in december 2025, OpenAI took equity in Thrive Holdings, Joshua Kushner’s $10B+ vehicle. OpenAI contributed its research and engineering teams and platform access in exchange for equity, and in one specific deal inside Thrive’s portfolio, Shield Technology Partners, the right to train its models on the platform’s IT services data.
Shield is an MSP rollup. managed service providers. the companies that run IT for small and mid-size businesses.
read that transaction carefully because it is the tell.
OpenAI is not buying MSP margin. they are buying telemetry into every SMB that runs their IT through an MSP.
meanwhile, General Catalyst led a $74M round in Titan, another AI-enabled MSP holding company, from GC’s $1.5B creation fund. roughly half earmarked for AI-driven consolidation in traditional services.
the surface argument for both plays is level two: buy MSPs, automate the manual work, capture the margin expansion. that is not what the smart money is actually paying for.
the read that’s missing
MSPs are the last delivery layer of AI into the 95% of the economy that will never adopt AI directly.
no small manufacturing firm is going to hire an AI transformation consultant. no regional accounting firm is going to build their own agent stack. they are going to call their MSP. and their MSP is going to install whatever their consolidated parent company tells them to install.
whoever owns the MSP layer owns AI adoption for SMB. not through better marketing. through the delivery pipe itself.
and every SMB an MSP serves generates operational data. every ticket. every network event. every incident. every deployment. every security posture check. at one MSP that is noise. at 200 MSPs across a national market that is a real-time panel on SMB operations at a level no incumbent tech vendor holds.
which sectors are hiring. which are laying off. which industries are getting hit by ransomware. which companies are consolidating. all of it visible before it appears in any published statistic.
the operators most bullish on MSP consolidation are explicitly not optimizing for near-term profitability. they are maximizing coverage. maximizing telemetry surface area. maximizing the panel density that only exists once you cross a threshold.
margin expansion is the story they tell the LPs. telemetry acquisition is the actual play.
Up close, a Chuck Close painting is meaningless blobs. Step back, and a face snaps into existence.
the pattern is not just MSPs
Metropolis is not a parking company. it is a real-time telemetry of urban movement. every vehicle in every lot generates data. entry time, dwell time, payment behavior, revisit frequency. the parking business pays for the sensor grid. the telemetry is the asset.
the same pattern runs through every vertical where a service operation crosses the density threshold at which its byproducts become more valuable than its outputs. a consolidated tax advisor network is a telemetry of the SMB economy. a consolidated property manager is a telemetry of residential real estate cash flow. a consolidated insurance broker network is a telemetry of SMB risk.
this is the inversion most operators miss. in the old model the software is the product. in the emerging model the software is the sensor and the panel is the product. the most valuable companies of the next decade will be indistinguishable from telemetries of the economies they operate in.
who else holds sensors
the panel will not be the only sensor grid pointed at the SMB economy. so the question that decides everything: what does it see that the others don’t?
the bank. sees its own book, not the market. a $50k inflow could be revenue, a loan, or an owner’s deposit. the classification lives one layer up, with the bookkeeper. and most structurally, the bank is a counterparty. the owner negotiates against the bank. the pre-event signal is systematically withheld from it. broad but blind.
the payment processor. real-time sight of the card-and-online fragment of revenue, in an economy that runs on invoices and bank transfer. no cost side. no payroll. no P&L. fast and shallow.
the payroll provider. ADP moves markets with one axis and decades of calibration. proof of concept, not competition.
the labs’ agents. every AI agent inside an SMB is itself a sensor. but the agents see how a company works, not what it earns, owes, and pays. workflow, not meaning.
everyone else sees movements. the multiplier layer, the tax advisors and brokers and property managers who actually do the books, sees meaning. classified, reconciled. and meaning plus calibration is the layer that turns out to be time-locked.
why the panel appreciates while everything else decays
now the objection. signals decay. official statistics catch up. any specific data point is worth a lot in july and worth nothing in eighteen months. so how does the panel appreciate?
the value was never the archive. an archive is a museum. the value is the freshest feed read against the longest calibration history.
one quarter of SMB cash-cycle data is noise. you can’t distinguish seasonality from structural shift. you don’t know what a normal june looks like so you can’t spot an abnormal one. five years of data plus this morning’s feed is a different asset entirely. panels are valued by length and freshness together. this is how alternative data has always been priced.
ADP has run the panel for decades and calibrated its signals against BLS revisions thousands of times. ADP’s asset is not this month’s employment data. it is the calibration history that makes this month’s data readable.
Bloomberg knows this. $30,000 per user per year. what a subscriber pays for is not the current price of every asset. every price is public. what they pay for is decades of aligned, calibrated, timestamped history that lets them price relative to something coherent.
Constellation proved it happens in verticals too. in may 2026 they acquired Keypoint Intelligence, the benchmark authority of the digital-imaging industry. every fact Keypoint produced in 2024 was stale by 2026. that was never the point. the calibration position had accumulated. that is what Constellation paid for.
nobody has ADP for SMB tax filings. nobody has Bloomberg for property-management operations. nobody has Keypoint for MSP incident data. the panels that would make those markets legible do not exist. the multipliers that would generate them are being consolidated for the first time now with ai.
time-locked
a competitor with unlimited capital cannot buy the years back. they can outspend today. they cannot replay 2026 through 2031 to build the calibration history the position requires.
this is what “time-locked” actually means. not “hard to build.” mechanically impossible to compress. the tool depreciates because AI writes it. the operation replicates because a competitor imitates it. the panel appreciates because history only runs forward.
the position is available in a way it will not be in five years. the panel that exists in 2032 will be the one someone started building in 2026. there is no shortcut.
and be honest about what each layer sees, because the exclusivities decay at different speeds. workflow exhaust is already contested. the labs are buying it. assume zero exclusivity within a few years. meaning is different. the labs’ agents see how a company works. they do not see what it earns, owes, and pays. and they cannot backfill the five winters of baselines that make a fresh number legible.
above meaning sits intent. the advisor hears “i’m thinking about succession.” no telemetry hears that. today. but run the trendline forward honestly: owners are beginning to tell their AI things they tell no human. intent’s half-life is shorter than meaning’s. it is a wasting advantage: real now, decisive now, and to be banked now.
meaning appreciates. intent burns. that is this thesis in four words.
the full thesis, with the complete evidence and every rival sensor analyzed, is at predictioneconomy.ai