Nvidia has spent billions on AI companies without actually buying them. That may be changing what an investment in these startups really means.
There is a new pattern emerging in the AI industry, and Nvidia is at the center of it.
Instead of acquiring an entire startup, Nvidia is increasingly taking the pieces it wants most: technology, intellectual property, licenses and key employees.
The companies remain independent on paper.
But the technology and people that made them valuable can end up inside Nvidia.
That raises a much bigger question for startup investors and employees:
If Nvidia buys the technology and the team, but not the company, what is actually left for existing shareholders?
The answer is not always straightforward.
Nvidia is buying the valuable pieces, not necessarily the company
Over the past nine months, Nvidia has been involved in two major reported transactions structured around this idea.
The first was Groq, followed by Poolside. Nvidia is also reportedly discussing an investment in Perplexity at a valuation above $30 billion.
Together, the reported Groq and Poolside transactions represent around $27 billion in value.
But there is an important distinction.
Neither Groq nor Poolside was acquired outright.
That means these were not traditional acquisitions where Nvidia buys the company, takes control and shareholders receive a standard acquisition payout.
Instead, the deals separated the pieces of the business.
Nvidia could gain access to critical technology and talent, while the original company continues to exist in some form.
That distinction matters enormously to shareholders.
Groq showed how this model works
In December, Nvidia reportedly entered into an approximately $20 billion transaction involving Groq’s inference technology.
The companies described it as a non-exclusive license, along with substantially all of Groq’s assets.
Groq’s founder and CEO Jonathan Ross, president Sunny Madra and other senior leaders joined Nvidia.
But Groq itself did not disappear.
GroqCloud remained behind, with Simon Edwards, previously the company’s finance chief, becoming CEO.
That created two very different things:
- Nvidia got the technology, assets and key leadership.
- Groq retained a continuing business.
- Existing shareholders reportedly received distributions tied to the transaction.
- The surviving Groq business later raised more capital at a much lower valuation.
And that last point is particularly important.
Groq had previously been valued at around $6.9 billion in its September 2025 funding round.
Nvidia’s reported transaction was worth roughly $20 billion.
Yet the remaining Groq business later raised money at a $3.5 billion valuation.
Those numbers are not contradictory.
They represent different things.
The $20 billion transaction reflected the value attached to what Nvidia was taking.
The $3.5 billion valuation reflected the business that remained after the transaction.
The headline deal value and the value of the surviving company were never the same thing.
Then came Poolside
Poolside offers another version of the same strategy.
According to reporting cited in the source, Nvidia agreed to:
- $6 billion for a non-exclusive license to Poolside’s Model Factory software
- $1 billion investment in Poolside
- Offers to 109 employees who built its Laguna model
Poolside reportedly described the arrangement as neither an acquisition nor an acquihire.
But the circumstances behind the deal are revealing.
Poolside had been trying to raise $2 billion to build a 40,000-GPU cluster.
It had six weeks to secure the financing.
It missed the deadline and lost the cluster.
That context changes how the transaction looks.
For Poolside, the deal may have provided a way to keep the company alive, bring in fresh capital and preserve value when the alternative could have been significantly worse.
This is where the reverse acquihire model becomes more complicated than a simple “Nvidia bought the startup” headline.
Perplexity could take the model somewhere new
The reported discussions with Perplexity are particularly interesting because the company does not appear to fit the same distressed-startup narrative.
Nvidia is reportedly discussing a direct equity investment in Perplexity at a valuation above $30 billion.
Perplexity was last valued at around $20 billion in September 2025.
Its reported annualized revenue has also grown from less than $250 million to more than $750 million this year.
There is no announced agreement yet, so the talks may not result in a transaction.
But if Nvidia does invest, it would show that these relationships are not limited to companies that need rescuing.
Nvidia may simply want strategic access to important AI companies without taking full ownership of them.
That could become an increasingly important part of how the AI ecosystem develops.
Why not just acquire the company?
This is where the structure gets interesting.
A traditional acquisition gives the buyer control over the company.
A licensing deal can give the buyer access to specific technology without requiring a full acquisition.
A talent arrangement can bring key people into the larger company.
An investment can give Nvidia an economic and strategic relationship with the startup while leaving the company independent.
The pieces can be combined.
Technology here.
People there.
Capital somewhere else.
For Nvidia, that can offer flexibility.
It can access technology and talent while allowing the startup itself to continue operating.
There can also be regulatory considerations. A licensing arrangement may not trigger the same premerger notification requirements as a traditional acquisition, although regulators can still examine these arrangements.
That does not automatically make the structure anti-competitive or improper.
It simply means the structure matters.
And this is not just an Nvidia story
Microsoft, Google and OpenAI have all been involved in similar arrangements.
One of the earliest high-profile examples came from Inflection AI in 2024.
Microsoft reportedly paid around $620 million for licensing while hiring the company’s co-founders and much of its staff.
Inflection continued operating under new leadership and shifted toward enterprise sales.
For investors, the outcome was unusual.
Those who invested in Inflection’s earlier $225 million round reportedly received around 1.5x, while investors from the later $1.3 billion round reportedly received around 1.1x.
The company had previously been valued at around $4 billion.
It was never acquired by Microsoft.
Character.AI followed a similar path
Google took another route with Character.AI.
Alphabet later disclosed a $2.7 billion cash payment for a non-exclusive license.
At the same time, co-founder Noam Shazeer and other employees returned to Google.
