Nvidia has found an interesting way to get access to some of the most valuable parts of AI startups without actually buying the startups themselves.
Over the past nine months, the company has been involved in reported multibillion-dollar deals with Groq and Poolside, while also reportedly discussing an investment in Perplexity.
The common thread?
Nvidia is buying technology, licensing intellectual property and bringing key people into the company, while leaving the original startup standing.
That may sound like a technical distinction. For private-market investors and shareholders, it is anything but.
When a company is not acquired, shareholders do not automatically get a traditional acquisition payout. Instead, the value can be split across licensing fees, investments, employee compensation and shareholder distributions.
That makes these deals much harder to understand from a headline valuation alone.
The new AI deal playbook
The traditional startup exit is easy to understand.
A large company acquires a startup, shareholders receive proceeds according to the deal terms, employees may receive payouts or accelerated vesting, and the acquired company becomes part of the buyer.
Nvidia’s recent deals look very different.
In reported transactions involving Groq and Poolside, Nvidia has pursued a structure where it gets access to the technology and talent it wants without taking control of the entire company.
The reported numbers are huge:
- Around $20 billion for the Groq transaction
- Around $7 billion in combined licensing and investment for Poolside
- More than $30 billion valuation reportedly being discussed for a potential Perplexity investment
The Perplexity discussions are separate and are not included in the $27 billion figure.
And importantly, the reported terms of these transactions have not been fully confirmed by the companies involved.
That distinction matters because a $20 billion transaction does not necessarily mean shareholders received $20 billion in the same way they would in a conventional acquisition.
Groq: Nvidia got the technology and the team
Groq is probably the clearest example of how this structure works.
In December, Nvidia reportedly agreed to a transaction worth approximately $20 billion.
But Nvidia did not acquire Groq outright.
Instead, the companies described the arrangement as a non-exclusive license to Groq’s inference technology, alongside substantially all of its assets.
At the same time, Groq’s founder and CEO Jonathan Ross, president Sunny Madra, and other senior leaders joined Nvidia.
The remaining Groq business continued independently.
That distinction is important.
Nvidia got access to the technology and people it wanted, while GroqCloud remained behind as a separate company.
The reported transaction was also significantly higher than Groq’s previous private valuation.
Groq had reportedly raised money at a $6.9 billion valuation in September 2025.
Just a few months later, the reported Nvidia transaction was approximately three times that figure.
But the company that remained after the transaction was no longer the same company investors had previously valued.
What happened to the Groq that remained?
This is where things get particularly interesting for private-market investors.
The continuing Groq business has since raised additional capital.
TechCrunch reported in August that Groq raised $350 million at a $3.5 billion valuation, following a reported $650 million round in June.
That means the surviving company has raised roughly $1 billion during 2026.
Its business has also evolved.
Groq now operates as an inference cloud company using its own processors alongside Nvidia hardware.
So there are effectively two very different numbers attached to the Groq story:
The reported $20 billion Nvidia transaction
and
the $3.5 billion valuation attached to the continuing Groq business.
These numbers are not directly comparable.
The $20 billion transaction involved technology, assets and talent that left the original company.
The $3.5 billion valuation reflects the business that remained.
That is the key lesson: the value of the transaction and the value of the surviving company can be two very different things.
Poolside shows another version of the strategy
Poolside followed a similar but slightly different route.
Newcomer reported a deal involving:
- $6 billion non-exclusive license for Poolside’s Model Factory software
- $1 billion investment at a reported $12 billion pre-money valuation
- Offers to 109 employees who built its Laguna model
Poolside itself described the transaction as not an acquisition and not an acquihire.
The company’s circumstances also highlight why a startup might agree to this kind of arrangement.
According to its shareholder letter, Poolside had roughly six weeks to raise $2 billion to build a 40,000-GPU cluster.
It missed that deadline and lost access to the cluster.
The Nvidia deal therefore provided a very different path forward.
Instead of selling the company outright, Poolside could license important technology, bring in an investor and continue operating as an independent business.
For shareholders, however, the important question becomes:
How much of the reported deal value actually reaches them?
That answer depends on the structure.
Perplexity could take the strategy even further
Perplexity is a particularly interesting case because the reported discussions appear to be moving beyond licensing.
According to reporting cited in the original piece, Nvidia is discussing a straight equity investment in Perplexity at a valuation above $30 billion.
Perplexity’s last reported valuation was $20 billion in September 2025.
Its reported annualized revenue has also grown sharply, from less than $250 million to more than $750 million this year.
No agreement has been announced, and the discussions may not result in a transaction.
