After GPT-3.5 launched and gained popularity, OpenAI decided on a $20 fee we would all pay for a premium subscription to get all the generative AI access we want. What went into these calculations? Unlikely much. It was a standard fee OpenAI executives decided would go well for a subscription-based product. Others like Claude and Perplexity were forced to follow suit by sticking with the $20 number as now nobody would pay a penny more. Except, it turns out that users really wanted to make a bang for the buck when paying for the subscription and with the usage, the GPU costs have been running too high — making the 20 bucks subscription loss-making for generative AI companies. OpenAI sought to correct this with a whopping $200 subscription that gave high-flying access to its top-tier models like o1 and o3 and AI agents like Operator and Deep Research. Except, Sam Altman revealed that the usage even on this tier is so high that OpenAI simply can’t make a profit. OpenAI set the bar and everyone else was forced to follow suit. The DeepSeek Surprise In all this, DeepSeek revealed a surprise — that it expects a theoretical profit margin of 545%. It’s theoretical because DeepSeek isn’t accounting for charging less during non-peak hours, differences in its models, and the fact that it’s providing a whole lot of access via the web interface on chat.deepseek.com and the mobile apps. So, how is DeepSeek able to turn a profit? Well, because it claims to be using GPUs that cost a lot less in a lot less quantity. It also carries a much smaller team — about 200 estimated against say OpenAI’s 2,000+ employees. Anyway, it’s a bit moot to talk about DeepSeek, owing to trust issues with how information from China usually comes about. Nevertheless, it’s the most exciting project around on two fronts — its breakthrough in reasoning model that remains unparalleled and its contribution to indiehacker and open-source ethos. What’s the right price to pay for AI? The real value for any company or industry comes about with an answer to a simple question: Do the users want to pay for your product or service? Venture Capitalists are pouring billions and billions into AI development with the hope for glorious returns. Leaders in the space like OpenAI struggling to strike the balance between costs and revenue is not a good sign. Some industries like Web3 are forever stuck in this “budding industry” zone and AI won’t want to be in this company. So, how would the profit come about? Sam is now toying with the idea that we perhaps pay as we go for what we are using. I am not sure where that proposal is coming from but I expect that to be a disaster if implemented. This isn’t dissimilar to how we pay for OpenAI APIs when building our own apps but counting tokens is just not the way people want to use an end-consumer AI chat app. The ‘Heavy User’ Burden The flat-free model currently implemented may be hurting right now but may just work out long-term. Why you ask? Because the early adopters are also likely to be the heaviest users. For a flat subscription fee to pay dividends, the companies need more occasional queriers and fewer AI nerds like me. A better solution would possibly be to keep the $20 subscription with existing unlimited base-models access (GPT-4o, o3-mini, o3-mini-high, etc.) and give limited premium features such as 5 Deep Research usage or 3 GPT-4.5 prompts, with ability to pay for more as you go. A rudimentary analysis of how different pricing models would stack up for OpenAI, generated with a custom script in Python. Pay-per-use makes a lot of sense with a heavy skew toward power users, which is the likely scenario at the time. With wider adoption and a changing skew, you are likely to get better results with a flat fee structure. Charge $20 plus extra for added usage, and you have the optimal scenario — both for the user and the company. At least, that is what the trend seems to be right now. That’s the thought I leave you with today. What’s New? Claude's parent Anthropic has raised another $3.5 billion in a funding round that values it at a whopping $61.5 billion!
OpenAI is bringing its most “emotionally intelligent” model GPT-4.5 to Plus users.
Deutsche Telekom is launching an “AI phone” in partnership with Perplexity. (I expect it to be nothing of note.)
Amazon may be working on its own hybrid-reasoning model, similar to what was launched by Claude recently in Sonnet 3.7.
Key Nvidia partner CoreWeave, which is on track for an IPO, has acquired AI developer platform Weights & Biases. After GPT-3.5 launched and gained popularity, OpenAI decided on a $20 fee we would all pay for a premium subscription to get all the generative AI access we want. What went into these calculations? Unlikely much. It was a standard fee OpenAI executives decided would go well for a subscription-based product. Others like Claude and Perplexity were forced to follow suit by sticking with the $20 number as now nobody would pay a penny more. Except, it turns out that users really wanted to make a bang for the buck when paying for the subscription and with the usage, the GPU costs have been running too high — making the 20 bucks subscription loss-making for generative AI companies. OpenAI sought to correct this with a whopping $200 subscription that gave high-flying access to its top-tier models like o1 and o3 and AI agents like Operator and Deep Research. Except, Sam Altman revealed that the usage even on this tier is so high that OpenAI simply can’t make a profit. OpenAI set the bar and everyone else was forced to follow suit. The DeepSeek Surprise In all this, DeepSeek revealed a surprise — that it expects a theoretical profit margin of 545%. It’s theoretical because DeepSeek isn’t accounting for charging less during non-peak hours, differences in its models, and the fact that it’s providing a whole lot of access via the web interface on chat.deepseek.com and the mobile apps. DeepSeek chat.deepseek.com So, how is DeepSeek able to turn a profit? Well, because it claims to be using GPUs that cost a lot less in a lot less quantity. It also carries a much smaller team — about 200 estimated against say OpenAI’s 2,000+ employees. So, how is DeepSeek able to turn a profit? Anyway, it’s a bit moot to talk about DeepSeek, owing to trust issues with how information from China usually comes about. Nevertheless, it’s the most exciting project around on two fronts — its breakthrough in reasoning model that remains unparalleled and its contribution to indiehacker and open-source ethos. What’s the right price to pay for AI? The real value for any company or industry comes about with an answer to a simple question: Do the users want to pay for your product or service? Venture Capitalists are pouring billions and billions into AI development with the hope for glorious returns. Leaders in the space like OpenAI struggling to strike the balance between costs and revenue is not a good sign. Some industries like Web3 are forever stuck in this “budding industry” zone and AI won’t want to be in this company. So, how would the profit come about? Sam is now toying with the idea that we perhaps pay as we go for what we are using. I am not sure where that proposal is coming from but I expect that to be a disaster if implemented. This isn’t dissimilar to how we pay for OpenAI APIs when building our own apps but counting tokens is just not the way people want to use an end-consumer AI chat app. The ‘Heavy User’ Burden The flat-free model currently implemented may be hurting right now but may just work out long-term. Why you ask? Because the early adopters are also likely to be the heaviest users. For a flat subscription fee to pay dividends, the companies need more occasional queriers and fewer AI nerds like me. A better solution would possibly be to keep the $20 subscription with existing unlimited base-models access (GPT-4o, o3-mini, o3-mini-high, etc.) and give limited premium features such as 5 Deep Research usage or 3 GPT-4.5 prompts, with ability to pay for more as you go. A rudimentary analysis of how different pricing models would stack up for OpenAI, generated with a custom script in Python. Pay-per-use makes a lot of sense with a heavy skew toward power users, which is the likely scenario at the time. With wider adoption and a changing skew, you are likely to get better results with a flat fee structure. Charge $20 plus extra for added usage, and you have the optimal scenario — both for the user and the company. At least, that is what the trend seems to be right now. That’s the thought I leave you with today. What’s New? Claude's parent Anthropic has raised another $3.5 billion in a funding round that values it at a whopping $61.5 billion! OpenAI is bringing its most “emotionally intelligent” model GPT-4.5 to Plus users. Deutsche Telekom is launching an “AI phone” in partnership with Perplexity. (I expect it to be nothing of note.) Amazon may be working on its own hybrid-reasoning model, similar to what was launched by Claude recently in Sonnet 3.7. Key Nvidia partner CoreWeave, which is on track for an IPO, has acquired AI developer platform Weights & Biases. Claude's parent Anthropic has raised another $3.5 billion in a funding round that values it at a whopping $61.5 billion! raised OpenAI is bringing its most “emotionally intelligent” model GPT-4.5 to Plus users. bringing Deutsche Telekom is launching an “ AI phone ” in partnership with Perplexity. (I expect it to be nothing of note.) AI phone Amazon may be working on its own hybrid-reasoning model, similar to what was launched by Claude recently in Sonnet 3.7. may be Key Nvidia partner CoreWeave, which is on track for an IPO, has acquired AI developer platform Weights & Biases. acquired