The reported payment provided liquidity for investors and employees.
Again, the company itself was not simply bought and absorbed.
Instead, technology, people and shareholder liquidity were handled through different parts of the transaction.
Windsurf made the structure even more complicated
Windsurf is perhaps the clearest example of how messy these deals can become.
OpenAI had agreed to acquire the company for around $3 billion.
That acquisition later collapsed.
Google instead reportedly paid roughly $2.4 billion to license the technology and hire the CEO and co-founder.
Cognition then acquired what remained of Windsurf.
The company said remaining employees would participate financially and receive accelerated vesting.
Three companies.
Multiple transactions.
Different groups receiving value in different ways.
This is why looking only at the headline transaction value can be misleading.
For shareholders, the headline number isn’t enough
This may be the most important lesson from these deals.
When a company is acquired, investors naturally focus on the acquisition price.
But in a reverse acquihire or licensing transaction, the money can go to very different places.
A license payment may go to the company.
Employee compensation may go directly to employees.
A primary investment goes into the company and can dilute existing shareholders.
A distribution may send some of the proceeds back to existing investors.
The result is that a $7 billion transaction does not necessarily mean $7 billion of value goes directly to shareholders.
The terms matter.
A lot.
Investors need to understand:
- Who receives the license payment?
- Does the company distribute any of it to shareholders?
- What happens to employee equity?
- How is vesting treated?
- Does the investment dilute existing holders?
- What assets remain with the company?
- What business is actually left after the deal?
These details can determine the difference between a meaningful liquidity event and a much less attractive outcome.
The Groq example makes the point
Consider what happened with Groq.
Nvidia reportedly paid around $20 billion in its transaction.
Existing shareholders reportedly received staged distributions.
Then the continuing Groq business raised another $350 million at a $3.5 billion valuation, following a reported $650 million round earlier in the year.
The surviving business had changed.
Its founder and senior leaders had left.
Its technology had been licensed to Nvidia.
Its business model had shifted toward an inference cloud using its own processors alongside Nvidia hardware.
So asking whether Groq was “worth $20 billion” or “$3.5 billion” misses the point.
Both numbers referred to different pieces of the story.
The real question is:
Which piece did you own?
A reverse acquihire can actually help shareholders
It would be easy to look at these transactions and assume shareholders are losing out.
That is not necessarily the case.
Sometimes, the structure can preserve value that might otherwise disappear.
Poolside is a good example.
The company reportedly needed $2 billion in financing for its GPU cluster and failed to secure it in time.
A deal with Nvidia provided substantial value through licensing and investment.
The alternative could have been a difficult fundraising round, a down round or potentially something worse.
The same logic applies to the continuing companies.
A startup does not necessarily have to disappear just because Nvidia takes its technology and hires some of its people.
A reverse acquihire can sometimes keep the remaining company alive while monetizing the assets that a larger company values most.
But whether that benefits shareholders depends entirely on the deal terms.
This changes how private-market valuations should be read
This trend also creates a problem for anyone looking at private-company valuations.
Suppose a startup raises money at a $12 billion valuation.
Soon afterward, Nvidia pays billions for licenses to its technology and hires a large portion of its technical team.
It would be tempting to assume that the startup’s equity has suddenly become worth more.
That conclusion may be wrong.
A license payment is not the same thing as a per-share price.
The company could receive significant cash without shareholders receiving the same amount.
And the assets that made the company attractive may no longer be fully controlled by the company.
That means secondary-market prices need to be treated carefully.
Reported private-company financials and secondary-market indications can be incomplete, unaudited or based on limited transactions.
They are not automatically fair value or executable pricing.
The bigger AI trend
There is a broader story here beyond Nvidia.
The AI industry is becoming increasingly capital-intensive.
The best models require enormous computing resources.
The best engineers are expensive and highly sought after.
And companies with valuable technology can become strategic targets long before they are mature enough for a conventional acquisition.
That creates an interesting middle ground.
You don’t always need to buy the company to control access to what makes the company valuable.
License the technology.
Hire the team.
Invest in the company.
Leave the remaining business independent.
That may be enough.
For Nvidia, this approach can provide access to important technologies and talent without necessarily absorbing entire organizations.
For startups, it can provide capital and an exit of sorts without completely shutting down the business.
For shareholders, however, it means the old rules of reading an acquisition headline don’t always apply.
The question investors should be asking
The next time you see a headline saying that Nvidia has committed billions to an AI startup, don’t stop at the number.
Ask what Nvidia actually bought.
Did it buy the company?
Did it buy the technology?
Did it license the intellectual property?
Did it hire the founders and engineers?
Did it invest fresh capital?
And most importantly, what happened to the shareholders who stayed behind?
Those questions may tell you far more than the headline valuation.
Because in today’s AI market, the biggest deal may not be the one that buys the company.
It may be the one that buys everything that made the company valuable in the first place.
Data point of the day
1.1x
That is the reported return received by investors in Inflection AI’s later $1.3 billion funding round when Microsoft licensed its technology and hired most of its staff.
Earlier investors reportedly received around 1.5x.
The company was never acquired.
That range is a useful reminder that these transactions can produce very different outcomes for different shareholders.
The headline deal value is only the beginning. The terms determine where the money actually goes.
Source: Augment Markets, “What Nvidia Buys Without Buying the Company,” published August 27, 2026.
If you want, I can also turn this into a more punchy Vested-style community article with a stronger opening hook, shorter paragraphs, and more discussion-provoking sections.