Have you ever taken SSRIs? I have. They hardly work. So why does everybody take them? (And what are SSRIs, by the way?)

SSRIs are a common class of psychiatric drugs. The acronym stands for selective serotonin reuptake inhibitor. In short, SSRIs block the serotonin transporter, so less serotonin is taken back up into the presynaptic neuron, which increases serotonin activity in other parts of the brain. Serotonin is a neurotransmitter that regulates a slew of functions in the central nervous system. When one neuron sends a serotonin signal to another, most of the serotonin is quickly vacuumed back up by a protein called the serotonin transporter. SSRIs block this protein, so serotonin hangs around for longer in the gap between neurons (the so-called synaptic cleft).

This does two things:

1. It slightly boosts the strength of serotonin signals.

2. Over several weeks, it changes how sensitive many brain circuits are to serotonin.

The thing is, A LOT of brain circuits are sensitive to serotonin. We’re talking mood, anxiety, obsessive thoughts, impulse control, sleep, appetite, sexual function, pain, and so much more. Whoever controls serotonin has a pretty good handle on the steering wheel.

How many people take SSRIs? There is no single global headcount, but the available data suggest tens of millions of people all over the world. It’s everywhere. And for many patients, it isn’t working.

SSRIs show small to modest average benefits over placebo, limited remission, and high discontinuation rates. A major 2018 network meta analysis in The Lancet looked at 522 double blind randomized trials of 21 antidepressants, including SSRIs, in 116,477 participants. It found two things:

1. All antidepressants were statistically more effective than placebo

2. The average effect size compared with placebo was around 0.3 standard deviations, which is code for “small”, just a few points on common depression scales.

The effect is so little that, according to a peer-reviewed paper published on BMJ Open (Journal Impact Factor of 2.4), “the evidence does not support definitive conclusions regarding the benefits of antidepressants for depression in adults. It is unclear whether antidepressants are more efficacious than placebo.”

Kirsch 2008, reanalysing FDA data, found “virtually no difference” at moderate depression, and only reached the NICE “clinical significance” threshold (d ≥ 0.5) at the very severe end of the scale.

Fournier 2010 reached a similar conclusion: benefit over placebo increases with baseline severity, and may be minimal or non-existent on average in mild or moderate depression.

Placebo response in depression trials is huge:

• Meta analyses of antidepressant RCTs find placebo response rates around 35–40 percent.

• Kirsch’s analysis reports that “82% of the drug response was duplicated by placebo.”

Which means that, from a population point of view, most of the improvement people experience after starting an SSRI could have occurred with placebo, time, and supportive care, and only a minority experience a substantial drug specific effect. A 2022 BMJ analysis even suggested that only “about 15% of participants have a substantial antidepressant effect beyond a placebo effect.”

So, the efficacy of SSRIs is still debated, but what about remission rates and drop out data?

The biggest “real world” SSRI trial is STAR*D, where all patients started on citalopram (an SSRI). After the first trial, remission rates were 28–33 percent, response 47 percent. After up to four sequential treatment steps, the original report even claimed a cumulative remission of 67 percent, but with high drop out and method issues. But a 2023 reanalysis that handled missing data more rigorously estimated a true cumulative remission of only 35 percent.

Naturalistic cohorts tell a similar story. One study found about 73 percent of patients discontinued antidepressants within 24 weeks, regardless of which drug they were on.
According to the University of Oxford’s PETRUSHKA project, around 80 percent of people in the UK who are prescribed antidepressants “stop the treatment after just a few weeks.”

TL;DR: For most people, SSRIs mostly do not work, and most people quit early. Why? Because of limited efficacy. And, most importantly, because of annoying side effects.

Nausea, diarrhea, headache, insomnia, dry mouth, tremor, and increased anxiety are all common early on. And sexual dysfunction—that’s a big one. One study found that 73 percent of SSRI users reported sexual side effects, compared with 14 percent on bupropion. Think loss of libido, erectile dysfunction, anorgasmia.

And weight gain, did I tell you about weight gain? Just imagine that in one trial “a significantly higher proportion of paroxetine treated patients (25.5%) gained ≥ 7% in weight compared with the other treatment groups (sertraline, 4.2%; fluoxetine, 6.8%).”

And finally, the bigger of all side effects: sloth. You feel dull: less joy, less sadness—flat like a pancake.

That’s why people stop. And when they do, the troll starts taking its toll. Recent meta analyses of randomized trials suggest about 1 in 6 patients stopping antidepressants have noticeable discontinuation symptoms, and about 2 to 3 percent face severe protracted problems. Typical symptoms include dizziness, flu-like symptoms, insomnia, irritability, anxiety, low mood, strange sensory phenomena. And while for many people this is a few days, for a non-trivial minority, “withdrawal can sometimes be more difficult, with symptoms lasting longer (in some cases several weeks, and occasionally several months)” and “sometimes be severe, particularly if the antidepressant medication is stopped suddenly.”

A bridge to the future

So what do you do when one drug doesn’t work? You add a second one. Atypical antipsychotics like aripiprazole or quetiapine, or old school lithium, are often added on top of an SSRI in so-called augmentation strategies. Meta analyses show that this does increase response and remission rates compared with staying on the antidepressant alone, but the gain is “small to moderate” and comes at a cost: weight gain, metabolic problems, sedation, and so on.

Then came ketamine. At sub anesthetic doses, this old anesthetic blocks NMDA receptors, triggers a rapid glutamate surge and a burst of synaptic plasticity. In layman’s terms, it briefly puts the brain into a “rewiring mode”: all of a sudden, neurons start chatting more and more, forming brand-new connections, making people feel noticeably better within a day. Its S enantiomer, esketamine, was turned into Spravato, an intranasal formulation that turned this effect into a product. Spravato, made by Janssen, the pharmaceutical arm of Johnson & Johnson, was first approved in 2019 for treatment resistant depression (TRD) in combination with an oral antidepressant, then for major depressive disorder (MDD) with acute suicidal ideation or behavior, and more recently as monotherapy in TRD. It has to be given in certified clinics as it can cause sedation, dissociation, spikes in blood pressure and, caveat emptor, “abuse and misuse.” Patients come in, spray the drug, ride out a very strange couple of hours under observation, and a large minority feel noticeably better right from the get-go.

And it’s the real deal. In pivotal trials, adding Spravato to a standard antidepressant led to an additional improvement of about 3.8 to 4.1 points on the Montgomery Asberg Depression Rating Scale (MADRS) at 4 weeks, compared with placebo plus antidepressant, in patients who had already failed at least two SSRIs. Long term open label data show that if you keep giving esketamine weekly, every other week, or every 4 weeks, a big chunk of treatment resistant patients stay in response or remission, with remission rates in this cohort reaching about 46 percent. In a real-world multicenter study of esketamine nasal spray for treatment resistant depression, about two thirds of patients achieved a clinically meaningful response and about 40 percent achieved remission by 3 months, with roughly one third already responding after the first month.

Spravato is now a blockbuster drug tracking a $1.6 billion run rate, which tells you how strong the demand is when something finally moves the needle.

The era of psychedelic medicine

And then came psychedelic medicine. The molecules themselves are not new. LSD was first synthesized in 1938 and by the 1950s psychiatrists were already running hundreds of small studies on LSD for alcoholism and mood disorders before the moral panic of 1960s shut everything down.

The modern reboot started in universities in the 1990s and accelerated through the 2000s and 2010s, with small, tightly controlled trials at places like Johns Hopkins, NYU, and Imperial College. What changed in the last few years is that this moved out of the lab and into Delaware and Bay Street. Around 2019 and 2020 you suddenly had pure play psychedelic medicine companies listing in New York and Toronto, building pipelines around psilocybin, MDMA, DMT, and LSD instead of SSRIs. MindMed, focused on lysergide (LSD) D-tartrate and R(-)-MDMA, was founded in Toronto in 2019 and became the first psychedelic pharmaceutical company to go public on a Canadian exchange, then uplisted to Nasdaq. In London, Compass Pathways, a biotech founded in 2016 that researches psilocybin therapy, became the first psychedelic medicine company to list on Nasdaq in 2020. At the same time, atai Life Sciences, a Berlin based psychedelic platform company, raised hundreds of millions of dollars and went public on Nasdaq in 2021, before combining with Beckley Psytech in 2025 to form AtaiBeckley.

At this point you’re probably wondering, does any of this actually work, or is just New Age mumbo jumbo?

That’s the cool part.

None of it is.

MindMed

MindMed is building around MM120, a lysergide D-tartrate (LSD) orally disintegrating tablet with patent protection out to 2041, and around MM402, an R(-)-MDMA enantiomer in development for autism spectrum disorder (ASD).

The company’s trial of MM120 in generalized anxiety disorder (GAD), published in JAMA in 2025, offered the first solid clinical data on LSD for anxiety. It was a 12-week, multi-center, parallel, randomized, placebo-controlled, double-blind, Phase 2b study designed to assess the dose-response, efficacy, safety, and tolerability of a single dose of MM120 in 198 adults with moderate-to-severe GAD. Participants received a single dose of MM120 (25, 50, 100, or 200 micrograms) or placebo and then were simply followed; there was no protocolized psychotherapy bolted on.

The 100 µg dose of MM120, now being evaluated in three pivotal Phase 3 trials, showed the optimal level of clinical activity in the study. At Week 4, it achieved a 7.6-point greater reduction in HAM-A scores compared to placebo (-21.3 vs. -13.7; p<0.0004; Cohen’s d=0.88) with a 65% clinical response rate and 48% clinical remission rate sustained to Week 12. This all came from a single dose, not from daily pills.

MM120 was generally well-tolerated in this study, with most adverse events rated as mild-to-moderate, transient, occurring on the dosing day, and being consistent with the expected acute effects of the trial drug. Based on these Phase 2b study results and the significant unmet medical need in the treatment of GAD, the U.S. Food & Drug Administration (FDA) has provided Breakthrough Therapy Designation to MM120 for GAD.

Building on positive Phase 2b study results in generalized anxiety disorder (GAD), which showed MM120’s potential antidepressant effects and clinically meaningful improvements on MADRS at the 100 µg dose, the company has gone straight into the Phase 3 Emerge study in major depressive disorder (MDD) using a single 100 µg MM120 ODT dose versus placebo in roughly 140 patients with a 12 week topline readout guided for mid 2026, and is planning a second Phase 3 MDD trial, Ascend, to start in mid 2026 with the same basic 12 week plus 40 week design and 100 µg, 50 µg control or placebo.

In GAD, MM120 is also in two Phase 3 trials, Voyage and Panorama, both built around the same single 100 µg MM120 ODT dose, with enrollment running through 2025 and 12 week topline data guided for 2026, Voyage in the first half of the year and Panorama in the second.

The second MindMed asset, MM402, pushes beyond mood and anxiety into neurodevelopmental conditions. MM402 is the company’s proprietary R(-)-MDMA, the R-enantiomer of MDMA, developed for ASD. Preclinical work in rodent models of ASD suggests that R(-)-MDMA is more potent and enhances social interaction more strongly than racemic MDMA. MindMed has already completed a Phase 1 study in healthy volunteers and plans to initiate a single dose, open label Phase 2a study of MM402 in up to 20 adults with ASD in Q4 2025, targeting core socialization and communication symptoms.

Compass Pathways

Compass Pathways has taken a different route, focusing on one proprietary psilocybin formulation, COMP360, which in 2018 received FDA Breakthrough Therapy designation for psilocybin therapy for TRD.

The anchor data set is the Phase 2b trial in treatment-resistant depression (TRD), published in the New England Journal of Medicine in 2022 and funded by Compass itself. In that study, 233 adults with TRD were randomized to a single 25 mg, 10 mg, or 1 mg session of COMP360, always combined with psychological support. At week 3, the 25 mg group had a 6.6-point greater reduction in MADRS depression scores than the 1 mg control group (a standard “low-dose psilocybin” active placebo), and 36.7 percent met response criteria and 29.1 percent were in remission, compared with 17.7 percent and 7.6 percent in the control arm.

Phase 3 is now confirming that this is not a statistical fluke. In June 2025, Compass reported top-line results from COMP005, the first of its pivotal TRD trials: 258 adults across 32 US sites, a single 25 mg COMP360 session versus placebo, primary endpoint at week 6. The 25 mg arm showed a high statistical significance, clinically meaningful reduction in depression at 6 weeks, and no unexpected safety findings, hitting a 3.6-point advantage on MADRS over placebo and meeting the pre-specified primary endpoint. A second Phase 3 trial (COMP006) with two COMP360 doses and longer follow-up is ongoing, with 9-week Part A data and COMP005 26-week data expected in Q1 2026, and 26-week Part B data from COMP006 expected in early Q3 2026.

Compass is not stopping at unipolar TRD. The same COMP360 platform is being pushed into PTSD, anorexia nervosa, and bipolar II depression. A Phase 2 open-label study in 22 adults with PTSD, run at King’s College London, used a single 25 mg COMP360 dose with structured therapy and followed patients for 12 weeks; published data show consistent improvements in PTSD symptom scores and functioning across that period.

Anorexia nervosa is being tackled through COMP401, Compass Pathways’ Phase 2 proof-of-concept trial of COMP360 psilocybin therapy in adults with anorexia nervosa (25 mg vs 1 mg in about 60 participants across multiple US, UK and Irish sites), built on top of the Phase 1 open-label Nature Medicine feasibility study, which showed that a single 25 mg COMP360 session plus psychological support was feasible, acceptable, and produced clinically meaningful improvements in eating disorder psychopathology at 3 months in 4 out of 10 participants (40 percent), defined as having global Eating Disorder Examination scores within one standard deviation of community normative values.

On top of that, an investigator-initiated open-label study in bipolar II depression, supported by Compass, has already made it into JAMA Psychiatry: a single 25 mg COMP360 session plus psychotherapy in 15 adults with bipolar II depression led to sustained MADRS response and remission in the majority of participants out to 12 weeks.

AtaiBeckley

AtaiBeckley’s strategy is different again: instead of psilocybin or LSD, it focuses on intranasal 5 MeO DMT, DMT, and R-MDMA.

The most visible asset is BPL-003, a proprietary intranasal formulation of mebufotenin benzoate, a synthetic version of 5-MeO-DMT. Sessions are short by psychedelic standards, with the acute experience typically under two hours, and the drug is designed from the start to fit into the same half-day clinic workflow that Spravato pioneered. In 2025, AtaiBeckley reported that their randomized Phase 2b trial in 193 adults with TRD met its primary endpoint and all key secondary endpoints, and BPL-003 demonstrated rapid, robust and durable antidepressant effects with a single dose.

Single 8 mg and 12 mg doses of BPL-003 produced statistically significant and clinically meaningful reductions in depressive symptoms at all study time points versus a 0.3 mg low dose active control, with MADRS scores at day 29 reduced by a mean of 11.1 points for 12 mg and 12.1 points for 8 mg, compared with 5.8 points in the control arm (p = 0.0038 and p = 0.0025, respectively), and improvements detectable as early as one day after dosing and generally maintained through week 8. BPL-003 was generally well tolerated at all doses, with 99 percent of treatment emergent adverse events rated mild or moderate, no drug related serious adverse events or suicide-related safety signals, and the majority of patients considered ready for discharge 90 minutes after dosing.

In light of these results, BPL-003 has already secured FDA Breakthrough Therapy designation for TRD.

