Soul Machines: How a $70M Digital-Human Darling Ended in Receivership

Published · Researched 2026-09-21

Soul Machines had the best-looking digital humans on earth, $135 million in venture funding, Academy Award pedigree, SoftBank on the cap table, and clients like Mercedes-Benz and Air New Zealand. On February 5, 2026, it entered receivership, with KPMG appointed to sell what was left. If you build anything on top of someone else's platform, this is the story you should study before you sign.

The rise: when photorealism was the whole pitch

Soul Machines didn't start as an AI company in the modern sense. It spun out of the University of Auckland in 2016, built around the work of Mark Sagar — a genuine Academy Award–winning animator — and CEO Greg Cross. The idea was a "Digital Brain": biologically inspired, autonomously animated digital humans that could hold emotionally responsive conversations for brands, banks, and healthcare. Where chatbot companies sold text, Soul Machines sold a face.

And for a while, the money story was everything a startup wants. A $7.5 million Series A in 2016 from Horizons Ventures. A $40 million Series B in January 2020, led by Temasek with Salesforce Ventures participating. Then, in February 2022 — right at the peak of the market — a $70 million Series B1 led by SoftBank Vision Fund 2, bringing total funding to $135 million across six rounds from eleven investors, according to NZ business press (b2bnews.co.nz, April 2026 deep-dive on the receivership).

That was the promise era. Enterprise customers got photorealistic "Digital People" for customer experience, brand ambassadors, training, and kiosks. The tech stack was real: a full pipeline for avatar design, voice configuration, gesture and behavior tuning, conversational AI integration, deployment to web, mobile, and in-store kiosks, plus a reporting dashboard. Review sites at the time weren't wrong about the product quality — squeezegrowth wrote in 2025 that "if you want next-level realism and customizability, Soul Machines creates incredibly life-like digital humans," and ilearnlot called out "unparalleled empathy and engagement." The realism benchmark really was Soul Machines.

The last big public moment came at SXSW 2024: "Digital Marilyn," an AI Marilyn Monroe built with the estate's blessing. It got the press. It was, in hindsight, the company's last headline before the decline.

There's a detail worth sitting with here. Soul Machines was still unveiling flagship digital humans at international conferences while its UK filing showed four months of runway. That's not a criticism of the engineers — the product work was genuinely good — but it tells you how the incentives ran: public storytelling stayed loud long after the balance sheet went quiet. In enterprise AI, the demo is marketing's weapon and the filing is management's diary. When they disagree, the filing wins.

The fall: eighteen months of silent alarms

Here's the uncomfortable part. The collapse wasn't a sudden market shock. The alarms were ringing for years, and the filings tell the story plainly.

In March 2023, UK financial filings showed just $12.3 million in cash against $38.1 million in net outflows for the year — roughly four months of runway. In September 2024, the UK entity was quietly deregistered, a full eighteen months before the formal receivership announcement. Those two facts, sitting in public records, are worth more than any analyst postmortem: the company was running on fumes while it was still unveiling digital humans at conferences.

The human cost followed the balance sheet. Headcount collapsed from 253 in July 2023 to 70 in July 2024 to roughly 45 immediately before receivership — an 82% reduction from peak. CEO Greg Cross resigned in September 2023. Co-founder Mark Sagar stepped down as director in June 2024. The founders were gone before the end.

And then the clients walked. Mercedes-Benz, ANZ, Air New Zealand — the marquee names that made the commercial case credible — all abandoned the technology. As b2bnews.co.nz put it: "When they walked, the commercial case walked with them."

One more detail that should make every enterprise buyer sit up: no revenue figures were ever publicly disclosed. For a company operating for seven-plus years with $135 million raised, the silence speaks. CB Insights lists no revenue data in available filings. When a venture-backed company won't tell you what it sells, it usually means it doesn't like the answer.

On February 5, 2026, receivership became official, with KPMG's Leon Bowker and Luke Norman appointed to run a sale process. The company was, in the press's words, "on the block."

What went wrong — the product and market autopsy

Strip away the drama and the failure has a clear shape: Soul Machines built for the old enterprise market, and the market moved.

It was services-heavy in an API era. Publicly reported enterprise deals clustered in the low-to-mid six figures annually (per competitor Anam's widely repeated 2026 comparison), and delivery meant multi-month bespoke implementations — kickoff to launch measured in quarters, versus days-to-weeks for API-first rivals. One review put it plainly: "the cost and complexity of implementation can be a barrier for smaller businesses" (mhasnain.net). Another competitor-authored but widely echoed framing describes Soul Machines as "an enterprise digital human platform with bespoke creative implementation — Fortune 500 programmes with multi-month delivery and six-figure contracts" (Anam blog — biased source, but the shape of the claim matches the broader record).