The is an opinion piece based on the author’s POV and does not necessarily reflect the views of HackerNoon.

Amazon

Deutsche Telekom

Why I Stopped Giving $20 A Month To ChatGPT

Google is Secretly Building an Edge in the AI Race

Subscribe Now

Dzambhala Finance

TushitaAI

About Author

Read My Stories

Too Long; Didn't Read

How Long Can AI Companies Maintain a $20 Monthly Subscription Fee?

How Long Can AI Companies Maintain a $20 Monthly Subscription Fee?

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

AI Agents Need To Come With An Emergency Button

The Noonification: The Battle Between Proprietary and Open Source AI (11/3/2023)

The Noonification: Breaking Axioms in Program Execution (10/29/2023)

A Data Scientist's Guide to Simplistic Time-Series Models

Aitana Unveiled: A Spanish Symphony of Innovation in the AI Revolution

Beyond the Algorithm: How Training Data Can Make or Break a Generative AI Model

AI Agents Need To Come With An Emergency Button

The Noonification: The Battle Between Proprietary and Open Source AI (11/3/2023)

The Noonification: Breaking Axioms in Program Execution (10/29/2023)

A Data Scientist's Guide to Simplistic Time-Series Models

Aitana Unveiled: A Spanish Symphony of Innovation in the AI Revolution

Beyond the Algorithm: How Training Data Can Make or Break a Generative AI Model

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps