Voice AI's Funding Lull Is Over — Track These Eight Names Into GTC Week

It is Friday afternoon in the city, the kind of late-winter afternoon where the light goes flat at 4:30 and the only honest thing left on the desk is the spreadsheet. Mine has eight rows. Eight names. One cap table for the entire voice-AI category as I see it heading into GTC week, which kicks off in San Jose on Tuesday. The coffee has gone cold twice. TechCrunch is open in the next tab, queued to whatever Jensen is going to say about agentic voice on the keynote stage. I am not interested in the keynote. I am interested in which of these eight rows survives the year with its TAM math intact.

Let me start with the only number that has actually moved.

The lull is over, and the math is now uncomfortable

CB Insights pulled the numbers and The Wall Street Journal and PYMNTS reported them out: venture investment into voice-AI startups went from $315 million in 2022 to $2.1 billion in 2024. That is roughly 8x in two years. ElevenLabs alone took down a $180 million Series C inside that 2024 figure. On the demand side, PYMNTS Intelligence is now showing 30.4% of Gen Z consumers shop by voice on a weekly basis. Read that twice. Almost one in three of the youngest cohort of buyers is transacting with their voice on a weekly cadence, not a quarterly one.

For two years the polite consensus on voice was that it was a feature, not a category. The 2022 funding totals supported that view. The 2024 totals do not. Eight times growth is not a feature curve. That is a category being repriced in real time, and the repricing is happening in front of an Nvidia keynote that will almost certainly mention agentic voice and almost certainly drag every multiple in the sector with it for a week.

So the operator question for the back half of March is not whether voice AI is real. The operator question is which vendor’s TAM math actually holds when the keynote glow wears off.

My base case heading into GTC: the vertical specialists hold. The horizontal pure-plays compress. Here is how I am ranking the eight names on my sheet.

The only public pure-play, and why the comp set is thin

SoundHound (NASDAQ: SOUN) is, for better or worse, the only public pure-play voice-AI name a generalist fund can actually own. That scarcity premium is doing a lot of work in the multiple right now. Strip it out, and what is left is a business with real automotive and restaurant traction, a credible independent agentic-voice stack, and a revenue ramp that is finally clearing the bar institutional buyers want to see.

The problem with being the only ticker in a hot category is that you become the proxy. Every dollar that wants exposure to voice AI but cannot buy ElevenLabs at a private mark ends up in SoundHound. That is great on the way up. It is less great on the way down, because the volatility is not really about the business. My base case is that SoundHound trades as a sentiment instrument through GTC and reverts to a fundamentals instrument by the May print. Anyone sizing the position should size for that, not for the headlines.

The second public-adjacent name on my sheet is Presto Automation (NASDAQ: PRST). I am keeping it in the eight because the drive-thru voice deployment footprint is real and the customer logos are not vaporware. The cap structure is the issue, not the product. Presto is the name where I want to see another two quarters of unit economics before I underwrite anything beyond a trading position.

That is two of eight. The other six are private, and that is where the interesting work is.

The vertical specialists I am actually paying for

ConverseNow is the one I have been tracking longest. The Deliverect partnership in late February locked in restaurant-management integration in a way that materially changes the channel story. Voice ordering by itself is a feature. Voice ordering wired directly into the same middleware that already runs the kitchen and the off-premise stack is a category. Per the fastcasual.com write-up, that integration is now live across the Deliverect customer base, which is the largest single distribution event voice ordering has seen this cycle.

My base case on ConverseNow: this is the name that gets bid in the next strategic round, and the bidder is somewhere in the POS-plus-middleware adjacency, not a horizontal voice platform. The TAM math holds because the unit of sale is per-location per-month and the locations exist. There is no speculative TAM here. There is a restaurant count.

Kea is the closer comp on the restaurant voice side. Slightly different go-to-market — Kea has historically leaned harder on independent and small-chain operators, where the per-location ARPU is lower but the sales cycle is shorter. The two names will not both win the same accounts. They will split the market by chain size, and the question for an investor is whether the small-chain economics ever clear without a channel partner. I am neutral on Kea standing alone and constructive on Kea as an acquisition target for a POS rollup.