The pricing ladder had a cliff. Even the self-serve "Studio" tiers, listed in 2026 tool directories, tell a story: $140/year for 480 conversation minutes on Basic, jumping to $1,069/year on Plus, then $29,160/year on Pro. And support, reviewers noted, vanished at the lower tiers — "the absence of technical support on lower tiers [is an] important factor for organizations planning production-scale deployments" (thetoolsverse, August 2026). Enterprise-grade prices with consumer-grade support at the bottom of the ladder is a combination that teaches buyers to look elsewhere.

The moat was photorealism, and the moat drained. The independent 178-participant avatarbenchmark.com blind study rated Anam's avatars highest among real-time providers. Soul Machines wasn't even evaluated in that sample, but the point stands: realism alone stopped being a defensible advantage. API-first platforms — Inworld for game-grade NPC conversation, Convai for real-time 3D characters, DeepBrain and Anam for real-time streaming avatars — offered transparent pricing, self-serve onboarding, and modern developer tooling. The developer conversation moved on. By 2026, people weren't building on Soul Machines; they were talking about Soul Machines as a cautionary tale and a pricing benchmark.

The timing of the last pivot never came. Six rounds of funding, SoftBank at the table, and still no publicly stated revenue by year seven. The venture model needs the revenue line to bend upward eventually; photorealism as a service, sold through long sales cycles to a handful of flagship brands, apparently never bent it enough.

What enterprise buyers should learn from this

If you're buying or building on digital-human technology, the Soul Machines arc is a checklist of things to verify before you commit:

1. Read the filings, not the funding announcements. The $70M SoftBank round got the headlines in February 2022. The March 2023 UK filing — $12.3M cash, $38.1M outflow — told you the company had roughly four months of runway. Funding announcements are marketing; statutory filings are the company's heartbeat. For any vendor you're depending on, check whether revenue is ever disclosed at all. Seven years of silence is an answer.

2. Treat delivery timelines as a competitive signal. Multi-month bespoke implementations versus days-to-weeks self-serve isn't just a convenience gap — it's a structural one. Services-heavy vendors carry cost structures their API-first competitors don't, and those costs get passed to you as six-figure contracts and long onboarding. If a vendor can't quote you transparent per-minute or per-seat pricing, that's data.

3. Ask who keeps the lights on. Headcount falling 82%, founders exiting, marquee clients leaving — each of those is survivable alone. Together, eighteen months before receivership, they're a forecast. If your vendor won't tell you their headcount trend or their cash position, discount their roadmap to zero.

4. Own your stack where it matters. The deepest lesson: Soul Machines' customers bought a fully integrated pipeline — avatar, voice, behavior, conversation, deployment, analytics. When the company collapsed, there was no clean way to keep any of it running, because every layer was proprietary. If you're building on a platform, know exactly what happens to your product if the vendor disappears. Open components (open-source SDKs, standard file formats like VRM, your own LLM layer) are an insurance policy. The cheapest vendor in year one can be the most expensive one in year three if it takes your whole deployment down with it.

5. Watch for the deregistration. A UK entity quietly deregistered eighteen months before receivership is the kind of detail that never makes the press release. Check the registries. It takes twenty minutes and it tells you what the blog posts won't.

Does anything survive?

Honestly: UNVERIFIED, as of September 20, 2026.

KPMG was running a sale process as of the February 2026 receivership. No acquisition or sale outcome has appeared in any observed source. PitchBook, crawled around August 2026, still lists the company as private and "Generating Revenue" with roughly 58 employees — consistent with continued operation under receivership or a quiet sale as a going concern, but that is inference, not fact.

Meanwhile, third-party tool directories (thetoolsverse, updated August/September 2026) still list Soul Machines Studio as an available self-serve product with free access, and competitor comparison pages from Tavus and Anam still treat it as a live competitor. Whether the self-serve Studio's signup and billing actually function after receivership is UNVERIFIED — the research flags it explicitly. Do not buy without confirming operational status directly with whoever is running the company now. If you have an existing Soul Machines deployment in production, the only honest move is to ask KPMG or the current operator, in writing, who provides support and for how long — community discussion and review directories cannot answer that for you.

The technology itself was real — the realism work, the Digital Brain architecture, the pipeline — and technology outlives companies all the time through asset sales. But as of this writing, anyone claiming to know the buyer's name or the final outcome is guessing.

The bottom line

Soul Machines wasn't a scam and wasn't a failure of ambition. It built genuinely impressive digital humans and sold them to real brands — right up until the market decided that photorealism delivered through six-figure, multi-month services engagements wasn't worth what it cost, and API-first platforms ate the space from below. $135 million raised, zero publicly disclosed revenue, four months of runway hidden in a 2023 filing, and a quiet deregistration eighteen months before the end.

The lesson isn't that digital humans are a bad bet — Inworld, Convai, DeepBrain, and Anam are all growing in the space Soul Machines defined. The lesson is that when you build on a platform, you're betting on the company's balance sheet, not its demos. Read the filings. Ask about the revenue line. And never assume the best-looking product in the room is the safest one to bet your roadmap on.