Why Most AI Startups Are BAD Businesses

It's never been easier to launch apps, with individuals who previously lacked coding skills now releasing AI applications and securing venture capital. However, the underlying economics of many of these AI software businesses reveal a less optimistic picture. While traditional B2B software boasts margins of 70-90%, AI-native SaaS, particularly those wrapping large language models, typically operates with significantly lower margins, ranging from 30-60%. This discrepancy is largely due to the substantial ongoing costs associated with AI, such as API calls, compute time, and licensing, which contrast sharply with the near-zero marginal cost of serving additional customers in traditional software.

The current AI landscape, especially for generative AI, is riding a wave of hype, which can mask underlying economic challenges. Many companies are seeing revenue from "dead subs"—users who pay but don't use the service—a practice that is unlikely to be sustainable. This trend mirrors past tech bubbles like blockchain and augmented reality, where initial excitement overshadowed a lack of widespread, monetizable use cases. Even leading AI companies face profitability hurdles, with high operational costs, such as daily expenses that can run into hundreds of thousands of dollars for services like ChatGPT, leading to situations where individual users can cost more than their subscription revenue.

For an AI business to achieve sustainability and profitability, it's crucial to focus on core value that exists even without the AI component. The most promising AI businesses are often those that integrate AI features into traditional software or tackle complex, industry-specific problems, particularly in text-heavy sectors like legal or accounting, rather than relying solely on AI novelty. Ultimately, the metrics that matter for long-term success are retention and conversion rates, which will endure long after the current AI hype cycle fades.

• Traditional B2B software has high margins (70-90%), while AI-native SaaS has lower margins (30-60%).
• AI businesses face significant ongoing costs per user, including API calls, compute time, and licensing, unlike traditional software.
• Many AI apps rely on "dead subs" (paying but not using) and are currently riding a hype wave, similar to past tech trends.
• Leading AI companies incur substantial daily operational costs, sometimes exceeding individual user revenue, necessitating usage controls.
• Sustainable AI businesses often integrate AI into existing traditional software or solve complex, industry-specific problems, with retention and conversion rates being key metrics for success.

Related Video Summary

• How Developers Find $1M Problems Everyone Ignores
• Rich People Are Quietly Preparing For This - And You Should Too
• Beat the 95%: Why AI Projects Fail—And How Builders Win
• I copied TikTok trends and made millions with AI

Traditional B2B Margins [00:36]

• Traditional B2B software margins range between 70% and 90%.
• Best-in-class SaaS applications achieve around 80% or higher margins.
• This high scalability is due to software businesses having near-zero marginal cost to serve additional customers after the platform is built.

This section highlights the exceptionally high profit margins common in traditional software businesses, attributing this to their inherent scalability and minimal additional cost per customer once the initial investment is made.

Traditional B2B margins range between 70 and 90%. With best-in-class SAS applications achieving around 80% or higher.

AI-Native SAS Margins [01:28]

• AI-native SaaS, particularly GenAI and LLM wrappers, have margins of 30-60%.
• 60% is considered best-in-class for these products.
• Even mature GenAI products like Anthropic's Claude are at 55% margins.

This part contrasts the profitability of traditional SaaS with AI-native applications, revealing significantly lower margin expectations for AI-centric businesses.

Genai native SAS and especially LLM rappers which the vast majority of them are have 30 to 60% margins with 60% being best-in-class products.

Dead Subscriptions [02:01]

• A dead subscription occurs when a user pays but does not use the service.
• This is particularly common when users purchase access in bundles, like bundles of AI tools.
• GenAI products are currently benefiting from dead subs due to the hype cycle, but this is not sustainable.

This section explains a revenue stream that currently inflates the appearance of success for some AI companies but is acknowledged as a temporary phenomenon.

A dead sub is when a user is paying but not using.

The Hype Cycle [02:31]

• GenAI is currently at the peak of the hype cycle.
• Being at the peak of the hype cycle has no correlation with long-term benefits.
• Past trends like blockchain and augmented reality also experienced similar hype cycles with limited sustainable use cases.

The speaker draws parallels between the current AI enthusiasm and previous technological trends, cautioning that peak hype does not guarantee lasting value or profitability.

Now this is where Genai is at the moment. It is very common among tech trends that are reaching the peak of the hype cycle.

Stark Margin Differences Explained [03:53]

• In traditional SaaS, after initial R&D, serving more customers adds very little extra cost.
• GenAI-native products have major ongoing costs per user: API calls, compute time, licensing, and moderation.
• These costs can become exponential in GenAI products.

This section delves into the fundamental reasons behind the significant difference in profit margins between traditional software and AI-native applications, highlighting the inherent cost structures.

In traditional SAS after initial R&D and platform investment serving more customers adds very little extra cost.

High Operational Costs for AI [04:06]

• GenAI native or GPT wrapper products incur substantial ongoing costs per user.
• These costs include API calls, compute time, licensing, and output moderation.
• Companies may need to cap usage to manage these escalating costs.

This part focuses on the direct costs associated with running AI services, illustrating how these expenses can quickly add up and impact profitability.

In Gai native or GPT wrapper products there are major ongoing costs per user API calls compute time licensing sometimes per output moderation.

Examples of Costly AI Services [04:44]

• ChatGPT cost OpenAI $700,000 a day in 2023, with daily costs ranging from $100,000 to several hundred thousand in the current year.
• Prolific AI users can individually cost OpenAI over $200 a month, leading to usage metering even at premium tiers.
• GitHub Copilot cost Microsoft almost $30 a month per user, with power users costing around $80 per developer per month.
• Midjourney's low-cost plan had strict image generation limits due to significant GPU resources consumed per image.

