIDIA

Certrec (2025).Energy demands for OpenAI stargate project Retrieved from https://www.certrec.com/blog/energy-demands-for-openai-stargate-project/

Foundation Inc. (2024).OpenAI partnerships list Retrieved from https://foundationinc.co/lab/openai-partnerships-list/

Carilu (2024).Hypergrowth marketing playbook: The OpenAI example Retrieved from https://www.carilu.com/p/hypergrowth-marketing-playbook-the

A. Age (2024).PwC to become OpenAI’s largest enterprise customer Retrieved from https://www.accountancyage.com/2024/05/30/pwc-openai-largest-enterprise-customer

Bloomberg (2024).OpenAI CFO: 75% of revenue comes from paying consumers Retrieved from https://www.bloomberg.com/news/articles/2024-10-28/openai-cfo-says-75-of-its-revenue-comes-from-paying-consumers

Boston Consulting Group (2024).Consumers know more about AI than businesses think Retrieved from https://www.bcg.com/publications/2024/consumers-know-more-about-ai-than-businesses-think

Similarweb (2025).Generative AI tools site traffic Retrieved from https://www.similarweb.com/website/generative-ai.tools

Writingmate (2025).ChatGPT vs gemini vs claude comparison Retrieved from https://writingmate.ai/blog/chat-gpt-gemini-claude

T. Verge (2025).OpenAI and jony ive working on AI hardware Retrieved from https://www.theverge.com/news/671838/openai-jony-ive-ai-hardware-apple

W. S. Journal (2025).OpenAI and microsoft tensions boil over Retrieved from https://www.wsj.com/tech/ai/openai-and-microsoft-tensions-are-reaching-a-boiling-point-4981c44f