Still, the potential deal matters because it shows that Nvidia’s interest in AI companies is not limited to acquiring companies outright.
It can choose from several structures depending on what it wants:
- License the technology
- Invest directly
- Bring key employees into Nvidia
- Acquire substantially all assets
- Or potentially pursue a traditional acquisition
The structure can be tailored to the specific company and the specific assets Nvidia wants.
Why wouldn’t Nvidia just buy the whole company?
That is the obvious question.
If Nvidia wants the technology and talent, why not simply acquire the startup?
One reason is flexibility.
A traditional acquisition brings the entire company into the buyer. That means taking on the full business, its employees, contracts, liabilities and operations.
A licensing arrangement can be much more targeted.
Nvidia can get access to the technology it wants without necessarily taking ownership of every part of the business.
The same applies to talent.
If the founders and key researchers are the real prize, bringing them into Nvidia can accomplish much of what an acquisition would accomplish from a technology-development perspective.
There is also a regulatory angle.
Because these arrangements do not necessarily involve a change of control, they can face different merger-control or premerger-notification requirements than a conventional acquisition.
That does not mean regulators cannot review them.
Competition authorities can still scrutinize arrangements that may affect competition.
But structurally, a license plus hiring can look very different from buying an entire company.
The shareholder question is where things get complicated
For anyone holding private shares, the biggest mistake would be to look at the headline transaction value and assume that is the value being paid for their shares.
A license fee is not the same thing as a per-share acquisition price.
In a traditional acquisition, the purchase price is generally tied directly to the company being acquired.
In these newer structures, the money can go to different places.
For example:
- A license fee may be paid to the company
- A new equity investment goes into the company and changes the cap table
- Employee compensation goes directly to people joining the buyer
- Some proceeds may eventually be distributed to shareholders
- Existing investors may remain invested in the company that continues operating
That means two investors in the same startup could potentially experience very different outcomes depending on their rights, preferences and the deal’s distribution mechanics.
The headline number tells only part of the story.
We’ve seen this before
Nvidia is not the first major technology company to use this playbook.
There are already several examples from 2024 and 2025 that show how these transactions can play out.
Microsoft and Inflection AI
In March 2024, Microsoft reportedly agreed to pay around $620 million for licensing, alongside a reported $33 million legal waiver, while hiring Inflection’s co-founders and much of its staff.
Inflection itself was not acquired.
Reported shareholder outcomes varied:
- Investors in an earlier $225 million round reportedly received around 1.5x
- Investors in a later $1.3 billion round reportedly received around 1.1x
The company continued under a new CEO and shifted toward enterprise sales.
The lesson?
Even without an acquisition, investors can still receive liquidity, but the outcome depends heavily on the transaction terms.
Google and Character.AI
Google used another version of the structure with Character.AI in August 2024.
Alphabet later disclosed a $2.7 billion cash payment for a non-exclusive license.
Character.AI co-founder Noam Shazeer and other employees returned to Google.
The reported payment provided liquidity for investors and employees at around $2.5 billion, above the $1 billion valuation from the company’s 2023 round.
Again, there was no conventional acquisition.
Yet shareholders and employees could still receive liquidity.
Windsurf shows how complicated it can become
Windsurf’s situation in 2025 was even more complicated.
OpenAI had agreed to acquire the company for roughly $3 billion, but that deal collapsed.
Google then reportedly paid around $2.4 billion to license Windsurf’s technology and hire its CEO and co-founder.
Days later, Cognition acquired what remained of the business.
Cognition said remaining employees would participate financially and receive accelerated vesting.
The sequence demonstrates how quickly ownership, technology rights and talent can become separated.
The startup itself can survive in one form while its technology and key people move elsewhere.
The biggest lesson: don’t confuse valuation with value
This may be the most important point for anyone following private markets.
Suppose you see a headline saying a company was involved in a $20 billion deal.
That does not automatically mean:
“The company is now worth $20 billion.”
And it certainly does not necessarily mean:
“Every shareholder receives a portion of $20 billion.”
You need to understand what the money is actually paying for.
Is it:
- A purchase of shares?
- A technology license?
- An acquisition of assets?
- An investment into the company?
- Compensation for employees?
- A combination of several of these?
Each answer can produce a very different outcome for shareholders.
The terms behind the headline matter more than ever
This is where private-market analysis gets interesting.
For a public company, investors have extensive disclosure around financials and transactions.
Private-company transactions can be much harder to evaluate.
