Selling the infrastructure that drove the artificial intelligence revolution allowed Nvidia to amass enormous profits.
Jensen Huang, the CEO, could now want a larger portion of the intelligence operating on top of it.
Nvidia (NVDA) is reportedly in early talks to buy Reflection AI, invest more heavily in the startup, or strike a deal that brings its employees and technology inside Nvidia. The chipmaker has already invested about $800 million in Reflection, which was seeking a roughly $25 billion valuation earlier this year, according to the Financial Times.
That could seem like a typical AI transaction.
It might be considerably larger than that.
Nvidia has had one of the safest positions in artificial intelligence for many years. OpenAI has a chance to succeed. Anthropic is able to prevail. A startup that no one has heard of yet has the potential to succeed. Nvidia may still sell all of those chips.
Buying deeper into Reflection could change that equation.
Nvidia may start participating in the AI model competition rather than just delivering it.
And it raises a question that investors have rarely had to consider during Nvidia’s incredible success: What happens when the business that sells everyone the picks and shovels starts looking for the gold itself?
Nvidia could be entering the fight over AI models.
Reflection AI is not your typical software startup.
Misha Laskin and Ioannis Antonoglou, two former Google DeepMind researchers, founded Reflection, which is creating robust open-weight AI models that anyone can download, alter, and run on their own.
Put simply, that increases the authority of governments and corporations.
A bank may not want private financial data to pass through the closed AI service of another party. The military would want a model that can function inside its own secure systems. A big business could want to tailor AI to private data without giving up control of that data.
Reflection is wagering that consumers will increasingly seek alternatives to closed systems from firms like Anthropic and OpenAI.
The startup unveiled Beam, its latest model, on Oct. 5. Reflection says Beam is designed for coding, reasoning and AI-agent tasks.
That timing is difficult to ignore. Reflection launches a key model. Reports surface days later that Nvidia may acquire the company.
That gives investors a reason to look beyond the acquisition price.
Nvidia may be deciding that selling the hardware underneath AI is no longer enough.
Reflection could give Nvidia another way to win
Nvidia already touches almost every level of artificial intelligence.
It offers networking hardware, whole data center systems, and graphics processing units, or GPUs. Additionally, it creates software that facilitates developers’ usage of such chips.
Nvidia would advance even farther with reflection.
More Nvidia:
- Nvidia stock flashes massive signal as Wall Street leans in
- Nvidia’s AI boom revives a Warren Buffett warning for investors
- Nvidia puts $500 billion on the table, challenges banking as we know it
Consider how Apple developed their ecosystem.
Apple doesn’t just sell iPhones and go. The software, services, and apps that users use later on also help Apple.
It’s possible that Nvidia is working on a comparable position in AI.
More computation is required for more powerful models. Increased computation often results in increased demand for the processors, networking hardware, and systems that Nvidia currently supplies.
Thousands of Nvidia GB300 GPUs were used to train Reflection’s Beam model.
This produces a very appealing loop.
Nvidia may support businesses creating AI models, provide the technology required for model training, and perhaps even control a portion of the model layer itself.
Businesses may still need large amounts of Nvidia compute to operate an Nvidia-backed open model, even if it is less expensive for them to use than a closed option.
That means Nvidia could potentially win twice. First, when the model is trained. Then again, every time somebody runs it.
Nvidia’s Reflection AI talks raise an awkward question
Nvidia could also create a problem it has never had before
There is a clear conflict in the plan. The fact that almost anyone can become a client has been Nvidia’s biggest advantage. OpenAI purchases computers. Computing is necessary for anthropic requirements.
Although Meta Platforms develops its own models, it still makes extensive use of Nvidia technology. Nvidia chips are also necessary for startups attempting to compete with all three.
Regardless of the winning model, Nvidia is compensated.
Having influence over or ownership of a significant model developer may complicate that relationship.
Imagine discreetly purchasing the third restaurant across the street while selling equipment to the first two.
Your clients can begin to question which side you are on.
This does not imply that Anthropic or OpenAI will abruptly quit purchasing Nvidia processors. Few alternatives can match Nvidia’s scale for AI firms.
However, it may create a competitive dynamic that hasn’t really existed since Nvidia’s ascent.
Some of the most successful software companies, in my opinion, are valued precisely because they operate under the competition rather than actively engage in it. Regardless of whatever consumer brand ultimately prevails, payment networks, cloud platforms, and chip vendors stand to gain.
For most of the AI boom, Nvidia has benefited from that edge.
Nvidia has another chance to expand by branching out into models, but doing so also erodes that neutrality.
The structure of a Reflection deal could matter
Investors should keep an eye on another section.
It’s possible that Nvidia won’t purchase Reflection via a conventional acquisition.
According to the Financial Times, options include further funding or an “acqui-hire,” in which Nvidia hires Reflection staff members and licenses its technology.
With that arrangement, Nvidia could get much of what it wants without actually buying the whole firm.
It is also becoming a more sensitive area.
Nvidia used a similar approach in its roughly $20 billion arrangement involving AI-chip developer Groq. Former Groq employees later challenged the transaction in court, while the Financial Times reported that U.S. authorities have examined aspects of the arrangement.
The same problems wouldn’t necessarily arise in a Reflection transaction.
However, given the pattern, investors should focus on how Nvidia closes deals rather than just if it does.
What investors in Nvidia should now pay attention to
Treating Reflection as a different business that Nvidia just happens to admire would be the simplest error.
The more significant possibility is that it indicates Huang’s preferred course for Nvidia.
For most of the AI boom, the Nvidia story was remarkably simple. Everyone needed more computing. Nvidia sold the computing.
This simplicity enables one of the most potent business models in technology.
Reflection indicates a more ambitious goal.
Nvidia already provides whole computer systems, finances AI infrastructure, builds its own model families, and supports model creators. A more comprehensive Reflection agreement may force Nvidia to compete directly for the adoption of AI models by governments and corporations.
Upside results from that.
Additionally, it generates a new kind of danger.
The business could compete with those that invested billions in hardware and extract more value from each layer of AI.
The talks are preliminary, and Nvidia may not buy Reflection. However, investors should not ignore the talks’ signals. Nvidia’s biggest advantage for years was not needing to know which AI company would win. Nvidia may make that bet if Huang continues the model race.
