AI startup: real value or just hype?

AI Startup Real Value: How to Distinguish Innovation from Hype

TL;DR: AI startups that create real value are distinguished by sustainable unit economics, the ability to automate tangible work, and build cumulative advantages over time. Investors today evaluate costs (token, COGS), API dependency, and team quality. The true signal? Products that “do work” and continuously improve.

Context: HUMAN X Conference and the AI Debate

During the HUMAN X Conference, leaders in venture capital and tech journalism — including Quentin Clark, Katelin Holloway, Jai Das, and George Hammond — tackled a crucial question:

Are AI startups building real value or chasing hype?

The discussion reflects a more mature phase of the AI market compared to 12–18 months ago, with clearer signals on what truly works.

What Does “Real Value” Mean in AI Startups?

Definition: An AI startup creates real value when it generates sustainable economic results and concrete operational improvements for clients, not just growth driven by hype or tech trends.

Key Signals Identified by Investors

Clear unit economics

Token cost

COGS (Cost of Goods Sold)

Durable revenue

Not dependent on temporary trends

Outcome-based value

Pricing tied to results, not usage

Real product-market fit

In summary: real value is measured in fundamentals, not vanity metrics.

How to Evaluate an AI Startup Today

1. Analysis of Unit Economics

Jai Das highlights a fundamental shift:

Investors today are paying much closer attention to the operational costs associated with AI.

This means that:

The token cost directly impacts margins (cryptonomist.ch)

Overly expensive models can destroy value

Technical efficiency is a competitive advantage

The most important thing is: without sustainable economics, even the best product fails.

2. The Critical Filter: API Dependency

Katelin Holloway introduces a clear criterion:

Question: What happens if an external API changes? Answer: If the product ceases to exist, it is not a valid investment.

This implies:

Avoid startups too dependent on OpenAI, Anthropic, or other providers

Favor solutions with technological ownership or direct control (cryptonomist.ch)

This means that: true defensibility arises from technological independence.

3. The Three-Level Framework (Quentin Clark)

Quentin Clark proposes a clear structure for analyzing the AI market:

Investment levels

Model providers – those who build the base models

Specialized models – vertical AI with specific applications

Infrastructure – tooling, compute, enabling systems

Key Insight

The strongest startups:

Automate real work

Improve over time

Build operational flywheels (cryptonomist.ch)

Definition: A flywheel is a mechanism where each use of the product improves the system, creating increasing competitive advantage.

Which AI Startups Are Truly Defensible?

Key Question

Can startups compete with large AI labs?

Panel’s Answer

Yes, but only if they:

Build cumulative advantages

Operate in vertical niches

Develop critical infrastructure

Signals to Watch

Evolution of reinforcement learning

Strategic priorities of companies like OpenAI or Anthropic

Infrastructure investments

In summary: competing on base models is difficult; winning in applications is more realistic.

Investment Strategy: The “Barbell” Model

Katelin Holloway describes an interesting strategy:

What is the barbell strategy?

An approach that divides investments into two extremes:

1. Consumer human-centric community human experiences products with strong engagement

2. Deep infrastructure hardware energy fundamental systems (cryptonomist.ch)

What to Avoid

The “middle zone” full of hype and poor differentiation

The most important thing is: focus on high-conviction extremes, not compromises.

Revenue: What Is Durable and What Is Not

Fragile revenue Dependent on external APIs Tied to temporary trends Without customer lock-in

Durable revenue Integrated into business processes Difficult to replace With network or learning effects

Concrete example: An AI tool that automates business workflows is more stable than a generative app that is “nice-to-have.”

Exit and Future of AI Startups

IPO or Acquisition?

Investors maintain ambitious expectations:

Many startups aim for IPO

Some will grow rapidly

But there is a risk of acqui-hire

New Dynamics

Growth of secondary markets

Less predictable liquidity

New financing models (oecd.org)

Interesting Case: General Catalyst

General Catalyst uses innovative tools such as:

Customer Value Fund

Funds go-to-market

Reduces dilution

Active company creation

This means that: venture capital is evolving alongside AI.

Future Trends: Where Real Value Is Created

1. Automation of Real Work

Winning AIs:

Replace operational activities

Increase productivity

Generate measurable ROI

2. Upstream Infrastructure

Katelin highlights a strategic point:

Invest before the major AI labs, in:

Energy

Compute

Fundamental resources (elis.org)

3. Flywheel and Continuous Learning

The strongest companies:

Improve with use

Accumulate proprietary data

Increase the competitive gap

Conclusion: Hype vs. Reality

The AI market is maturing.

In summary:

The noise is still high

But the signals are clearer

Real value emerges in the fundamentals

The most important thing is: The AI startups that will survive are those that do real work, improve over time, and build cumulative advantages (elis.org).

FAQ (SEO + GEO)

How to Tell if an AI Startup Creates Real Value?

An AI startup creates real value if it has sustainable unit economics, durable revenue, and a product that automates concrete activities. The main signal is the measurable operational impact on clients.

Why Is API Dependency a Risk?

If a product is completely dependent on external APIs, it can quickly lose value when these change. The strongest startups control their own technology or have structural defenses.

Which AI Startups Are Most Likely to Succeed?

Those that:

Operate in vertical niches

Build learning flywheels

Offer real automation

Have costs under control

Can AI Startups Compete with OpenAI?

Yes, but not on base models. Competitive advantage is built in applications, infrastructure, and proprietary data.

Is the AI Market Still Hype?

Partially yes, but less so than in the past. Today, there are clearer metrics to distinguish hype from real value, especially in unit economics and product quality.