The latest AI infrastructure deals are getting bigger, more interconnected and more complicated. Nvidia’s $105 billion guarantee for an OpenAI data center, Groq’s sharply lower valuation and Anthropic’s rapid revenue growth offer three very different windows into where the AI market is heading.
The AI boom is no longer just about who has the best model.
Increasingly, the bigger question is who is financing the infrastructure needed to run these models, who owns that infrastructure, who supplies the chips, and what happens if the economics do not work out as planned.
That is what makes Nvidia’s latest agreement with OpenAI particularly interesting.
Nvidia has agreed to provide a conditional guarantee of up to $105 billion tied to the first phase of a major data center campus in Ohio that OpenAI plans to lease for 20 years. In return, Nvidia gets exclusive chip-selling rights for that first phase and a $1.5 billion stake in SB Energy, the SoftBank unit building and operating the campus.
At the same time, another Nvidia-linked deal is showing the other side of the AI story.
Groq has raised $350 million at a $3.5 billion valuation, roughly half the valuation it reached in September 2025. Meanwhile, Anthropic’s annualized revenue run rate has surged to $65 billion, highlighting just how quickly the commercial side of AI is scaling.
Put together, these developments show a market where capital, chips, data centers and AI companies are becoming increasingly intertwined.
Nvidia is taking a much bigger role in OpenAI’s infrastructure
The Ohio project is being developed by SB Energy, a SoftBank unit, at the PORTS-Pike site in Portsmouth, Ohio.
The location is notable because the land includes a former federal uranium enrichment plant.
The planned campus is enormous. It is designed for 8 gigawatts of IT capacity, supported by 10 gigawatts of new generation.
OpenAI plans to lease the campus for 20 years.
But the headline figure surrounding the agreement needs some context.
Nvidia is not simply writing a $105 billion check to OpenAI.
The guarantee applies to the first phase of the project, which represents roughly half of the site. It would only pay out if OpenAI defaults on the lease and SB Energy is then unable to re-lease or resell the campus at a similar valuation.
That makes this a residual value guarantee, rather than an immediate $105 billion commitment.
The distinction matters.
A headline guarantee and the amount Nvidia might ultimately have to pay are two very different things.
There are multiple conditions that have to occur before the guarantee is triggered.
So what does Nvidia get in return?
The arrangement is interesting because Nvidia is not taking the infrastructure risk without receiving something in return.
It gets exclusive rights to sell chips into the first phase of the Ohio campus.
Nvidia is also investing $1.5 billion in SB Energy, joining SoftBank Group and OpenAI as investors in the landlord.
That creates a particularly tight relationship between the companies.
Nvidia is helping support the landlord.
OpenAI is the tenant.
Nvidia is also the exclusive chip supplier for the first phase.
In other words, the same infrastructure project brings together the AI model company, the chip supplier and the company responsible for the data center.
That is a sign of how capital-intensive AI infrastructure has become.
The economics of building an enormous AI campus cannot be separated from the economics of the companies expected to use it.
And Nvidia is increasingly participating directly in those economics.
The $105 billion headline is not the same as $105 billion of exposure
This is probably the most important detail to understand.
The guarantee only comes into play under a specific set of circumstances.
First, OpenAI would have to default on the lease.
Second, SB Energy would have to be unable to re-lease or resell the campus at a similar valuation.
Only then would the guarantee potentially be called upon.
That means the $105 billion figure represents the maximum amount covered by the guarantee under the agreed conditions, not necessarily the amount Nvidia expects to lose.
This is also why the term residual value guarantee matters.
A residual value guarantee is essentially a promise to cover a shortfall if an asset ends up being worth less than an agreed amount.
For a project as large and specialized as an AI data center, that protection can become particularly important.
The infrastructure is expensive, highly specialized and designed around rapidly changing technology.
If demand for AI infrastructure continues growing, the assets could retain significant value.
If demand weakens or the economics change dramatically, the ability to re-lease or resell that infrastructure becomes much more important.
Nvidia is becoming more than a chip supplier
The Ohio agreement also fits into a broader pattern.
