8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

The AI Bubble. A Reality I Just Realized in the

GenAI Landscape
Jing Hu · Follow
Published in AI Advances
7 min read · Aug 6, 2024

Listen Share More

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 1/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Y ou heard AI everywhere. You know that billions are invested in AI. But here’s
the thing: while some AI companies are raking in billions, others are
struggling to keep the lights on.

The most recent example, Character.AI — it’s a good highlight of what’s happening
in the AI world right now. The industry is in a paradoxical moment:

On one hand, we’re witnessing record-breaking investments in AI startups, from the

LLM layer to the application layer. On the other hand, high-profile exits and struggles

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 2/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

of once-promising AI companies are raising eyebrows across the tech world.

What Happened To Character.AI?

Google has hired the founders of chatbot maker Character.AI and struck a deal to license
its models, the latest in a series of deals in which Big Tech companies have scooped up
teams and technology from promising start-ups.

Noam Shazeer and Daniel De Freitas, who co-founded Character.AI in 2022, will join
Google along with a number of other staff from the start-up. Shazeer, who has had two
previous stints at Google, will join its AI research division, DeepMind, which develops
large language models and the Gemini suite of AI products such as chatbots…

… The majority of Character.AI’s team will remain at the start-up where they “will
continue to build the Character.AI product and serve our growing base of users”, the
company said in a blog post on Friday. — FT

This development is particularly striking given that Shazeer and De Freitas had
originally left Google in November 2021, dissatisfied with the company’s
bureaucratic culture. They founded Character.AI with the vision of creating
personalized super-intelligence. Now, just three years later, they’re returning to
their former employer.

The deal values Character.AI today at $2.5 billion, a significant drop from the $5
billion valuation discussed with investors last year.

This marks the third instance in just five months where an AI startup CEO has
jumped ship to a tech giant, following similar moves by leaders from Inflection and
Adept.

As part of the agreement, Google will license Character.AI’s language model and
provide additional funding to support the company’s growth.

This situation exemplifies the broader challenges facing the AI industry,

particularly in the realm of AI applications like AI companions. Despite high user
engagement, these companies are struggling to maintain investor interest and
achieve sustainable business models.

Character.AI’s Struggle: Trending ≠ Funding

High user engagement vs. low investor interest

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 3/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

AI companions offer personalized AI-driven interactions.

For example, you can engage with historical personalities, craft personalized
chatbots, get a personalized teacher to teach you a new language, etc.

Compare traffic between character.AI and other popular AI applications.

Yes, these products have shown impressive user engagement metrics. Or else you
can say in the Google trend above that Character.AI is much more popular
compared to other applications.

In May, AI companions saw a total of 432 million visits, with a year-on-year growth
of 13.87%, second only to chatbots in terms of growth rate.

Character.AI exemplifies this trend.

At its peak, the company boasted 4 million mobile users and 14.8 million monthly
website visitors, reaching up to one-fifth of Google’s search query volume. In May, it
recorded 277 million visits, an 18.89% month-on-month increase. —
https://whatsthebigdata.com/character-ai-statistics/

However, this impressive user engagement has not translated into investor
enthusiasm.

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 4/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

While AI search engines like Perplexity AI have secured large funding rounds at
increasing valuations (reaching $3 billion in its latest round), AI companion
companies are struggling to raise capital.

Character.AI’s Funding Challenges

Character.AI’s funding journey illustrates the challenges faced by AI companion
startups. Despite its strong user metrics, the company has not secured new funding
since March 2023, when it raised $150 million at a $1 billion valuation.

Soon, Character.AI engaged in another negotiation with multiple investors but failed
to close a new round. This funding drought is particularly concerning given the
company’s high operational costs due to its large-scale user interactions.

The stark contrast between Character.AI’s user growth and its funding struggles
suggests that the mobile internet era’s strategy of raising capital based on user
metrics may not be as effective in the AI era.

Why Is It Hard for AI Companions to Profit?

AI companions face a unique commercialization challenge. Their current business
model, primarily based on subscription services, creates a paradox:

AI Companions Depend On IP That They Do Not Own

Character.AI’s traffic relies heavily on IP from games and anime characters,
accounting for 70% of interactions.
However, the company doesn’t own these IPs, limiting its ability to monetize this
engagement effectively.

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 5/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

The Subscription Revenue Barely Covers Its Token Consumption

The low subscription fee is insufficient to cover the costs associated with heavy
users.

Estimates suggest that Character.AI’s monthly inference costs could be around

$20-$30 million. This estimation is based on 200 million visits per month, and users
spend an average of 29 minutes per visit with the chatbot.

Note.

I started by trying to calculate the cost of the token based on the monthly visit and
usage. Thanks to the comment @Eduard Ruzga I realized it wasn’t too wise, given
not enough data on hands. For example, if I am diving into the details of the cost, I
can’t just look at the usage and completely ignore the training and the characters’ IP
cost.

Hence the best I can do (for now) is to highlight the 5000ft high cost of running a
company like Character.ai:

1. Infrastructure: Expenses for server hosting, data management, and processing

power to handle large-scale AI training, fine-tuning, and deployment.

2. Licensing and Data Annotation: Fees for acquiring or licensing datasets used to
train models, as well as for characters’ IP.

3. R&D and Development: Salaries for AI researchers, data scientists, and

engineers developing new models and algorithms. Software Development: Costs
related to building and maintaining the platform.

4. Operational Costs, such as customer Support: Maintaining a team to handle user

inquiries and issues.

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 6/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

5. Marketing and Sales (extremely expensive for B2C products like character.ai):
Costs for customer acquisition, advertising, etc.

6. Compliance and Legal: It is hard to say how much these AI companies invest in
legal and compliance.

7. General Overhead, eg., Office Space, administrative (Accounting, HR, Legal, and
others)

While its annual revenue in the previous year was only about $15.2 million.