Hi Auto sits in the same restaurant lane but with a stronger speech-stack story and a more defensible technical moat. The deployments are concentrated in QSR drive-thru, which is the highest-volume voice surface in the entire category. If you believe — as I do — that drive-thru is the first surface where voice AI clears positive ROI without a subsidy from the vendor, Hi Auto is the cleanest expression of that thesis on my sheet. I am long this name in the private book.

That is three vertical specialists, all restaurant-adjacent, all with defensible per-location economics. This is the cluster I am paying for through GTC and into the May reporting cycle. For the longer arc of how restaurant voice fits into the broader operator stack, see the /blog/posts/the-ai-premium-in-hospitality-m-a-broker-story-or-real-number piece I filed earlier this quarter.

The early-stage names where I am paying for optionality, not revenue

Vox AI is the early-stage name on my sheet, and I am holding it for optionality rather than revenue. The team has a credible technical thesis on multilingual voice for cross-border QSR — a surface that nobody else on this list is genuinely covering — and the round structure is friendly enough that the downside is bounded. My base case is that Vox AI either gets bought inside 18 months for the team and the multilingual stack, or it raises a real Series A on the back of a single anchor customer. Either outcome is fine. I do not need it to be a standalone outcome to underwrite the position.

Loman is the other name I am holding for optionality, though for different reasons. The product is genuinely good on the small-business side — voice agents for independent restaurants and service businesses — but the unit economics at that ARPU tier require a channel that does not exist yet. The thesis here is that one of the small-business SaaS aggregators reads the same PYMNTS deck I just read, notices the 30.4% Gen Z voice-shopping number, and decides it needs a voice answer. Loman is the cleanest acquisition. I am sizing accordingly.

Incept rounds out the eight. The pitch is a horizontal agentic voice platform, and I want to be honest about my skepticism: this is the seat on my sheet that I think compresses hardest if the GTC keynote brings agentic voice into the Nvidia tent. Horizontal platforms are the names that get re-rated downward when the hyperscaler signals it intends to commoditize the layer. I am keeping Incept on the sheet because the team is excellent and the product is real, but my base case is that this is a sale, not a standalone.

What I am actually watching at GTC

Three things, in order of how much they move my sheet:

First, any signal from Nvidia on agentic voice tooling at the infrastructure layer. If the keynote includes a developer SDK for voice agents that runs natively on the Nvidia stack, the horizontal pure-plays compress and the vertical specialists hold. That is the trade. If Nvidia stays at the model layer and does not push into agent orchestration, the entire sector trades up for a week and then gives most of it back.

Second, any partnership announcement from a restaurant POS or middleware name. ConverseNow already has Deliverect. Watch for a Toast, Olo, or PAR Technology mention adjacent to a voice vendor on this list. That is the real signal for the vertical specialist thesis.

Third, the ElevenLabs follow-on. The $180 million Series C clears for a few quarters, but the question is which strategic shows up on the next round. If it is a hyperscaler at a $5B-plus mark, every horizontal pure-play on this list gets a synthetic comp it cannot earn. If it is a quieter inside round, the entire private cap table for voice AI breathes out.

The trade I would actually put on

If I had to compress the eight rows into a single position, here is the framing. Long the vertical specialists, principally ConverseNow and Hi Auto in the private book. Tactically long SoundHound into the keynote with a tight stop on the other side. Cautious on the horizontal pure-plays and on any name whose pitch deck leads with platform rather than deployment. Watch the /blog/posts/what-the-doordash-sevenrooms-deal-actually-buys note on how restaurant tech multiples have decoupled from broader SaaS comps — that decoupling is precisely what is letting the vertical specialists clear the bar that the horizontals cannot.

The eight names on my sheet will not all be on my sheet in twelve months. My base case is that two get acquired, one goes public, one fails to raise, and the remaining four are still on the sheet but in different slots. The category is real. The cap table is in motion. The TAM math is the only thing that survives the keynote.

That is the sheet. The coffee is cold for the third time. The light is gone. I will be in San Jose on Tuesday.

— Oliver writes The Bottom Line on M&A and valuations. Tips: [email protected].