How SMEs Should Decide in an Era Where Over $3 Billion in AI Investments Are Sent to the “Graveyard”

TechCrunch has compiled a list of disappointing AI projects under the title “AI Graveyard.” Delays in the refresh of Apple’s Siri and the confusion surrounding OpenAI’s overly ambitious “super app” concept—well-known companies are stumbling one after another on projects they invested heavily in.

On the other hand, LangChain has reached a valuation of $3 billion (approximately ¥450 billion), with over 7,000 customers.

What accounts for this disparity? Is it technological prowess? Financial strength?

No. It’s the difference in “whose and what kind of hassle has been eliminated.”

And this question isn’t just for large corporations. It applies equally to local SMEs considering a ¥50,000 monthly investment in AI tools.

—

Common Traits of Projects Sent to the “AI Graveyard”

First, let’s look at the failed projects.

Apple’s Siri Refresh. iPhone users must have experienced it. What are the odds of getting a proper response when you ask, “Hey Siri, what’s on my schedule today?” Apple announced a significant AI enhancement for Siri at the 2024 WWDC, but implementation has been significantly delayed. They have been completely outpaced by ChatGPT and Google Gemini. This is from a company with the highest market capitalization in the world.

OpenAI’s “Super App” Concept. By cramming in code execution, image generation, file analysis, plugins, and all sorts of features into ChatGPT, users ended up in a state of confusion, thinking, “In the end, I don’t know where to start or what I can do.” Although the features increased, the number of users who could effectively utilize them decreased.

What these projects have in common is that they were driven by a “technology-first” approach.

• “Look at all the amazing things we can do” came first.
• “So whose and what becomes easier?” was pushed to the back.
• As a result, users were left behind.

Large corporations can survive even if they send projects to the graveyard due to their resources. However, SMEs do not have that luxury. If a ¥1 million investment fails, it’s equivalent to losing the salary of one new hire for the next fiscal year.

—

The Structure Behind LangChain’s $3 Billion Valuation

Now, let’s look at the winning side.

Examining the valuation trend of LangChain reveals its rapid growth.

• April 2023: Series A, valuation of approximately $200 million
• Late 2024: Valuation of $3 billion (about 15 times)

A 15-fold increase in just a year and a half. Why?

LangChain’s offering, “LangSmith,” is a tool that centrally manages the development, testing, and monitoring of AI agents. The key point is that it does not depend on any specific AI model. It can be used with OpenAI’s GPT, Google’s Gemini, or Anthropic’s Claude—any model works.

What this means is that users can hedge against the uncertainty of the “AI model supremacy battle.”

This month, GPT-4o may be optimal, but next month Claude 4 might surpass it. At that point, a system that is heavily reliant on a specific model would need to be rebuilt. With LangChain, you only need to swap out the model.

In other words, LangChain is not selling “AI technology”; it is selling a mechanism that makes the cost of switching AI models zero.

This is why it resonated with 7,000 companies. They sold structure, not technology.

—

Another Success Story: ElevenLabs’ “Narrow and Deep” Strategy

ElevenLabs is also worth noting. Known for its voice AI technology, the company has entered the market for meeting note applications.

The meeting note app market is already crowded with competitors like Otter.ai and Fireflies.ai. However, ElevenLabs is differentiating itself by leveraging its voice recognition and synthesis technology with a unique approach of “automatically generating minutes and reading summaries aloud.”

What’s noteworthy is that they did not aim to create an “AI that can do everything.” They focused on their strength in voice technology and specialized in specific use cases that derive from it.

This “narrow and deep” approach is one that SMEs should take the most inspiration from.

—

Three Questions That Separate Success from Failure in AI Investment for SMEs

Now, onto the main topic. When local SMEs are deciding on AI investments, it’s not about checking off “three checklist items” from a textbook. It’s about three questions they should ask themselves.

Question 1: “Whose and what hours does it eliminate?”

When I receive inquiries about wanting to implement AI, the first question I ask is this.

It’s not about abstract discussions like “understanding customer needs.” It’s much more specific: “Who in the company is currently spending how many hours each month on which tasks that will be eliminated?”

For example, in a manufacturing company, if a staff member spends 20 hours each month preparing a quality report, that’s about ¥40,000 at an hourly wage. That’s ¥480,000 annually. If this can be automated with an AI tool costing ¥5,000 a month, that’s a savings of ¥420,000 annually. The return on investment would be positive from the first month.

AI investments that cannot make this calculation should not be pursued.

Projects that start with “It seems impressive” usually end up in the graveyard. They follow the same structure as Apple’s Siri.

Question 2: “Will it still be usable if the model changes next year?”

The speed at which AI models evolve is extraordinary. GPT-3.5 was cutting-edge just two years ago. Now, GPT-4o is the norm, and next year, an even newer model will be released.

What SMEs often do is heavily rely on a specific AI service. “I created a Q&A bot for our operations manual using ChatGPT”—that’s great. But what happens if that bot is entirely dependent on OpenAI’s API, and the API costs double? What if another model becomes more accurate?

Systems with high switching costs become liabilities.

Remember why LangChain was valued so highly. Its design does not depend on specific models. SMEs need to adopt the same mindset.

Specifically:

• Document and manage prompts (instructions to AI).
• Avoid direct dependency on specific vendor SDKs by incorporating an abstraction layer.
• Conduct comparative tests with other models at least once every six months.

Even for tools costing a few thousand yen a month, this perspective should be maintained.

Question 3: “Can it operate without the CEO?”

The most common failure pattern in AI implementation for SMEs. The CEO personally takes an interest in the implementation, becomes the only one who can utilize it, and when the CEO loses interest, it’s over.

AI utilization that is personalized is not true AI utilization; it’s just a hobby.

Successful AI implementations become systems. They produce the same results regardless of who does them. There are procedures in place. Handover is possible.

What ElevenLabs aims for with its meeting note app is precisely this. AI automatically processes the personal task of writing minutes, which is dependent on who attended the meeting. It creates a system that does not rely on individuals.

For SMEs, this means:

• Teach AI the procedure for creating estimates so that even newcomers can handle it on the same day.
• Delegate FAQ responses to an AI chatbot so that operations can continue even when the responsible person is absent.
• Automate the summarization of daily reports so that managers can grasp everyone’s status in five minutes.

All of these can be achieved within the range of a few thousand to tens of thousands of yen per month. There’s no need for a ¥3 million system development.

—

Separate “Investment” from “Experiment”

Finally, one last point.

For SMEs, it’s better not to think of AI implementation as an “investment” from the outset. Start with it as an “experiment.”

Try a tool costing ¥5,000 for one month. If it eliminates 10 hours of work, consider full implementation. If it doesn’t, cancel it the following month. That’s all.

While large corporations are spending hundreds of millions on creating “AI graveyards,” SMEs can run ten experiments at ¥5,000 each. That’s ¥600,000 a year for ten experiments. If two of them succeed, that’s hundreds of thousands of hours saved annually.

This is the strategy that SMEs can uniquely adopt.

There’s no need to make a huge investment and go all in like large corporations. Experiment on a small scale and only scale up what works. LangChain’s $3 billion valuation also started from a small open-source library.

The dividing line between “failed AI investments” and “recoverable AI investments” is not about the quality of technology.

It’s about whether it’s clear “whose hours are being eliminated,” whether it can be used even if the model changes, and whether it can operate without the CEO.

If you can answer these three questions, then go ahead. If you can’t, it’s still too early.

Let’s start with a ¥5,000 experiment.