Dental AI companies raised $800M in 2025. Where did it go?

On January 28, 2025, VideaHealth closed an oversubscribed $40 million Series B led by Threshold Ventures, pushing its total raised to $70 million, and spent the money scaling an X-ray-analysis platform into chains like Heartland Dental and Dentalcorp, according to SiliconANGLE. That round was the template for the year. The widely repeated figure for 2025 dental AI funding, somewhere around $800 million, is a press-release aggregate that nobody audits, but the composition of it is not in dispute: the capital went into the radiograph. Eight hundred million dollars taught software to read a bitewing, and not one dollar of it will answer your phone at 4:50 on a Friday.

Here is the claim we will defend, and plenty of owners will hate it: for a solo or two-doctor practice in 2026, caries-detection AI is the lowest-return AI purchase on the market, and the money chased it anyway because the buyer the venture capital was underwriting was never you. It was a 1,000-location group with a diagnostic calibration problem you don't have.

Half the Clearances Point at One Body Part

Follow the regulatory trail and the concentration is obvious. As of December 2024 there were 52 FDA-cleared dental AI and machine-learning devices, and Oral Radiology alone accounted for 48% of them, 25 clearances, per a peer-reviewed review in PMC. Nearly half of everything the FDA has ever cleared in dental AI looks at a picture of a tooth.

That's not an accident of science. It's an accident of what is fundable. An image-analysis model has a clean training target, a labeled dataset, a measurable sensitivity and specificity, and a regulatory path that already exists because radiology software has been clearing for two decades. A model that reduces your 38% hygiene reappointment gap has none of that. It has no clearance pathway, no benchmark, no clinical journal, and no way to prove to a Series B investor that the lift came from the software and not from the office manager you hired in March.

Capital goes where the proof is cheap. It is a reasonable thing for capital to do. It is a terrible reason for you to buy.

The Standards Body Confirmed the Pile-Up

If you want a second confirmation that the money bunched in one place, look at what the ADA had to write rules about. ANSI/ADA Standard No. 1110-1:2025, the first U.S. standard on AI in dentistry, was approved in 2025 and governs validation datasets for AI-based 2D radiograph image-analysis systems, per the ADA. Not AI in dentistry broadly. Two-dimensional radiograph image analysis, specifically.

Standards bodies are lagging indicators. They write rules where the products, the dollars and the liability have already stacked up high enough to make a mess. The first American standard for dental AI being a bitewing standard tells you exactly where the 2025 wave landed, and it tells you something useful about timing: an area gets standardized right around the time it stops being a differentiator and starts being a commodity.

Scheduling Is Tied With Diagnostics, and Nobody Funded It

Now look at the demand side, which the capital largely ignored. ADA Health Policy Institute data shows 43.3% of responding dentists are using AI for at least one task. Imaging and diagnostics sits at 22.8%. Appointment scheduling also sits at 22.8%.

Dead even. Dentists adopted AI for booking appointments at exactly the same rate they adopted it for reading radiographs, and roughly half the FDA-cleared device landscape is pointed at one of those two jobs while the other got the leftovers. That gap between where the money went and where practices are actually putting AI to work is the single most useful fact in this entire story.

It also tells you something about price. Imaging AI is crowded, standardized and venture-subsidized, which means the per-month price is going to fall and the feature sets are going to converge. Scheduling, recall and phone-coverage AI is fragmented and underfunded, which means the good ones are still expensive and the bad ones are plentiful. You are shopping in a buyer's market for the thing you need least and a seller's market for the thing you need most. Budget accordingly.

What Most Owners Get Wrong: Buying the DSO's Tool on the DSO's Logic

The specific mistake, and we see it most in owners of two to four operatory practices and small groups of three locations or fewer: they look at which AI the big DSOs bought, assume it must be the proven one, and buy the same thing. Heartland and Dentalcorp are sophisticated buyers. They are also solving a problem you do not have.

A group with hundreds of dentists has enormous variance in diagnostic thresholds. One associate watches an incipient lesion, another restores it, a third misses it, and the group has no way to see which. A radiograph AI that applies one consistent threshold across every operator is a governance tool. It standardizes a workforce. It also creates a defensible paper trail when a patient asks why the last office didn't mention this.