Nvidia is increasingly appearing on both sides of major private AI transactions.
It is supplying the chips that power AI infrastructure.
It is investing in AI companies.
It is providing financial support for infrastructure.
And it is becoming directly involved in the economics of the companies building and operating that infrastructure.
The Groq deal is another example.
Nvidia recently licensed Groq’s technology and hired founder and CEO Jonathan Ross along with much of the team.
Nvidia is also an investor in Groq’s latest funding round.
That means Nvidia is not simply benefiting from AI companies buying its chips. It is becoming increasingly connected to the companies and infrastructure that make up the wider AI ecosystem.
Groq shows that AI valuations can move sharply
While Nvidia’s Ohio agreement highlights the enormous amount of capital flowing into AI infrastructure, Groq provides a useful reminder that not every AI valuation is moving higher.
Groq has raised $350 million at a $3.5 billion valuation.
That is roughly half the $6.9 billion valuation the company reached in September 2025.
At first glance, that looks like a major down round.
But there is an important wrinkle.
Groq says it does not view the new valuation that way because the business itself has changed.
After Nvidia’s licensing agreement and personnel moves, Groq has shifted away from being primarily an AI chip designer.
The company now operates data centers and sells AI inference capacity.
So the businesses being valued in September 2025 and August 2026 are not identical.
That makes a simple comparison between the two valuations less useful.
It is another example of why private-market valuations can be difficult to interpret.
A number on a funding announcement can look straightforward, but the underlying business, ownership structure, financing terms and strategic relationships can change significantly between rounds.
Then there is Anthropic
While Groq’s valuation has moved lower, Anthropic is showing just how quickly AI demand can translate into revenue.
Anthropic told investors that its annualized revenue run rate reached $65 billion at the end of July.
That compares with:
- $47 billion in May
- Around $9 billion at the end of 2025
The company also reported more than $11.5 billion in preliminary second-quarter revenue.
The growth is striking.
But there is one important distinction.
A revenue run rate is an annualized estimate based on recent performance. It is not the same thing as revenue actually generated over a full year.
That distinction becomes especially important when discussing rapidly growing private companies.
A company can have an enormous run rate without having generated that amount of revenue over the previous 12 months.
Still, the pace of Anthropic’s growth provides another indication of how quickly demand for AI services is expanding.
Three very different signals from the same market
Put these developments together and the AI market looks much more complicated than a simple story of valuations going up.
Nvidia and OpenAI show the infrastructure side.
Huge amounts of capital are being committed to data centers, power generation and chips. The scale of these projects is becoming almost difficult to comprehend.
Groq shows the valuation side.
A company can see its valuation fall substantially even during an AI boom, particularly when its business model changes.
Anthropic shows the revenue side.
AI companies with strong commercial demand can grow revenue at extraordinary rates.
These three stories can all happen at the same time.
That is important because the AI market is not one single trade.
There are chip companies, model developers, cloud providers, data center operators, infrastructure financiers and investors, all with different economics.
The bigger question is who is carrying the risk
The Ohio agreement raises an important question for investors and the broader market.
As AI infrastructure gets bigger, where does the risk ultimately sit?
Building an 8-gigawatt data center campus requires enormous upfront capital.
Someone has to finance the construction.
Someone has to provide the chips.
Someone has to sign the lease.
Someone has to finance the power generation.
And someone has to bear the risk if demand turns out to be lower than expected.
In this case, the risk is distributed across several parties.
SB Energy is building and operating the campus.
OpenAI is the tenant.
Nvidia is providing the guarantee and supplying the chips.
SoftBank is involved through SB Energy.
That interconnected structure can make large AI infrastructure projects possible, but it also means the financial relationships between companies become increasingly important.
AI infrastructure is starting to look like its own financial market
The scale of these commitments is also changing the way investors need to think about AI.
It is no longer enough to ask whether AI demand is growing.
The next questions are about how that demand is being financed and how the infrastructure is being funded.
The source material points to roughly $3 trillion in off-balance-sheet AI commitments across nine large technology companies, including about $1.2 trillion in leases that have not yet started and $1.9 trillion in purchase commitments for chips, power and infrastructure.
