Earlier this week, in Part One of this report, I made the case that Brookfield’s decision to commit $500 million to the newly formed OpenAI Deployment Company—DeployCo—was one of the more important signals I have seen in the enterprise AI economy. Not because Brookfield suddenly discovered artificial intelligence, not because OpenAI needs another famous institutional investor and cohort of banks behind it, and certainly not because every CEO should run out tomorrow and standardize their company on one model provider.
The signal was much more interesting than it was loud.
Some of the most sophisticated owners of businesses in the world have concluded that AI deployment itself is becoming an asset class.
For most of the last several years, nearly all of our attention has been directed toward the model layer. Which model is smartest? Which has the largest context window? Which scores highest on the benchmark? Who has the most compute? Who is going to win? Meanwhile, something considerably more durable has been forming underneath all of that noise: companies have started putting these models to work.
Once that happens, the economic center of gravity begins to move.
The model may generate the intelligence, but the deployment process begins generating something potentially much more valuable: proprietary intelligence about the enterprise itself. It begins capturing how the company works, how decisions get made, how customers behave, where exceptions occur, which rules matter, which approvals are required, what the organization knows that is not written down, how its best people solve problems, which workflows produce economic value, and eventually how thousands or millions of interactions can be converted into institutional memory.
That is why I believe the most important question emerging from the Brookfield/OpenAI transaction is not, “Should I invest in OpenAI?” It is not even, “Should my company use OpenAI?”
The much more important question is:
Who is going to own the deployment layer through which my company’s intelligence is being created?
There are really three different transactions hiding inside this one deal:
A capital deal,
A distribution deal, and
An intelligence deal.
Brookfield may have found a way to make all three attractive simultaneously.
Most enterprises and almost none of my readers, will not have that luxury. So why should you keep reading or care?
The Μost Ιmportant Νumber Μay Νot Βe $500 Μillion
Start with the number everyone saw. Brookfield announced in May that it had agreed to invest $500 million in The OpenAI Deployment Company, a newly formed AI deployment platform created with OpenAI and a group of global investment firms, consultants, and systems integrators.
OpenAI described a coalition of 19 partners led by TPG, with Advent, Bain Capital, and Brookfield serving as co-lead founding partners. Other named participants include B Capital, BBVA, Emergence Capital, Goanna, Goldman Sachs, SoftBank Corp., Warburg Pincus, and Welsh, Carson, Anderson & Stowe, alongside consulting and systems-integration firms including Bain & Company, Capgemini, and McKinsey & Company. OpenAI said the initial investment behind the company exceeds $4 billion, and public filings state that the post-money valuation of Deploy Co is around $14B.
That alone is remarkable. But I do not think it is the most interesting number.
The number I keep coming back to is less publicly known: 17.5%.
Press reporting around the formation of DeployCo indicated that OpenAI was offering private-equity investors a guaranteed minimum annual return of approximately 17.5% over five years as part of the economic package surrounding the venture. There are important qualifications here. We do not have the operating agreement. We do not have the full preferred-equity waterfall. We do not know from public disclosures or the hundreds of footnotes researched this week, whether that reported return functions as a coupon, preferred return, IRR floor, redemption obligation, make-whole, or some combination. We do not know whether it compounds. We do not know all of the side-letter economics.
And I would not state that Brookfield itself has publicly confirmed that its specific security carries precisely the reported 17.5% term.
What Brookfield has subsequently confirmed in their June filing is almost as interesting. In its second-quarter disclosure, Brookfield Business Corporation said the investment had closed and described its position explicitly as a preferred equity investment providing an “attractive contracted return” while also giving Brookfield access to OpenAI’s models, technology, and engineering talent.
That changes how I think we should interpret this deal because Brookfield is not simply making a directional bet on an AI consulting business. It appears to be buying a structured financial return, strategic access, and potential operating leverage across a massive portfolio of businesses.
Those are very different economics from those facing the ordinary enterprise customer.
Signal, Noise, and Subsidy
There are three buckets I would put the DeployCo story into.