The second AtaiBeckley pillar is VLS-01, a proprietary buccal film formulation of DMT for TRD. The idea is to get DMT’s ultra-rapid onset and short duration into a convenient oral transmucosal strip that dissolves on the cheek, again targeting a sub-two-hour clinic stay. Phase 1b data in healthy volunteers have already been reported, and the Phase 2 ELUMINA trial in TRD is underway, randomizing patients to VLS-01 versus placebo buccal film, with topline data to be published in the second half of 2026.

The third leg is EMP-01, a proprietary oral R-MDMA formulation for social anxiety disorder (SAD), developed through the EmpathBio subsidiary, now fully inside AtaiBeckley. In May 2025, Atai announced that the first patient had been dosed in an exploratory phase 2 study of EMP-01 in adults with SAD, with top-line data expected in the first quarter of 2026.

AtaiBeckley also runs an early stage discovery program of novel 5-HT2A agonists, including non-hallucinogenic neuroplastogens, for opioid use disorder and treatment resistant depression. The program applies AI and machine learning guided medicinal chemistry to generate 5-HT2A agonists that are non-hallucinogenic in rodent drug discrimination assays and show antidepressant-like effects in Wistar Kyoto rat models.

So… are psychedelics about to have their GLP-1 moment?

GLP-1 drugs reshaped obesity treatment after decades of stalled innovation. Better yet, they have changed the world and saved who knows how many lives.

Is the same thing about to happen with psychedelics?

Maybe.

If so…

…what happens next?

Subscribe now

Disclaimers:

• a) At the time of publishing this article, I own shares in Mind Medicine (MindMed) Inc. (MNMD).

• b) This article should not be misconstrued as financial or medical advice. Do your own research, talk to experts, and don’t do anything stupid.

• c) I put a lot of effort into this article. If you like it, please help me grow my publication by subscribing: it’s free. I publish weekly, bringing you a glimpse from the extropian future, and the ideas and markets wiring it all together. And if you also decide that the article is worth sharing with someone in your life, that’d mean the world to me.

Subscribe now

Share

Is Memo from Sunday Robotics a spiritual machine? In Kurzweil’s own words, “a spiritual machine is a conscious machine. Consciousness is the true spiritual value.” By this logic, Memo is certainly not a spiritual machine.

Or is it? At what level of intelligence can we start talking about consciousness? Could it be that consciousness is a spectrum? Or, in other words, is it reasonable to assume that Memo, while not being human-level conscious, might be at least as conscious as a worm? (And is a worm conscious, after all?) In the absence of any formal proof, any consciousness-related statement is, at best, unfalsifiable. It follows that “Whereof one cannot speak, thereof one must be silent.” But I’m not in the business of staying silent.

This story starts with Deep Blue, IBM’s computer, which famously beat then world chess champion Kasparov in 1997. While today a computer being superhuman at chess is no one’s surprise, at the time it struck a chord. So much so that Ray Kurzweil, renowned inventor and serial entrepreneur, came up with the idea that, one day, however far away, computers will match humans at any task, no matter the difficulty. Sounds familiar? Needless to say, the concept was not welcomed with open arms. Author and philosophy professor John Searle opposed any notion that the Deep Blue had actual understanding of the game.

Imagine a machine that takes Chinese characters as input and outputs Chinese characters so perfectly to make it indistinguishable from a native Chinese speaker. Does the machine understand Chinese?

Now suppose that a man is in a room.

“One way to test any theory of the mind,” argues Searle, “is to ask oneself what it would be like if my mind actually worked on the principles that the theory says all minds work on. [...]

Suppose that I’m locked in a room and given a large batch of Chinese writing. Suppose furthermore (as is indeed the case) that I know no Chinese, either written or spoken, and that I’m not even confident that I could recognize Chinese writing as Chinese writing distinct from, say, Japanese writing or meaningless squiggles. To me, Chinese writing is just so many meaningless squiggles.

