Will the AI tools you use still exist next year?

I want you to ask yourself this question right now.

Groq, which had garnered attention for its AI chips, has abandoned its chip business to pivot to “Neo Cloud.” They raised 34 billion yen, with a valuation of 350 billion yen. At first glance, this seems impressive. However, it indicates that they concluded they could not compete in the much-hyped “Nvidia killer” chip market.

In the same week, AI translation startup Relay shut down. The entire team was absorbed into Google’s Chrome team. The product has disappeared, leaving users stranded.

For large corporations, these two pieces of news are merely part of an “industry restructuring.” However, for small and medium-sized enterprises (SMEs), this is a matter of survival. Why? Because the vendor they entrusted their operations to suddenly disappears or becomes an entirely different company, and this is happening in reality.

“Disappearing vendors” are not exceptions

First, let’s look at the numbers.

According to data from CB Insights, about 90% of AI startups will disappear within ten years. This survival rate is nearly the same as that of startups in general, but the pace of technological generational change in the AI field is extraordinarily fast. ChatGPT was released in November 2022. In just two and a half years, the landscape of generative AI has been rewritten multiple times.

When SMEs adopt AI tools, they often think, “It has a good reputation,” “It’s cheap,” or “The salespeople were friendly.” However, there is no guarantee that the company will still exist a year later.

Take the example of Relay. AI translation seems to be a field with stable demand at first glance. However, if Google decides to get serious, the translation accuracy of a startup can be surpassed in no time. Google Translate, DeepL, and the multilingual capabilities of ChatGPT and Gemini. The market Relay was competing in was one that giants could easily cover as a “side offering.”

Groq is a bit more complicated. They marketed the speed of inference using their proprietary LPU (Language Processing Unit). Indeed, the demos were overwhelmingly fast. However, the management concluded that it was impossible to continue competing solely on the chip itself, given Nvidia’s massive moat with its CUDA ecosystem. Thus, they pivoted to cloud services. The chip became a “means,” and the product transformed into “inference infrastructure.”

This is a rational decision in itself. However, the problem is that the premise has completely changed for companies that were considering adoption based on the speed of Groq’s chips.

Five things SMEs should really check

The typical “vendor selection checklist” is written for large corporations. Market share, Gartner ratings, and so on. Frankly, these are irrelevant for local SMEs.

Here are practical checkpoints:

1. Where does the vendor’s revenue come from?

Look at the source of revenue, not just the amount of funding raised. Companies that survive on VC funding are entirely different from those that rely on customer revenue. Groq raised 34 billion yen, but the scale of its revenue at that point was not disclosed. This means they are still in the stage of “running on investors’ money.”

What SMEs should look for is, “Is this company actually generating revenue from customers of a similar scale to us?” A SaaS that charges a few tens of thousands of yen per month and is used by hundreds of companies is much more reassuring than a startup that raised hundreds of billions but has only ten customers.

2. What are the costs if you decide to stop using it?

Anyone can calculate the cost of implementing a tool. However, very few SMEs calculate the cost of stopping.

Companies using Relay likely built their translation workflows around it. Suddenly losing that tool raises questions. What about data migration? API rewrites? Employee retraining?

Before implementation, you should ask, “If this tool disappears tomorrow, will our operations come to a halt?” If the answer is Yes, you need to design it to reduce dependency on that tool or always have alternatives ready.

Specifically, check if data export is easy. This alone will cover about 80% of your judgment. Tools that do not allow data extraction, no matter how convenient, should not be used.

3. Is there a chance that major platforms will “offer this feature as a side service?”

The fundamental reason Relay failed lies here. Translation was a function that Google and OpenAI could provide “as a side offering.”

When SMEs choose AI tools, they should consider, “Is there a chance that Google, Microsoft, or OpenAI will offer this feature for free in six months?” Automatic meeting notes, email summarization, image generation—these are already being integrated as standard features by major players.

Conversely, industry-specific AI is harder for the giants to offer as a side service. AI for safety management on construction sites, AI for quality inspection in seafood processing, voice recognition that accommodates regional dialects. Such “niche but deep” tools are less likely to be swallowed by the giants.

4. Does the founder want to pursue this business or just seek an “EXIT”?

The founding team of Relay was absorbed by Google. This is known as an “acqui-hire” (acquisition for talent), where the product itself had no value, only the talent was acquired.

You can check the founder’s LinkedIn or interviews. Are they “betting their lives to solve this problem” or “wanting to create good technology to sell to a big company”? In the latter case, they likely have no intention of supporting your business from the start.

5. Can you understand the value after using it for 30 minutes?

Finally, it’s simple. If you can’t think “this is usable” after 30 minutes of using the tool, it won’t stick even if you implement it.

SMEs often lack IT departments. Tools that require reading manuals and undergoing training before they can be used will not take root in the field. A free trial for 30 minutes. If you can’t see the value in that time, the tool is not suitable for your company.

So, what should you do?

Just remember three things.

① Do what major platforms can do on those platforms.
Google Workspace, Microsoft 365, ChatGPT. Make full use of the AI features offered by major companies for a few thousand yen a month. This is the minimum standard for “vendors that won’t disappear.”

② If you use a startup, just check if “data can be extracted.”
You can use convenient startup tools. However, ensure that even if that tool disappears tomorrow, your data remains accessible. This alone significantly reduces risk.

③ For tools costing over 50,000 yen per month, reassess every six months whether they are still necessary.
Considering the speed of AI evolution, there’s a chance that a feature you paid 50,000 yen for six months ago is now available for free. The capabilities of ChatGPT’s free tier surpass those of paid plans from six months ago. If you don’t regularly take stock, you’ll continue to incur unnecessary costs.

Neither Groq nor Relay lacked technical prowess. The market structure simply changed. In the AI market, this “structural change” occurs on a semi-annual basis.

For SMEs, choosing an AI vendor is not a game of “selecting the best tool.” It’s a game of “creating a state where it doesn’t matter if they disappear.”

If you design it this way, no matter which vendor disappears, your operations will not come to a halt.