The headline may tell you the transaction value, but not necessarily:
- Who receives the money
- How much is distributed to shareholders
- Whether distributions happen immediately or over time
- How employee vesting is handled
- What happens to existing investors
- Whether the company continues operating independently
- What technology remains with the company
- What technology moves to the buyer
For investors looking at secondary-market pricing, these details can make a huge difference.
A share in the continuing company after a major technology license may represent something very different from a share in the company before the transaction.
A reverse acquihire is not a traditional exit
The term increasingly used for these arrangements is reverse acquihire.
In simple terms, it describes a situation where a large company:
licenses a startup’s technology and hires its founders or key employees without acquiring the startup itself.
The consideration can then be divided between several buckets:
- Technology licensing payments
- Employee compensation
- New investment
- Shareholder distributions
- Other negotiated payments
Because the company is not formally acquired, the transaction can leave a separate business behind.
That business might continue to raise capital, change its strategy or build around whatever assets remain.
The startup doesn’t necessarily disappear. It can emerge as a smaller or fundamentally different company.
Groq is a useful example of that evolution.
Is this good or bad for shareholders?
There is no universal answer.
It depends on the terms.
There is a strong argument in favor of these transactions.
If a startup is struggling to raise enough capital, a major licensing deal can provide a lifeline.
Poolside’s situation illustrates this clearly.
The company had a large infrastructure funding requirement and missed its original deadline. A major deal with Nvidia gave it another path rather than forcing an outright sale.
For shareholders, that can be better than a distressed financing or a much lower valuation.
There are also potential advantages to remaining independent.
The company can continue operating, raising capital and pursuing new opportunities while monetizing technology that a larger player wants.
But shareholders also need to understand what they are left owning.
If the founders, senior team and core technology move to another company, the continuing startup may be very different from the business investors originally backed.
The $1.1x data point tells an important story
One of the most useful figures in the original analysis comes from Inflection AI.
Investors in the company’s later $1.3 billion funding round reportedly received about 1.1x when Microsoft licensed the technology and hired most of the staff.
Earlier investors reportedly received around 1.5x.
That is a striking difference.
It shows why a headline transaction cannot tell you what an individual investor actually makes.
The outcome depends on when the investor entered, what rights they had and how the transaction was structured.
The reported 1.1x and 1.5x figures are transaction-specific outcomes, not expected returns or a template for future deals.
What should private-market investors watch?
When a startup enters one of these transactions, the most useful questions are not simply “What is the valuation?”
Instead, look deeper.
1. What exactly is being sold or licensed?
Is the buyer getting a narrow technology license, substantially all of the company’s assets or something in between?
2. Who is moving to the buyer?
If the founder, CEO and key technical team are leaving, the continuing company may look very different.
3. Where does the money go?
A $6 billion license fee paid to the company is not the same as $6 billion paid directly to shareholders.
4. Are shareholders receiving liquidity?
Some structures provide distributions. Others primarily strengthen the company’s balance sheet.
5. What happens to existing ownership?
A new equity investment can dilute existing holders.
6. What remains of the original business?
This may be the most important question.
What you own after the transaction may not be the same economic opportunity you originally invested in.
Nvidia’s strategy says something bigger about AI
There is a broader story here beyond individual startups.
The AI ecosystem is becoming increasingly competitive for three things:
Compute.
Technology.
Talent.
The largest technology companies do not necessarily need to own every company developing those assets.
Sometimes it is enough to secure access to the technology and people that matter most.
That creates a new kind of strategic transaction.
Instead of:
“Buy the company.”
The question becomes:
“What part of the company actually matters to us?”
If the answer is the model, the inference technology, the intellectual property or the engineering team, a company can potentially structure a deal around those pieces.
The private-market takeaway
Nvidia’s reported deals with Groq and Poolside, along with the reported discussions around Perplexity, highlight a shift worth watching.
Big AI deals do not always look like acquisitions anymore.
A company can remain independent while its most valuable technology and people move to a much larger player.
For private shareholders, that makes transaction analysis more important than ever.
A headline like “$20 billion deal” can sound like a simple valuation event.
It isn’t.
The real story is in the details:
- What was licensed
- What assets changed hands
- Which employees moved
- How much money went to the company
- How much reached shareholders
- What remained behind
- What the continuing company is ultimately worth
That is why the Groq story is particularly revealing.
Nvidia reportedly paid around $20 billion for a transaction involving its technology and assets, while the continuing Groq later raised money at a $3.5 billion valuation.
Those two numbers are not contradictory.
They represent different pieces of the same company after its value was split apart.
And that may be the most important thing to remember when the next multibillion-dollar AI deal hits the headlines.
Don’t just ask how much the deal is worth. Ask what, exactly, was bought.