Sovereignty, Dominion, and Return

On the third of June, on a stage at Computex in Taipei, Robert F. Smith did something that told you exactly where the patient money is headed. He did not raise another fund. He turned on a cloud.

Vector Core Compute, built with Cambium and anchored by a three and a half billion dollar compute commitment to SambaNova, went live that same afternoon from a facility in Los Angeles. He called it the first enterprise AI inference cloud. Strip away the keynote staging and the message was plain. The man who built one of the largest software fortunes in history had decided the next decade gets won at the layer almost nobody toasts at dinner: inference, the unglamorous cost of actually running the models long after the launch confetti is swept up.

I do not write this from a hundred billion dollar perch, and I will not pretend to. I write it as someone who watches these tools closely and believes the shift Smith is betting on reaches well beyond his own portfolio. Let me use his move to explain what I believe is coming next, and where I think it lands first in private equity, which is diligence.

What Smith actually saw

For two years he has argued one idea with unusual consistency. The largest returns in this cycle will not accrue to the companies pouring concrete for data centers. They will accrue to the enterprise software companies that push AI directly into the workflows their customers already depend on, in banking, in insurance, in the regulated corners where being wrong is expensive and proprietary data cannot be copied. Vista calls the machine it built for this an agentic factory. By his own count at the launch, more than half of Vista's ninety plus portfolio companies have already converted to agentic solutions, and he expects most of the rest to follow. He has sized the wave behind it at roughly three trillion dollars flowing into AI infrastructure.

What I admire is not the size of the bet. It is the location of it. Smith does not chase the model of the week. He keeps asking a quieter question, whether a company truly owns its workflows and its data, because that is the part no competitor can replicate. He has put it without a shred of hedging:

Robert F. Smith, on Bain & Company's Dry Powder

That principle has a consequence worth sitting with. If a company's workflows and its data are its sovereign territory, it will open that territory to very few. Large institutions will share their most valuable data only with partners who are a proven quantity, firms with a track record they trust not to leak it, misuse it, or mishandle it. In the agent economy, access to the data follows trust, and trust is earned over time, not bought with a pitch. That is the quiet gate standing in front of every firm that wants to do this work, and it is why the proven keep winning the work even after the technology is everywhere.

Vector Core Compute is that same question pushed down one layer. If inference is the recurring cost that decides whether agentic software is genuinely profitable or merely impressive, then owning the economics of inference is not a side wager. It is the board. And cheap inference is not a technical footnote. It is what lets agentic software run efficiently and at a far lower cost, which is what finally makes it economical to put an agent on a thousand workflows instead of a chosen few. Vector Core Compute is, underneath the branding, an efficiency engine. It exists to drive the cost of doing the work down. He also told a room at Brainstorm Tech to stop gutting their intern programs in the rush to automate, which is the sort of thing a person says when he is thinking about the next twenty years and not the next quarter. That instinct, capacity now, judgment kept human, is the whole argument of this paper.

Notice, too, how he did it. Not alone. He built Vector Core Compute through partnership, with Cambium and with SambaNova, and he chose to unveil it from Intel's stage in Taipei, at the center of the world's chip supply. The agentic era is being assembled in alliances, across the companies and the borders that can actually deliver the compute. That is its own lesson, and it is not a small one. The firms that win this will be the ones that partner well, not the ones that insist on owning every layer alone.

The disciplined other half

Orlando Bravo, who built Thoma Bravo into a software house managing more than a hundred and eighty one billion dollars, holds the other half of this temperament. He will tell you flatly that AI valuations are in bubble territory. "Venture firms are just piling into any AI story they can," he has said, and that line is a warning as much as an observation. In the same breath he insists that enterprise software itself is far from dying. He has gone so far as to declare the so called SaaSpocalypse over, arguing the market wrote software's obituary far too early, and pointing to a tailwind he measures in the trillions. Inside his own portfolio, close to half of new revenue now comes from AI and agentic tools. And he is not sitting out the era he is skeptical of. Thoma Bravo's own thesis is that software and AI tools will merge into what the firm calls agentic solutions for corporate customers, and it has kept acquiring software companies precisely as AI fears push their valuations down, taking the workforce software maker Dayforce private in a deal worth more than twelve billion dollars. Bravo is buying into the agentic future on his own terms. Two of the most successful software investors alive are saying the same thing in two keys. The capability is real. The price is frequently absurd. Learn to tell the difference and you keep your shirt.

