Conclusion

Let’s get straight to the point: The tectonic shifts in the “cost of using AI” have begun.

Last week, three significant news items emerged simultaneously in the AI industry:

1. OpenAI announced its in-house chip, “Jalapeño.”
2. AI chip company Cerebras saw its stock plummet by about 20% after its earnings report.
3. NVIDIA declared its full-scale entry into the agricultural robot market.

While these three may seem unrelated, they are all connected by a single thread: the simultaneous movement of three layers—chip layer, API layer, and on-site application layer—a first in recent years.

For small business owners, this signals the beginning of the end for the assumption that “AI is still expensive, right?” Let’s examine each layer in turn.

—

Layer 1: Chip Layer—What OpenAI’s “Jalapeño” Means

OpenAI’s “Jalapeño” is a custom-designed chip optimized for inference processing, specifically for the part that returns responses when users hit the API, rather than for training.

Why is this important? It becomes clear when considering OpenAI’s cost structure.

Every time ChatGPT or the API service is used, OpenAI relies heavily on NVIDIA’s GPUs running in the cloud. This “inference cost” is the heaviest part of OpenAI’s operational expenses. Reports suggest that the introduction of Jalapeño could reduce inference costs by up to 30%.

Hearing “30%” might make you think, “That’s not too bad.” However, OpenAI’s annual infrastructure costs are estimated to be in the billions of dollars. That 30% represents a significant impact in the hundreds of millions of dollars.

Now, here’s where it gets relevant for small businesses.

If OpenAI’s costs decrease, the API usage fees will also drop. OpenAI has already reduced API prices multiple times over the past year. In fact, inference at the GPT-4 level is cheaper than it was two years ago with GPT-3.5. Once Jalapeño is fully operational, this downward price trend is likely to accelerate.

Specifically, a process that currently costs 50,000 yen per month using the GPT-4 API could potentially drop to 20,000 to 30,000 yen in one to two years. Small businesses are definitely getting closer to the point where they can say, “Let’s try automating some tasks with AI.”

The fact that OpenAI is creating its own chip signifies a management decision to reduce dependence on NVIDIA while also creating room for further price reductions in API fees.

—

Layer 2: Chip Market—Cerebras’ Plunge Reflects the Reality of “Winner Takes All”

Cerebras is a startup that has challenged the AI chip market with its unique wafer-scale technology. After its first quarterly earnings report post-IPO, the CEO had to clarify that “margin forecasts were misunderstood by the market,” leading to a temporary stock drop of about 20%.

At first glance, this news may seem like just a “single company’s earnings misstep.” However, a structural reading reveals a deeper story.

The AI chip market is currently transitioning from a state of “NVIDIA dominance” to one of “NVIDIA plus in-house chip makers.” OpenAI’s Jalapeño, Google’s TPU, Amazon’s Trainium/Inferentia, and Microsoft’s Maia are all examples of major platforms developing their own chips.

In this context, business models like Cerebras’, which rely on “selling chips to others,” face the risk of dwindling buyers. If potential big clients like OpenAI or Google have their own chips, the incentive to purchase Cerebras’ chips diminishes.

Cerebras’ stock plunge is evidence that investors are beginning to notice this structural change.

For small businesses, this means that increased competition among chip manufacturers will ultimately lead to lower AI computing costs. Who will win is uncertain, but one thing is clear: costs are heading downward. This is nothing but a tailwind for small businesses.

The key is not to focus on specific chip brands. It’s essential to operate under the assumption that “the AI cost estimates from six months ago may already be outdated.”

—

Layer 3: On-Site Application Layer—The Significance of NVIDIA Entering the Agricultural Robot Market

NVIDIA has announced its entry into the agricultural robot market, planning to provide a platform that utilizes its GPUs in edge devices for tasks such as crop image recognition, automated harvesting, and growth management.

Reading this as “NVIDIA is getting into agriculture” misses the essence of the matter.

What NVIDIA truly aims to do is increase the “destinations for GPU shipments.” The growth of GPUs for data centers won’t last indefinitely. The next markets they want to embed GPUs into are factories, logistics, and agriculture. Agricultural robots are merely the entry point.

However, from the perspective of small businesses, a different picture emerges.

Agriculture is one of the largest industries in Japan’s rural areas and also one of the sectors facing the most severe labor shortages. If a giant company like NVIDIA establishes a platform, the development costs for applications and services built on top of it will dramatically decrease.

For instance, developing an agricultural image recognition system independently could cost several million to tens of millions of yen, including cameras, edge devices, AI models, and cloud integration. If NVIDIA’s platform becomes standardized, one would only need to build applications on top of it. A world where development costs are reduced to one-tenth is within reach.

The question is: “When NVIDIA builds the infrastructure for agricultural robots, who will create the services that run on top of it?”

Large corporations move slowly and are often unaware of on-site challenges. This creates an opportunity for local small businesses. Individuals who understand the field can assemble services on the now more affordable AI infrastructure. This is a move that large corporations cannot replicate.

—

What Happens When All Three Layers Move Simultaneously

Let’s summarize.

All three are decreasing simultaneously. This is not just a matter of “AI becoming cheaper.” It represents a structural shift where the “cost of not using AI” is rising relatively.

When competitors halve their operational costs with AI, if your company continues with traditional methods, you will lose competitiveness simply by doing nothing. We are on the brink of an era where the decision not to adopt AI itself becomes a cost.

—

So, What Should Small Businesses Do This Week?

I have three suggestions:

1. Discard AI estimates from six months ago.
“It was expensive when I checked before” no longer holds true. API prices can drop by half in just six months. I encourage you to recalculate based on current prices.

2. Ignore the “chip talk” and focus only on “API price changes.”
Small businesses don’t need to track chip technology trends. What you should monitor are the price lists of major APIs like OpenAI, Google, and Anthropic. If these prices drop, your AI utilization costs will also decrease. It’s a simple equation.

3. Leverage your knowledge of the field as an asset.
NVIDIA builds infrastructure, and OpenAI makes APIs cheaper. What remains is the “last mile of solving on-site challenges.” Here, small businesses have a significant advantage over large corporations. Those who understand the field can combine AI tools on the now cheaper infrastructure faster and more cost-effectively than hiring large corporate consultants.

—

Summary

What happened this week is that the three cost layers of AI—”chip, API, and on-site”—have begun to crumble simultaneously. OpenAI is cutting inference costs with its in-house chip, competition in the chip market is driving prices down, and NVIDIA is starting to build infrastructure for on-site applications.

All of this is aligned in the same direction. AI costs are going down. Irreversibly.

While you say, “It’s still too early for us,” the neighboring company is automating estimates with AI, semi-automating customer inquiries, and optimizing inventory management. That era is already upon us.

Start by checking the current API price list. A different world should be unfolding before you.