You, with one or two doctors who have been calling the same lesions the same way for eleven years, have no variance problem. You have a capacity problem, a collections problem, or a new-patient problem. Buying a standardization tool to fix a capacity problem is like buying a torque wrench because your car is out of gas. Our view: if your treatment plans aren't the constraint, diagnostic AI is a cost, not a system.

Running the Numbers on a $12,000 Caries AI Subscription

Take a practice collecting $1.1 million, four clinical days a week, 18 hygiene patients a day. That's 3,312 hygiene exams a year.

Say half of those visits include bitewings, so 1,656 radiograph series a year run past the model.

Be generous to the vendor and assume the AI surfaces a restorable lesion you would otherwise have watched on 6% of those series: 99 additional findings a year.

At a 55% acceptance rate on a single-surface composite, that's 54 accepted restorations.

At $230 collected per composite after PPO write-offs, that's $12,420 in new collections.

Against a $1,000 a month subscription, you net $420 for the year.

And that's before the chair time. Fifty-four restorations at 40 minutes each is 2,160 minutes, or 36 doctor hours.

If your doctor schedule is already 85% full, those 36 hours displace something. Displace even a third of them with crown, endo or implant time producing $600 an hour in contribution and you've burned $7,200 of capacity to earn $420.

Now run the same $12,000 at the other end of the practice. Say you're carrying $180,000 in diagnosed, unscheduled treatment, which is ordinary for a practice that size. Recover 12% of it with automated recall, reactivation texting and after-hours call coverage and that's $21,600.

Net $9,600 on the identical spend, with zero new diagnosis required and treatment the patient already agreed to.

Then look at why the DSO math is completely different. A group collecting $400 million that moves diagnosis just half a percent picks up $2 million a year, which pays for a lot of per-location licenses.

Same product. Same price per chair. One buyer is printing money and the other is renting a second opinion on an image they were already reading correctly. We wouldn't sign the caries AI contract this year at a single-doctor practice unless the vendor is pricing it under $300 a month or you have genuinely open restorative time you can't fill.

The Funding Wave Still Helps You, Just Later

None of this means the $800 million was wasted. It means the benefit arrives to you as a price cut rather than an advantage. Twenty-five FDA clearances in one category is what commoditization looks like in its early stage. A published validation standard is what it looks like in its middle stage. The end stage is the feature getting folded into your imaging software or your practice management system at no separate line item, which is where this is heading within about 24 months.

The owners who wait will get a better product for a fraction of the money, and they'll get it with a standard behind the validation dataset instead of a vendor's internal claim. The owners who sign three-year contracts in 2026 at 2025 prices will be paying for a feature their competitors get bundled. That's not a technology judgment. It's a procurement judgment, and procurement is the part of this most owners skip.

What Would Change Our Mind

Three things. First, hard payer behavior: if a major PPO starts treating AI-flagged pathology as supporting documentation that reduces denials, the ROI stops depending on new production and starts depending on avoided write-offs, which changes the math completely. Second, pricing: if per-location imaging AI drops under $300 a month, the capacity argument weakens and it becomes a cheap insurance policy against the lesion you missed. Third, the adoption split: if ADA HPI's next read shows imaging pulling clearly away from scheduling rather than sitting even with it, that's evidence practices are finding returns we're not seeing.

We're also watching whether any 2026 round goes to the unglamorous side of the operation at real scale. The first dental AI company to raise $40 million for answering the phone and filling tomorrow's holes will tell you the market finally priced the actual problem.

Do This Before You Sign Anything

Pull your diagnosed-but-unscheduled treatment total out of your practice management software this week, and pull your doctor chair use for the last 90 days. If use is above 80%, you do not have a diagnosis problem and no imaging AI will help you. Then take every AI quote on your desk and divide the annual cost by the collections it would have to generate at your real net-of-write-off rate, not your fee schedule. If a vendor won't quote month to month, treat that as information about their confidence, not yours.

Before you spend $12,000 a year on any software, you should know what your overhead actually tolerates. Our free overhead calculator breaks your spend down by category so you can see whether a new technology line item fits or just pushes your doctor compensation down another point: run your own numbers.