This is a lose-lose scenario: Heavy users consume substantial resources without

proportional revenue, while light users are more likely to churn to apps like
ChatGPT.

Challenges in Broad AI Application Monetization

Limited Revenue Streams for AI Startups
We just covered this: Most AI startups, particularly in the AI companion space, rely
heavily on subscription models for revenue.

The High Costs of AI Model Training and Operation

As mentioned above, the scale of investment required is staggering.

Tech giants are projected to spend up to $1 trillion on data centers, real estate, chips,
and other equipment to build AI models, tools, and products.

However, as a recent Goldman Sach report highlighted, the demand for these AI
capabilities might not match the investment.

Tech giants and beyond are set to spend over $1tn on AI capex in coming years, with so far
little to show for it. So, will this large spend ever pay off? MIT’s Daron Acemoglu and GS’
Jim Covello are skeptical, with Acemoglu seeing only limited US economic upside from AI
over the next decade and Covello arguing that the technology isn’t designed to solve the
complex problems that would justify the costs, which may not decline as many expect. But

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 7/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

GS’ Joseph Briggs, Kash Rangan, and Eric Sheridan remain more optimistic about AI’s
economic potential and its ability to ultimately generate returns beyond the current “picks
and shovels” phase, even if AI’s “killer application” has yet to emerge — goldmansachs

A Tough Road Ahead. The Unsustainability of Current Business Models

You see, many AI startups, especially those making AI companions, are in a tight
spot. They’re spending a lot but not making enough. It’s a recipe for trouble in the
long run.

Understanding the Current AI Plateau

I’ve been reading and writing about AI for a while now. It is becoming obvious that
AI is hitting a wall. Here’s why:

Prediction Machines: These AI models? They’re just really good at guessing the
next word. It’s impressive, but it’s not real understanding or reasoning.

Expensive to Run: Running these AI models costs a fortune. It’s making it hard
for companies to grow or make their tech widely available.

Data Isn’t Free: Companies like OpenAI must pay for their training data.
Cleaning and processing all that information? It’s not cheap.

Environmental Worries: AI uses a lot of energy. With most electricity still

coming from fossil fuels, it’s not great for the planet.

What This Means?

AI’s come a long way, but it’s got limits.

To move forward, we need new ideas in algorithms, better hardware, and smarter
ways to make money from AI.

Some experts, like NYU’s Gary Marcus, think the AI bubble might burst soon.

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 8/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

One thing’s for sure: the AI world is changing fast.

Whether you’re building AI, investing in it, or just using it, keep your eyes open.
Things could get interesting.

AI Ai Investment Character Ai Chat Prediction Machines

Artificial Intelligence

Follow

Written by Jing Hu
153 Followers · Writer for AI Advances

Scientist -> Product (IT) / entrepreneur -> Full-time entrepreneur. I wake up, sell, market, gym, lunch, write,
make dinner, do fuck all, and repeat.

More from Jing Hu and AI Advances

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 9/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Open in app

Search

Jing Hu in AI Advances

AI Hospital Simulation. This Might Save NHS from Its Doom.

How AI-Powered Agent Hospitals Could Solve the NHS Crisis and Save Billions For the Labour?

3d ago 363 4

Kris Ograbek in AI Advances

If I started learning AI Engineering in 2024, here’s what I would do.

The exact path I would choose.

Jun 3 1.2K 18

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 10/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Gavin Li in AI Advances

Crazy Challenge: Run Llama 405B on a 8GB VRAM GPU

I’m taking on the challenge of running the Llama 3.1 405B model on a GPU with only 8GB of
VRAM.

Aug 2 1K 11

Jing Hu

360 Analysis Of The Great Straw Debate: How Many Holes a Straw Has!

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 11/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Just like any other evening, Klaas and I were chatting during afternoon tea, sharing the
complaints of our week at work.

Apr 2, 2023 160 3

See all from Jing Hu

See all from AI Advances

Recommended from Medium

Christopher Tao in Towards AI

Do Not Use LLM or Generative AI For These Use Cases

Choose correct AI techniques for the right use case families

Aug 10 1K 12

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 12/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Andrew Zuo

AI Has Already Peaked

AI kills sales say Washington State University researchers. And I was quite surprised by this
new report because I love AI. AI is…

6d ago 350 16

Lists

Generative AI Recommended Reading

52 stories · 1288 saves

AI Regulation
6 stories · 542 saves

What is ChatGPT?
9 stories · 422 saves

ChatGPT prompts
48 stories · 1910 saves

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 13/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Lakshmi narayana .U in Stackademic

STORM: Stanford’s Revolutionary Research Tool Harnessing the Power of

Agents and Agentic Workflows
In the rapidly evolving field of artificial intelligence, Stanford University has unveiled a
groundbreaking project that promises to…

Jul 28 426 1

Saman (Sam) Rajaei in Towards Data Science

Multi-Agent-as-a-Service — A Senior Engineer’s Overview

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 14/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

A technical consideration of how AI agents could function within enterprise production

environments

5d ago 249 3

Ignacio de Gregorio

Grokking, a New Form of Reasoning

Here’s a bold statement for you: This article will make you more ‘AI smarter’ and, in the
meantime, will shed light on the crude reality of…

5d ago 893 23

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 15/16
8/19/24, 4:35 PM Why AI Industry Is Struggling to Make Money | AI Advances

Golden Knight in Generative AI

Will This New AI Tool Kill ChatGPT?

ChatGPT might get obsolete unless it ramps up its game

Aug 11 402 10

See more recommendations

https://ai.gopubby.com/the-ai-bubble-some-reality-i-just-realized-in-the-genai-startup-landscape-c837a567ae3e 16/16