That is a huge amount of future spending.
It also highlights why financing structures are becoming increasingly important.
A company can announce massive AI capacity without paying for the entire infrastructure upfront.
Leases, guarantees, purchase commitments, debt and outside investors can all spread the financial burden across different parties.
That can accelerate construction.
But it can also make the underlying risks harder to see.
The next thing to watch: Nvidia’s private-market footprint
Nvidia’s role in private markets is becoming increasingly significant.
Crunchbase News counted 59 known funding rounds involving Nvidia so far this year, compared with 53 for all of 2025.
The combined value of rounds involving chip-company participation was reported at more than $250 billion year to date.
That does not mean Nvidia is responsible for all of that capital.
But it does show how frequently the company is appearing in major AI and technology financing rounds.
For investors, the interesting question is whether this trend continues.
If Nvidia keeps investing in companies across the AI stack, its position becomes broader than that of a semiconductor company.
It becomes part of the financing ecosystem supporting the AI buildout.
Debt could become the next major story
Another area worth watching is the growing role of debt in financing data centers.
Nvidia has announced a $500 billion financing initiative involving major financial firms including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR.
The structure has drawn comparisons to securitization markets because it could allow data center debt to be packaged and distributed among investors.
That is potentially significant.
AI infrastructure requires huge amounts of capital, and traditional corporate balance sheets may not be enough to finance everything that needs to be built.
If debt markets become a major source of funding, the AI infrastructure boom could increasingly resemble other capital-intensive industries where specialized assets are financed through structured debt.
The key question will be how that debt is priced and who ultimately holds the risk.
Private AI valuations need more context than ever
The Groq example is a useful reminder here.
A valuation is not always an apples-to-apples measure from one funding round to another.
The company’s business changed.
Its technology relationship with Nvidia changed.
Its personnel changed.
Its operating model changed.
So while $3.5 billion is materially lower than $6.9 billion, the two figures do not necessarily represent the same business at two different points in time.
The same caution applies more broadly to private AI companies.
Reported valuations can be based on limited transactions and may not provide the same transparency as public-market prices.
That is why revenue growth, cash generation, ownership structure, financing terms and strategic relationships all matter.
What this means for the AI boom
The latest developments do not necessarily point to a bubble or a collapse.
They point to something more interesting.
The AI boom is becoming a financial ecosystem of its own.
Nvidia is financing infrastructure.
OpenAI is committing to long-term capacity.
SoftBank is building data centers.
Anthropic is generating rapidly growing revenue.
Groq is reshaping its business after a major strategic deal.
Banks and asset managers are looking for ways to finance the infrastructure buildout.
And investors are increasingly being asked to evaluate not just AI companies, but the complex web of commitments connecting them.
That creates both opportunities and risks.
The biggest AI companies may continue to grow rapidly, but the infrastructure supporting them requires enormous upfront investment.
The more interconnected the system becomes, the more important it will be to understand who owns what, who owes what, who guarantees what and who ultimately carries the downside.
The bottom line
Nvidia’s $105 billion Ohio guarantee is less about Nvidia simply betting $105 billion on OpenAI and more about the increasingly complex way the AI industry is being built and financed.
The guarantee gives Nvidia access to exclusive chip sales while tying it financially to the landlord building OpenAI’s infrastructure.
Groq’s lower valuation shows that even in a booming AI market, individual companies can see valuations reset when their businesses change.
Anthropic’s $65 billion revenue run rate shows the other side of the equation: demand for AI services is growing at an extraordinary pace.
The bigger story is the connection between all three.
AI is moving from a software story into a massive infrastructure and financing story.
And as the numbers get larger, understanding the financial structure behind the headlines may become just as important as understanding the technology itself.
Reported private-company financials and secondary-market indications may be unaudited, incomplete, non-standard, or based on limited transaction activity. They should not be treated as fair value, executable pricing, or a basis for an investment decision. This article is for informational purposes only and does not constitute investment, legal, tax, or accounting advice.