What agentic AI actually does, and where it is going

It helps to be concrete, because the word gets thrown around loosely. A generative tool writes you a memo. An agentic system does the work. It pulls the filings, runs the model, drafts the synthesis, checks itself against the evidence, and hands a senior reviewer something close to finished. Smith's portfolio is the proof of scale, with more than half of ninety plus companies already converted and agents now earning real revenue in banking, insurance, and the regulated places where being wrong is expensive. The near future is not a chatbot answering questions. It is software that carries the load of an analyst or a junior associate, at a fraction of the cost, around the clock, and improving with every cycle.

Make no mistake about what this is and what it is not. This is not the decline of software. Software is the bread and butter of this entire economy, and it is not going anywhere. The shift is that software now has to act, not merely inform. The companies that make that leap, the ones that stay current, keep their value and often expand it. The ones that treat today's product as a permanent state get repriced by the market. Staying current is no longer a competitive nicety. It has become the price of staying relevant.

There is a deeper idea running under his bets, and he returns to it often. Every business paradigm works only inside a given equilibrium, and equilibria do not hold still. The on prem world had one. The cloud created another. Agentic AI is collapsing that one and forming the next. His conclusion is not to defend a paradigm but to treat evolution as the natural state, and to build for the process of change rather than for any single instance of it. For an investor, that is the entire posture shift. You are no longer buying a company's present equilibrium. You are buying its capacity to reach the next one. And the same test now applies to the firms you hire to evaluate it.

Why diligence is the first domino

Bring it back to diligence, because that is where this reaches most of us first, and the compression is already measurable. Deloitte's 2025 study found that eighty eight percent of private equity firms have committed a million dollars or more to generative AI for deals, and that thirty five percent of adopters already use it to screen targets and run diligence. Some teams report cutting financial modeling time by seventy percent. McKinsey puts the gain in deal origination at up to thirty percent. Bain, surveying investors who hold three point two trillion dollars in assets, found most portfolios already testing the technology and close to a fifth running it in production. A research primer that once took a week now takes an hour. A data room that took a fortnight is read before lunch.

The premium diligence model was built for the opposite world, where information was scarce and junior hours were the product, sold by the week at prices only the largest funds ever paid without flinching. Both of those inputs just inverted. A firm still selling the old model is charging a premium for the very part the machine now does on its own.

Smith has lived through a rerating like this before, and he says so plainly. "I think we've converted more companies from on-prem to the cloud than any institution on the planet," he has noted, "but what it did was it ushered in a major rerating of enterprise software ... our sense is this is going to be the same sort of dynamic." He was talking about software. The same rerating is now coming for the firms that analyze it.

The trigger is a person, not a model

So what actually resets the market. Not a model release, and not a benchmark score. A person. The reset arrives the morning a managing partner routes a live deal to a firm that pairs that new speed with senior review, because that firm sees more and can stand behind every line of it. That is a human decision about trust, and trust does not move with a logo. The technology turned out to be the easy part. Deciding whom to believe is the hard part, and it always was.

Who finally gets into the room

Here is the part I care about more than the league tables. For thirty years, the institutions most in need of rigor could not buy diligence at six figures a week. The historically Black colleges and the minority serving institutions. The community foundations. The public agencies. The sovereign nations. Rigor at that level was simply not for them, and everyone in the room understood it. The same compression that squeezes the incumbents quietly rewrites who can afford rigor at all. When a serious, senior reviewed study costs a fraction of what it once did, the people who were always told this work was not meant for them can finally commission it. Most coverage frames this era as a contest over which brands survive. The larger story is who walks into the room for the first time.

Speed without governance is a liability

One caution keeps the whole thesis honest. Speed without governance is not an edge, it is exposure. The same system that reads a data room in minutes can also invent a covenant, miss a concentration risk, or quietly inherit the bias sitting in its training data. The firms that earn a partner's trust will be the ones that can show their work, where the data lived, how the model reasoned, what a senior practitioner reviewed, and whose name is on the result. Owning your inference layer and owning your review process are the same instinct at different altitudes. Control the part that decides whether the output can be trusted.

The part that stays human

For all the speed, a surprising amount of this work stays stubbornly human, and it is worth being honest about which part. A model can draft a memo in minutes. It cannot care about the outcome, read the politics in a boardroom, sit with a nervous founder, or carry the accountability when a call goes wrong. Someone senior still has to own the judgment and put a name on the result. That layer does not disappear in the agentic era. It becomes the scarce thing, and the valuable one. The right model is not the machine replacing the practitioner. It is the machine carrying the load underneath, and a senior person standing on top of it, signing the work.

The close

Smith committed billions to own the economics underneath the agentic era, and he was right to. The rest of us own it with discipline, with senior eyes on the work, and with a refusal to mistake a fast answer for a sound one. The firms that pair real agentic capacity with real judgment will not be waiting for the reset. They will be the reason it happens. And for the first time in my career, the institutions that were always priced out will be in the room when it does.