The Signal
Deployment has become the bottleneck. Brookfield said it directly. The company described the next opportunity as “execution at scale”—moving beyond pilots and using AI across operating businesses to improve productivity, decision-making, and efficiency.
OpenAI is saying essentially the same thing. DeployCo is being constructed around Forward Deployed Engineers, or FDEs, who work inside customer environments alongside business leaders, operators, and frontline employees. Their job is not simply to install ChatGPT. It is to identify valuable problems, connect models with company data and tools, redesign workflows, and turn experiments into durable production systems.
The model race is becoming a deployment race.
The Noise
Much of the shorthand around this transaction obscures what actually happened. Brookfield did not simply “invest $500 million in OpenAI.” The investment is in DeployCo. No public evidence shows that Brookfield’s investment gives it meaningful ownership of OpenAI’s parent-company equity. There is also no public evidence that Brookfield has permanently selected one model architecture for every operating company it controls.
Additionally, I couldn’t find any public basis for concluding that DeployCo’s investor economics prove that DeployCo itself will generate extraordinary operating margins, and there is certainly no basis for concluding that because Brookfield partnered deeply with OpenAI, every enterprise should do the same thing.
Those are very different propositions.
The Subsidy
Then comes the most interesting part. OpenAI is not merely building a consulting company. It is creating an economic coalition around deployment.
The investment partners collectively sponsor thousands of businesses. The consulting and systems-integration partners touch thousands more. That potentially gives OpenAI something far more valuable than another advertising campaign: economically motivated enterprise distribution.
OpenAI is doing exactly what a rational platform company should do. It wants its intelligence embedded deeply inside real operating businesses. Brookfield is doing exactly what a rational asset owner should do. It wants financial return, operating leverage, engineering access, and potentially higher EBITDA and enterprise value throughout its portfolio.
In my humble opinion, there is nothing nefarious about either incentive, but if you are the end-user client sitting between them, you should understand those incentives before deciding what architecture to build.
Follow the incentives before you follow the characters’ architecture blindly.
What Brookfield is actually buying
Consider the potential return stack. First, Brookfield owns a preferred-equity security with a contracted return. Second, it gains access to frontier AI capabilities and deployment expertise. Third, it can potentially deploy those capabilities across hundreds of operating companies. Fourth, if those deployments improve margins, revenue growth, labor productivity, or capital efficiency, Brookfield may benefit from higher EBITDA and therefore higher enterprise values. Fifth, it gets something much harder to quantify: knowledge optionality.
Imagine the informational advantage created when an owner operating across infrastructure, industrials, real estate, energy, and business services repeatedly learns which AI deployments actually produce measurable economic value. Brookfield does not necessarily need DeployCo itself to generate the entire economic return. Some of the value can materialize elsewhere.
A few points of EBITDA improvement across a sufficiently large portfolio can dwarf the economics of the deployment company that helped produce them.
That is why I do not think the right conclusion is that Brookfield made a mistake by concentrating around OpenAI. Brookfield may have made an extraordinarily rational deal for Brookfield.
The problem begins when everyone else assumes Brookfield’s deal architecture should also be theirs.
The Missing Pages
This is where investigative discipline matters. There is a temptation in technology and financial journalism to fill gaps in disclosure with certainty. I would rather identify the gaps.
Here is what we still don’t know:
We still do not publicly know the complete preferred-equity waterfall. We do not know all redemption rights.
We do not know the exact economics of the reported minimum-return arrangement. We do not know every side letter.
We do not know the extent of customer exclusivity provisions, if any.
We do not know whether portfolio companies will have minimum spending obligations.
We do not know all revenue-sharing arrangements.
We do not know exactly how reusable workflow intellectual property will be treated.
We also do not know who ultimately owns every layer of knowledge created by FDEs working inside a customer’s organization, nor do we yet know how portable that accumulated intelligence will be if an enterprise eventually wants another model provider doing the inference.
Those are not accusations. They are open questions that could dramatically evolve the analysis of this transaction. Throughout the rest of this analysis, I think we need four labels: KNOWN. REPORTED. INFERRED. UNKNOWN. If you are making a seven-, eight-, or nine-figure AI allocation decision, the differences between those words matter.