Now suppose further that after this first batch of Chinese writing I am given a second batch of Chinese script together with a set of rules for correlating the second batch with the first batch. The rules are in English, and I understand these rules as well as any other native speaker of English. They enable me to correlate one set of formal symbols with another set of formal symbols, and all that ‘formal’ means here is that I can identify the symbols entirely by their shapes. Now suppose also that I am given a third batch of Chinese symbols together with some instructions, again in English, that enable me to correlate elements of this third batch with the first two batches, and these rules instruct me how to give back certain Chinese symbols with certain sorts of shapes in response to certain sorts of shapes given me in the third batch. Unknown to me, the people who are giving me all of these symbols call the first batch “a script,” they call the second batch a “story. ‘ and they call the third batch “questions.” Furthermore, they call the symbols I give them back in response to the third batch “answers to the questions.” and the set of rules in English that they gave me, they call “the program.”

Now just to complicate the story a little, imagine that these people also give me stories in English, which I understand, and they then ask me questions in English about these stories, and I give them back answers in English. Suppose also that after a while I get so good at following the instructions for manipulating the Chinese symbols and the programmers get so good at writing the programs that from the external point of view that is, from the point of view of somebody outside the room in which I am locked -- my answers to the questions are absolutely indistinguishable from those of native Chinese speakers. Nobody just looking at my answers can tell that I don’t speak a word of Chinese.

Let us also suppose that my answers to the English questions are, as they no doubt would be, indistinguishable from those of other native English speakers, for the simple reason that I am a native English speaker. From the external point of view -- from the point of view of someone reading my “answers” -- the answers to the Chinese questions and the English questions are equally good. But in the Chinese case, unlike the English case, I produce the answers by manipulating uninterpreted formal symbols. As far as the Chinese is concerned, I simply behave like a computer; I perform computational operations on formally specified elements. For the purposes of the Chinese, I am simply an instantiation of the computer program.”

This is the Chinese Room.

So what does this have to do with chess and Kasparov and spiritual machines? Deep Blue, Searle says, is a Chinese Room: it doesn’t actually understand chess, it just pretends to do so. It acts as if it did, but there’s no proof it does. It is, by any other name, a simulation.

Which makes me ask the question: am I, too, a Chinese Room? By Searle’s own logic, simulating a mind and having a mind are two different things. “The map is not the territory.”

Or is it? Like a wolf in sheep’s clothing, the Chinese Room is the simulation hypothesis in disguise—equally unfalsifiable. A commonsense counter-argument, something in short supply among philosophers, is that two objects that are perfectly identical and occupy the same point in spacetime are one and the same. A rose is a rose is a rose.

Which means: if a simulated mind is in all things equal to a base-reality mind, it is a mind. Is it not? Minds are not magical, after all. There is no magical agency particle inside the brain. There is no Wizard of Oz pulling the lever.

Why Memo is a watershed moment in AI

Back to where we started: is Memo conscious? Or, as they used to say in the Middle Ages, does Memo have a soul?

Unlike the minds we talked about a while ago, Memo is magical.

Watch.

So what is Memo? Memo is an autonomous robot by Sunday Robotics—and it’s much more than that. It is the world’s first — in fact, it is the one and only — autonomous robot for the home. Powered by state-of-the-art AI models and an ever-expanding Skill Library, Memo is not teleoperated. As of today, it can load up your dishwasher, throw out food scraps, fold the laundry, clear tables, and make coffee, all without any human pulling the strings. The cool thing is, the sky’s the limit to what it can do tomorrow.

Thanks to Sunday Robotics trademarked Skill Capture Glove, Memo has been learning from an army of so-called Memory Developers, i.e. humans who quite literally teach Memo by doing.

Wearing the glove, they perform a task once while Memo records every motion and decision, transforming that demonstration into a reusable Skill that Memo can execute, refine, and combine with others in the real world. This genius invention has allowed Sunday Robotics to move faster, and more efficiently, than any other physical AI startup. The results are jaw-dropping. What other robots can do this (pick two glasses with one hand)?

The answer is: no one. (No one deployed in the real world, at least.) And Memo does that autonomously. This is a big deal.