This section provides concrete examples of how expensive it is to operate AI services, demonstrating that costs can easily outstrip revenue, even for well-known products.

For example Chad GPT cost OpenAI $700,000 a day in 2023.

The Paradox of Usage Controls [06:35]

• When AI companies implement usage controls to manage costs, users dislike the experience.
• This creates a paradox: unmetered plans lead to runaway costs from heavy users, while metered plans can alienate customers.

This segment discusses the difficult balancing act AI companies face between managing high operational costs and providing a user experience that doesn't deter customers.

But the paradox is that when they do put in usage controls, the users don't like the experience.

Low Conversion Rates to Paid Plans [07:07]

• ChatGPT has a user base of approximately 700 million to 800 million users.
• Only about 10 million users are on the paid ChatGPT Plus plan, representing less than a 2% conversion rate.
• Even alternative reports suggesting 5-7% conversion are considered alarmingly low for a major disruptor.

This section presents data on the low conversion rate of free users to paid subscribers for a leading AI product, indicating a potential lack of product-market fit in terms of paid adoption.

Now of those 7 to 800 million users, how many of them are paid? 10 million on Chad GPT plus plan.

Vanity Metrics in AI [08:36]

• An increase of 100 million weekly active users for ChatGPT was reported, while site traffic remained unchanged.
• The use of "weekly" active users instead of the more common "monthly" metric is questioned.
• These metrics might be vanity metrics, offering large numbers without necessarily translating to revenue.

The speaker critiques the reporting of certain AI usage statistics, suggesting they might be inflated or misleading, focusing on impressive-sounding numbers rather than genuine business impact.

Personally, I think that's getting a little bit into the territory of vanity metrics.

API Revenue and Market Share [08:49]

• API revenue data suggests AI is not currently positioned to replace all human roles.
• ChatGPT leads with 300-400 million monthly active users, significantly outpacing competitors like Claude (3 million) and Gemini (47 million, boosted by Google).
• Copilot has 33 million users, boosted by Microsoft's reach.

This section provides comparative user numbers for major AI platforms, emphasizing ChatGPT's dominant market share and hinting at its limited revenue generation from free users.

At least not right now. And now to put this in perspective, let's look at OpenAI's biggest competitors.

The Unsustainable Economics of Foundational AI [10:55]

• The economics of foundational AI companies like OpenAI and Anthropic are currently "royally unprofitable" and unsustainable.
• As LLMs become a commodity, drastic changes in pricing, bundling, and growth strategies are expected.
• The current situation is a "rat race on the way to profits" for these foundational model creators.

This part offers a critical assessment of the financial viability of the companies building the core AI technologies, deeming their current business models unsustainable.

The economics of foundational AI companies like OpenAI or Anthropic are royally unprofitable and the way they exist today.

AI Price Wars and Expansion Strategies [12:13]

• The AI SaaS space is experiencing rapid price wars, further squeezing margins, especially for wrapper applications.
• AI companies are likely to expand product lines and upsell users more aggressively to compensate for squeezed margins.
• As the initial fascination fades, companies and investors will focus more on the actual economics of AI apps.

This section discusses the competitive landscape and the strategies AI companies are employing to survive financially, including aggressive pricing and product expansion.

The AI SAS space is experiencing rapid price wars, further squeezing margins as much as possible, especially for rappers and tools layered on top of foundational models.

Identifying Sustainable AI Businesses [13:32]

• A good litmus test is a traditional SaaS startup that incorporates AI features for automatable tasks.
• These companies should be benchmarked against traditional SaaS metrics, not inflated AI metrics.
• The product's value should be demonstrable even without AI; AI should enhance, not define, the product.

This segment provides guidance on how to identify AI companies with genuinely sustainable business models, emphasizing a focus on core value and traditional business metrics.

A traditional SAS startup with AI features for tasks that can be automated.

Promising AI Business Models [14:26]

• The most successful GenAI startups often work with large amounts of text-based data in industries like accounting, HR, sales, or legal.
• These businesses automate complex workflows by integrating data from various sources (e.g., contracts with invoices).
• They can be sold as standalone products, plugins, or APIs, offering sustainable revenue, though not necessarily billions.

This part highlights specific areas where GenAI startups are finding practical application and building viable businesses, focusing on automation of complex data-driven tasks within industries.

The vast majority of genai startups that work are the apps that work with large amount of textbased data and documents in various industries be accounting, HR, sales or legal.

Solving Difficult, "Boring" Industry Problems [15:12]

• The most challenging AI problems are often found in lengthy, effort-consuming tasks within traditionally "boring" industries.
• Legacy enterprise tech offers opportunities for those whose jobs might be automated by AI.
• Solving prevalent problems in industries like accounting, legal, or pharma could lead to building a unicorn company.

This section points to the significant opportunities in tackling deeply ingrained, complex problems within established industries, suggesting these are where true innovation and value creation lie, rather than superficial AI applications.

The most difficult problems that I personally haven't seen any AI startup solve yet are those lengthy, effort consuming, old school problems in very boring industries.

The True Measure of AI Value [17:06]

• Getting users to try an AI product with grand promises or FOMO is easy; delivering tangible value is difficult.
• The true indicators of value are retention and conversion rates, which will outlast hype cycles and VC rounds.
• If an app's core problem-solving ability doesn't rely on AI, users are likely safe from AI-driven job displacement in the long term.

The conclusion emphasizes that genuine value delivery, measured by retention and conversion, is the ultimate determinant of an AI product's success, not the hype surrounding it.

What's really difficult is to deliver value. Value that a few can replicate.