The Man Building a Universal Translator for the Human Mind

There’s a moment in every good sci-fi story where two beings who share no common tongue suddenly understand each other perfectly. Star Trek called it the universal translator. Phoenix Peng calls it Tuesday.

“Language is a compression layer,” Peng said during our fireside chat at BEYOND EXPO Macau 2026. “Sometimes we need to find the right word to express ourselves, but when you have the word, your intention has already been compressed.” He argues that every sentence you speak is a lossy version of a much richer thought, and he wants to read the original file instead of the compressed one.

That’s the pitch behind Gestala, the Chengdu-based startup Peng founded this year to build brain-computer interfaces that don’t require drilling a hole in your skull. Instead of electrodes, Gestala uses focused ultrasound, sound waves tuned to reach deep into neural tissue, and eventually both read and stimulate brain activity from outside the head.

Overcoming the skull barrier

Peng isn’t working in an empty field. Neuralink has already implanted its electrode arrays in nine human volunteers, some of whom can now control a cursor or robotic arm with their thoughts. Synchron has taken a different, less invasive route, threading electrodes through blood vessels into ten patients so they can send emails and control smart home devices. And in January 2026, OpenAI backed Sam Altman’s own BCI venture, Merge Labs, with the largest single check in its USD250 million seed round, a company explicitly betting, like Gestala, that ultrasound can do what electrodes can’t: reach neurons at scale without opening the skull.

Peng has watched this race from an unusual vantage point. Before founding Gestala, he ran NeuroXess, an implantable BCI company, meaning he’s now competing against his own former approach. “I don’t think we will always be non-invasive,” he admits of the ultrasound method. The skull scatters and weakens ultrasound signals, and getting a clean read through solid bone is, in his words, "still very challenging in engineering.”

That kind of candor is part of what makes Gestala’s pitch land. The company’s balance sheet backs it up too. Its USD21.6 million angel round closed just two months after founding, co-led by Guosheng Capital and Dalton Venture, with gaming billionaire Tianqiao Chen among the backers. Gestala arrived weeks after Merge Labs went public with its own ultrasound ambitions, which looks less like coincidence and more like two sides of the Pacific reaching for the same idea at the same time.

The pace has only picked up since. On July 3, 2026, Gestala closed a RMB 420 million (USD62 million) Series Angel+ round, led by Meridian Capital Asia with a long list of Chinese institutional investors joining in, pushing its total raised to roughly USD84 million in under six months. The same week, the company inaugurated a second headquarters in Shanghai, adding to its Chengdu base and newly operational manufacturing plant, and beefed up its bench with a former Neuralink clinical lead and two regulatory veterans from visual-BCI company Nano Retina. For a company not yet a year old, that's an unusually fast build-out, and it signals that investors are betting on Gestala's execution speed as much as its underlying science.

Tokens made of blood flow and sound

Ask an AI researcher what a large language model learns from, and they’ll say tokens, chunks of text. Ask Peng what his models learn from and the answer gets stranger. “Our token is not language,” he said. “Our token is a biometric index.” Gestala’s systems ingest electrical signals, ultrasound reflections, and blood flow patterns at the same time, then run them through transformer-based architectures to find the pattern connecting neural noise to observable behavior: what someone is looking at, saying, or reaching for.

Peng is candid about the limits of generalization too. Because every human brain is wired slightly differently, a model trained on one patient’s signals doesn’t transfer cleanly to the next. “The foundation model structure is similar,” he said, “but the input-output is different.” Each user effectively needs a translator retrained just for them.

Chronic pain as the first wedge

Gestala’s first target is chronic pain, which affects roughly a fifth of adults. The underlying science is real and peer-reviewed: a 2024 study in the Journal of Neuroscience showed that low-intensity focused ultrasound aimed at the dorsal anterior cingulate cortex could noninvasively reduce acute pain perception in humans, and a separate exploratory clinical trial targeting the same region in chronic neuropathic pain patients found meaningful reductions in reported pain scores over a multi-week follow-up. That’s the scientific basis Peng is building on. He says Gestala’s own device can target that region and cut pain intensity by roughly half, with effects lasting one to two weeks. That could eventually reduce patients’ reliance on opioids and other pharmaceuticals. From there, the roadmap extends into stroke recovery, depression, PTSD, OCD, and eventually Alzheimer’s, a six-to-eight indication pipeline that reads more like a pharmaceutical company’s roadmap than a typical hardware startup’s.

Worth flagging clearly: Gestala’s own pain results, the roughly 30-participant trial the company points to, have not been published in a peer-reviewed journal; they’re company-reported preliminary results, not independently verified data. And as of this spring, Gestala had not yet filed for regulatory approval with China’s National Medical Products Administration, targeting a submission by the end of 2026, meaning its device isn’t cleared for clinical use anywhere yet.

None of it moves fast, and Peng doesn’t pretend otherwise. BCI devices have to clear the same regulatory path as any medical device: animal testing, human trials, and approvals. He puts the full cycle at “seven to eighteen years.” That’s a sobering number in an industry prone to hype, and it lines up with where Gestala actually is in that process: pre-filing, with unpublished early data.

A Rosetta Stone for carbon and silicon

Peel back the medical roadmap, and Peng is chasing something bigger: a shared translation layer between minds and machines. “There will be two different foundation models,” he said, “silicon-based and carbon-based. These two models can probably explain each other.” He imagines machine language and neural language becoming mutually legible, each side able to interpret the other.

It’s a future measured in decades. Peng estimates 15 to 30 years before brain-to-brain communication bandwidth meaningfully exceeds spoken language. But if even part of it arrives, the implications for how CDOs think about human-computer interaction are significant: interfaces that don’t wait for typing or talking, diagnostics that catch neurological disease before symptoms appear, knowledge that downloads in an instant (imagine the implications on education) and enterprise tools that finally speak the brain’s native format instead of ours.

Star Trek’s Federation took around three centuries to build a universal translator. Peng, Altman, and Musk are trying to do it in one generation, through sound waves, silicon, and a lot of regulatory paperwork. However it plays out, the era of typing what we mean instead of simply thinking it may be closer to its final chapters than most people assume.

Image credit: iStockphoto/SvetaZi