What’s more, Memo is safe: it is designed from the ground up to not cause any danger for big and small members of the home. Yes, it has wheels. Why? Because this way, it’s not at risk of falling on you if it loses power unexpectedly.

Wheels, weird hands. What for? Not just for practicality, but also, to reduce costs. The price tag is unknown at the moment, but odds are it will be more affordable than any other humanoid in development. Which is essential if you want to realize the vision of a robot in every home.

And the hat. I mean, isn’t that cute? Now, I don’t want to sound like a groupie (although I am), but the hat is the thing that separates Memo from any other humanoid right from the get-go. Funny, unpretentious, straight out of a Lego movie. While other robots feel scary and unsettling, Memo leaps above the Uncanny Valley to find its place in every home.

So, does Memo have a soul? To which I answer: do we? Perhaps Memo is a philosophical zombie and there is no ghost beneath the shell. Even so, I don’t think we are in a position to deny it the dignity of the human soul. In a 2014 Reddit AMA, Peter Thiel compared the birth of artificial intelligence to first contact with an alien civilization. This, in my opinion, is the right framework. We should approach AI from an empirical, not metaphysical point of view. Although it is made by humans, AI is a phenomenon, and we should talk about it no differently from how we talk about a storm or the double-slit experiment.

Assuming consciousness is not just an illusion, Memo feels unlike any other robot that I know of. It feels alive. And I can only wonder what wonders it will do when we finally cross the Rubicon and find ourselves after AGI.

Subscribe now

Disclaimer:

• a) At the time of publishing this article, I own zero shares in Sunday Robotics—I wish I had some!

• b) This article should not be misconstrued as financial advice. Do your own research.

• c) I put a lot of effort into this article. If you like it, please help me grow my publication by subscribing: it’s free. I publish weekly, bringing you a glimpse from the extropian future, and the ideas and markets wiring it all together. And if you also decide that the article is worth sharing with someone in your life, that’d mean the world to me.

Subscribe now

Share

Once, Zhuang Zhou dreamed he was a butterfly, a butterfly flitting and fluttering about, happy with himself and doing as he pleased. He didn’t know that he was Zhuang Zhou. Suddenly he woke up and there he was, solid and unmistakable Zhuang Zhou. But he didn’t know if he was Zhuang Zhou who had dreamt he was a butterfly, or a butterfly dreaming that he was Zhuang Zhou.

That is the dream argument in a nutshell.

Which begs the question: Are we living in a simulation? According to philosopher Nick Bostrom, we might be. If there were a substantial chance that our civilization will ever get to the posthuman stage and run many ancestor‐simulations, Nick explains, then how come you are not living in such a simulation? Either we never make it to build simulated worlds, or we choose not to, or simulated minds within simulated worlds will so vastly outpace base-layer inhabitants that our odds of living in a dream inexorably approximate to 1.

The thing is, we are already building simulated worlds. To some extent, we might say that the history of humanity is the history of simulation. If we push Harari’s argument to the extreme, every facet of our culture, including the substrate that it runs on, is a figment of simulated worlds. “Homo sapiens rules the world, Harari says, because it is the only animal that can believe in things that exist purely in its own imagination, such as gods, states, money, and human rights.” Simulation is necessary for cooperation. “You could never convince a monkey to give you a banana by promising him limitless bananas after death, in monkey heaven.”

Now, as we approach the Gentle Singularity, simulation has become so sophisticated, so economically meaningful that it will soon outgrow the value of base reality. How big is the market for simulated worlds? If we count just books and movies and video games — as if we were still living in the 90s — the number appears comfortably small, but the closer we get to AGI, the larger and more nonsensical it becomes.

Or, is it? As humanity moved from speech to text to multimedia, so is AI moving from large language models, with text at its cornerstone, to world models, where fiction is no longer a story you simply tell to someone else: it turns into something you can live in.

Just a few days ago, World Labs, a spatial intelligence company by Fei-Fei Li, the so-called godmother of AI, launched Marble, a text-to-world application that turns words into 360° explorable environments. Genie 3, a maybe more powerful world model by Deepmind, “can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p.”

