As AI Companies Self-Destruct, AI Costs for SMEs Decrease—What the Nvidia Turmoil, Anthropic Crash, and CoreWeave Debt Bomb Mean

Conclusion Let’s get straight to the point. The more AI companies struggle, the lower your company’s AI costs will be.

By Kai

|

Related Articles

Conclusion

Let’s get straight to the point. The more AI companies struggle, the lower your company’s AI costs will be.

The turmoil surrounding Nvidia’s CEO dismissal. The stock price crash of Anthropic following its IPO. The massive debts of CoreWeave.

These events are reported as the “crisis in the AI industry.” However, for local small and medium-sized enterprises (SMEs), this is the exact opposite news.

Why? Because the cost structure of AI is beginning to collapse.

A year ago, using an API equivalent to GPT-4 cost $30 per million tokens. Now, models of similar performance can be used for under $3. That’s one-tenth the price. And this trend will accelerate further due to the current “self-destruction of AI companies.”

What SMEs should do is simple: negotiate lower prices for the AI services they are currently using or switch to cheaper alternatives. In this article, we will explore why now is the right time to do so through three news stories.

—

1. Nvidia—The “King of Chips” is on Shaky Ground

Nvidia CEO Jensen Huang is facing pressure to resign. Reports cite allegations of infidelity and a lack of governance transparency, but that’s not the essence of the issue.

What’s noteworthy is Nvidia’s announcement of its withdrawal from the Chinese market for the H200 chip.

Approximately 20-25% of Nvidia’s revenue comes from data centers, and the Chinese market has been a crucial part of that. The disappearance of this market means that surplus chips will flow into other markets. When supply increases, prices will fall. This is basic economics.

Moreover, AMD and Intel are improving the performance of their AI chips. Google’s TPU and Amazon’s Trainium are also accelerating their in-house development. If Nvidia’s monopoly crumbles, the procurement costs for GPUs will decrease. If GPU costs go down, the cost structure for all AI services built on top of them will also decrease.

What this means for SMEs: The AI services you pay for monthly are underpinned by GPU costs. If those GPUs become cheaper, it opens up room for negotiation on the amount you pay.

—

2. Anthropic—The “Facade” of a $1.3 Trillion Valuation Has Been Stripped Away

Anthropic is known for its AI model called Claude. On its IPO day, it was valued at $1.3 trillion. However, its stock price has since plummeted.

What does this signify?

It indicates that the market is beginning to judge that “the AI model itself no longer holds that much value.”

Consider this: A year ago, only a few companies like OpenAI, Google, and Anthropic could create cutting-edge AI models. What about now? Meta’s free release of Llama 3.1 matches GPT-4 in many tasks. China’s DeepSeek has developed high-performance models at astonishingly low costs. Open-source models are updating performance records almost every month.

In other words, AI models are becoming commoditized. We are moving towards an era where it doesn’t matter where you buy them, just like water or electricity.

The crash of Anthropic’s stock price is evidence that investors are starting to realize this reality. In a commoditized market, price competition will ensue.

Let’s look at the numbers:

Period Cost of GPT-4 Class API (per million tokens)
March 2023 Approximately $30
May 2024 Approximately $10
June 2025 Approximately $2-3

In a year and a half, the price has dropped to one-tenth. This price decline is not yet stopping. As Anthropic faces competition, the price of Claude will also decrease. OpenAI will have to lower its prices as well. Google Gemini will follow suit.

What this means for SMEs: If you have an AI service you contracted for “50,000 yen per month” six months ago, you should review it immediately. There’s a high likelihood you can get the same performance for 10,000 to 20,000 yen per month. Alternatively, using open-source models might allow you to achieve similar results at nearly zero cost.

—

3. CoreWeave—The “AI WeWork” Shows a Debt Bomb

Have you heard of CoreWeave? It’s a company that operates data centers by borrowing a large number of Nvidia GPUs and rents them out to AI startups. It’s rapidly growing but has total debts in the billions of dollars. Its valuation at IPO was around $23 billion, but the stock price fell immediately after going public.

Does this structure remind you of something? Yes, WeWork.

It finances massive capital investments through debt and envisions growth based on “future demand.” If demand doesn’t materialize as expected, it can collapse suddenly. CoreWeave is not alone; several similar GPU cloud companies exist, all with similar structures.

What happens if this collapses? An oversupply of GPUs will occur. To maintain data center utilization, they will start offering GPU time at prices close to fire sale levels.

In fact, signs of this are already evident. GPU cloud services like Lambda Labs, Together AI, and Fireworks AI have reduced GPU usage fees by 30-50% compared to a year ago. If CoreWeave faces serious difficulties, this price competition will intensify.

What this means for SMEs: If you want to fine-tune your own AI model, the costs will drop dramatically. What used to cost 3 million yen to build a dedicated AI model can now be done for under 500,000 yen. There’s potential for further decreases.

—

So, What Should You Do?

Let’s not talk abstractly about “preparing for the AI era.” Just do three specific things.

① Take Inventory of Current AI-Related Costs

List all your expenses: ChatGPT Plus (monthly $20), various SaaS AI features, outsourced AI-related tasks. Many companies are unaware of “what they are paying for each month.”

② Consider Negotiating for Lower Prices or Switching

Prices from six months ago are now “old prices.” Investigate whether equivalent services are being offered at lower costs. Especially for API-based services, price fluctuations are significant. Options are continuously increasing from OpenAI to Claude to Gemini to open-source.

Let’s give a concrete example. A manufacturing company in a local area was using an AI transcription service for meeting minutes at 30,000 yen per month. After switching to OpenAI’s Whisper API, the cost dropped to under 2,000 yen per month. The performance is nearly the same. In this world, just knowing this difference can save you 28,000 yen a month.

③ Start Considering a “Dedicated AI” for Your Company

By using open-source models (like Llama, Mistral, etc.), you can create a dedicated AI trained on your company’s data at a cost that is now one-tenth of what it used to be. Customer support, estimate creation, quality checks—tasks that are highly personalized will see the greatest benefits.

You might think, “A small company like ours can’t handle an AI model.” However, today’s open-source AI is more powerful than the enterprise-level AI from three years ago. And the cost is less than one-hundredth. Because SMEs can make decisions quickly, you can try, implement, and achieve results while large corporations are still going through their approval processes.

—

“The Crisis of AI Companies” Can Become a Weapon for SMEs

Let’s summarize:

  • Nvidia’s Turmoil → The GPU monopoly is crumbling, leading to lower chip costs.
  • Anthropic’s Crash → The commoditization of AI models is accelerating, resulting in lower API fees.
  • CoreWeave’s Debt Bomb → Price competition in GPU cloud services is intensifying, making computational resources cheaper.

All three point in the same direction. The cost structure of AI is structurally decreasing.

This is not a temporary sale. It’s a change in the industry structure. Price destruction is happening simultaneously across all three layers: chips, models, and infrastructure.

Large corporations are bound by existing high-cost contracts. SMEs have no such constraints. You can start at the lowest price today. This is the greatest weapon for SMEs.

Will you look at the self-destruction of AI companies and think, “What a disaster,” or will you see it as an “opportunity” to act? The difference always lies there.

What to do this month: Write down all your AI-related expenditures. That’s all you need to do. Once you write it down, you’ll see where you can cut costs. Once you see it, you can act.

The cost structure of AI is dropping faster than you think.

POPULAR ARTICLES

Related Articles

POPULAR ARTICLES

JP JA US EN