Old Huang, I really don’t want to just be the one selling shovels!
Today, multiple foreign media outlets reported that NVIDIA plans to invest $6 billion (approximately RMB 43 billion) to build the world's most powerful open-weight AI model.
NVIDIA has finally dropped the pretense—they're now stepping in directly to dominate the mining market.
To realize this ambition, Huang Lao unveiled a comprehensive series of moves.
First, he invested $6 billion to acquire the "model factory" technology from the AI startup Poolside.
He then aggressively recruited the core team, directly inviting all 109 lead engineers from Poolside to join NVIDIA’s Nemotron project.
Finally, he made a massive additional investment in Poolside, pouring in $1 billion at a pre-money valuation of up to $12 billion!
With this seamless series of moves, a "soft acquisition" was accomplished in no time.
The industry is shocked: the competitive landscape for large AI models is about to change dramatically.
Old Huang is no longer the hidden powerhouse behind the scenes; now, he’s wearing a leather jacket and stepping right onto the stage.
Master Trading: Securing a $1.2 Billion Unicorn
So, why was Poolside chosen?
This company has an extremely impressive background.
Founded in 2023, the company was co-founded by Eiso Kant, a seasoned software developer, and Jason Warner, former CTO of GitHub. Their Laguna coding model is highly renowned in the Western open-source community.
However, even the most brilliant tech genius must earn their daily bread.
In a letter sent to shareholders on Thursday by Poolside, a heartbreaking account of survival was revealed.
The letter mentioned that, at the end of last year, they had a six-week window to urgently raise $2 billion to pay for the 40,000 GB300 supercomputing clusters set to launch in January.
What was the result? They failed to complete the funding on time and lost the cluster.
The founding team has finally come to terms with this harsh reality: without substantial capital and top-tier data center support, the company's computing power will be completely depleted by next year.
Without computing power, it’s equivalent to a death sentence.
At that very moment, NVIDIA descended from the heavens.
NVIDIA not only has money but also possesses the world’s largest and most advanced computing infrastructure. Poolside explicitly stated in the letter that NVIDIA is a “perfect partner” because they face almost no limitations in terms of computing infrastructure or cash.
More interesting is NVIDIA's trading strategy.
Instead of directly acquiring Poolside in full, they opted for a model involving a 6-billion-license fee, the recruitment of 109 employees, and a 1-billion-investment.
Thus, Poolside’s founders, Kant and Warner, along with operations executive Margarida Garcia, will remain in their roles, and the company will continue to operate independently.
Keep in mind, this is already the third time NVIDIA has adopted this trading model! It previously took the same approach with Groq and Enfabrica.
This clever "soft acquisition" strategy allows for the rapid acquisition of cutting-edge technology and a hundred-person team while skillfully avoiding the stringent scrutiny that regulatory authorities apply to "full acquisitions."
Targeting trillions of parameters! Nemotron 4 is about to take off.
What exactly are these 109 top engineers, poached with huge salaries, going to do?
There is only one answer: Go all out for Nemotron 4!
In fact, NVIDIA has long launched its open-weight model project—Nemotron. However, in the past, this project seemed more like a demonstration of NVIDIA’s superior GPU computing power.
But things are different now. Nemotron has now been elevated by Huang to a strategic level.
Earlier this month, NVIDIA released the lightweight Nemotron 3.5 Lightning model.
But this is just an appetizer—the real main course is the massive Nemotron 4, currently under secret development.
According to insiders, with the addition of Poolside’s core team of a hundred, Nemotron 4 is reaching a staggering scale of "trillion-level training parameters."
From now on, it is likely to directly join the world's top tier of large models.
NVIDIA's ambition is to leap straight to the top by creating the strongest American open-source alternative capable of competing with the most powerful models.
In his letter, the founder of Poolside explained that the purpose of this transaction is to ensure that AGI’s future "will not become a closed technology controlled by a few, but rather a technology jointly built by many in an open environment."
Open source is the right path! This time, NVIDIA has bet on open-source large models.
The "Sputnik moment" brought by open-source large models has made Old Huang anxious.
Why has NVIDIA, which has long been quietly reaping huge profits, suddenly decided to invest heavily in open-source models?
Due to the most advanced open-source large models, U.S. tech giants are experiencing unprecedented pressure.
Although Huang has consistently publicly supported open-weight models, most of the resources in the U.S. AI industry are focused on the proprietary large models of giants like OpenAI, Anthropic, and Google.
This presents an excellent opportunity for open-source AI.
January 2025 is destined to be a landmark month in AI history. When DeepSeek When it released its first low-cost, high-performance open-source weight model, the entire U.S. tech stock market cried out in anguish.
Investors suddenly realized that top open-source models not only approach the capabilities of the most advanced U.S. technologies but also surpass them in cost control and open-source ecosystems.
Silicon Valley renowned venture capitalist Marc Andreessen even called that moment America’s “Sputnik moment” for AI.
Subsequently, more open-source models caught up.
Old Huang can no longer sit still.
Last month, Jensen Huang published his first long-form post on X—“Open Weights and American AI Leadership.”
He passionately argued in the article: “Whether the United States can win cannot be judged solely by a cutting-edge proprietary model, but rather by whether the U.S. can build a robust, industry-wide open ecosystem.”
To this end, NVIDIA must step in directly.
As early as March this year, NVIDIA spearheaded the formation of the "Nemotron Alliance," bringing together Mistral, Perplexity, and others to share data and computing power; at the same time, NVIDIA made significant investments in Reflection AI, which has been dubbed the "American version of" DeepSeek ".
Today’s $6 billion acquisition of the Poolside team is a natural next step for NVIDIA.
Standing up to OpenAI—Is Huang not afraid of alienating a major client?
NVIDIA's strong entry inevitably creates a delicate situation—
NVIDIA is becoming a direct competitor to its largest customers.
Big companies like OpenAI and Anthropic are NVIDIA's largest buyers of GPUs. Now that NVIDIA is investing its own money to develop the world's most powerful model, isn't it afraid of alienating big names like Sam Altman?
Netizens commented wittily: “The people selling shovels now want to own the gold mine themselves.”
Others have offered sharp analysis: “I don’t believe they’ll try to completely eliminate their core customer base. Instead, they’ll strive to find a balance—keeping their model’s performance just below a certain threshold, while still providing enough challenge and pressure to compel large customers to buy more chips.”
This is a highly sophisticated "catfish effect."
The deeper reason is: major tech companies have long been turning against NVIDIA.
OpenAI, Google, Microsoft, and even Amazon are aggressively developing custom AI chips to reduce their heavy reliance on NVIDIA GPUs.
If that's the case, why shouldn't NVIDIA get into the model business?
As some have said: "NVIDIA isn't doing this out of charity. On one hand, it’s to increase market demand for its GPUs; on the other, they may be targeting enterprise customers eager to own an AI technology stack entirely built on 'American foundations.'"
After all, no such massive entity can run on standard consumer-grade hardware.
The larger the model, the higher the computing power demand, and NVIDIA's empire remains as solid as ever.
In this era of compute power, NVIDIA is reshaping the open-source large model landscape with a straightforward and forceful approach.
Reference materials:
https://x.com/mark_k/status/2091605721611723115
https://www.bloomberg.com/news/articles/2026-08-20/nvidia-to-pay-ai-startup-poolside-a-6-billion-license-newcomer-says
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation; edited by Aeneas.