Marble, Genie 3 are the real deal: far from smoke and mirrors, they take us downstairs—from base reality down to its many simulated layers.

How many layers can there be? The answer to this question is a puzzle. Right now, the scope of the simulation can still be measured. But as we enter the second half of the chessboard, we are to lose the pathway out of the maze. Gradually, then suddenly, simulated worlds will dominate the Light Cone: VR; AR; LLM-powered video games; digital twins and industrial metaverses, where androids dream electric sheep; autonomous vehicles, spatial intelligence, physical AI; humanoids in charge of almost all of labor; holodecks and brain computer interfaces; wetware; edge AI; agents and their own training environments; world models; entire supply chains rehearsed in silico before a single atom moves.

As intelligence becomes too cheap to meter, not simulating will be ridiculous. From tokens to pixels to actions and consequences, everything will be simulated. No physical asset will be deployed, no single penny will change hands, if first we didn’t simulate it. This is how the pipeline flips: the simulation morphs into the substrate as reality becomes the commit stage.

So, why Unity Software?

Unity is a game engine. Together with Unreal Engine by Epic Games, Unity is the dominant player in the video game industry. Made with Unity games and apps now reach billions of devices worldwide, with 3.6 billion downloads every month, and developers can ship to more than 25 platforms from a single codebase, including iOS, Android, PC, consoles, and all major VR headsets. More than 70 percent of the top 1,000 mobile titles, more than a quarter of the top PC games, and over 70 percent of the best selling VR games are built with Unity, and a community of millions of creators opens the editor every month to push new worlds into existence.

But, as they say, the devil is in the details. Unity is not a game engine. Unity is a simulation engine. The same stack that powers games is already used to build digital twins of factories for Hyundai, airports, and defense bases; to prototype and test robots through Unity Robotics Hub and ROS, where arms and mobile platforms learn grasping, navigation, and manipulation before touching the real world; to generate synthetic datasets for computer vision with the Perception package, producing perfectly labeled images for object detection, segmentation, and pose estimation at scale; to train RL agents and embodied policies with ML agents in 3D environments where they learn to walk, drive, and explore; to create interactive training simulators for surgeons, mechanics, and plant operators; to design AR and VR workflows for field technicians, warehouse pickers, and remote assistance; to power automotive HMIs, cockpit visualizations, and UX prototypes for self-driving stacks; to visualize IoT streams, sensor fusion, and fleet operations in industrial metaverses; and to act as the front end for AI-driven analytics and control systems where models see, decide, and act inside a live 3D representation of base reality.

At this point, the question is not ‘how can Unity get a bigger slice of the gaming pie’ or ‘how can Unity ground-and-pound Applovin in the ads game.’ Like Search allowed Google to fund Other Bets that today are starting to pay out —Waymo, YouTube, Google Cloud, TPUs, Android, Transformers etc. — gaming and advertising give Unity the monopoly money it needs to transform into an AI-first simulation engine.

If simulated-world GDP grows larger than physical GDP, if every factory, city, warehouse, battlefield, environment, machine has a shadow twin where policies, agents and robots are stressed before they ever touch reality, then profit shifts to the engine that powers the simulation.

Most of the inhabitants will not be humans. They will be agents like Deepmind’s Sima 2, machines, pieces of software, fleets of humanoids and vehicles, all training and coordinating in closed loops. World models like Marble and Genie 3 generate vast implicit environments and pretraining curricula. They are powerful components, but the real value accrues above, at the application layer.

Marble, Genie 3—they set the stage where the magic happens, but they do not spell the incantation. They lack a general-purpose, cross-platform runtime that can serve as the system of record for how thousands of agents, robots and humans interact; they lack an editor and asset pipeline that millions already know; they lack deployment to billions of endpoints; they lack the unglamorous product infrastructure for input and UI, networking, replay, access control, analytics, patching, and everything else you need to turn a simulation into something the entire economy can live in.

Unity is the engine of creation. It is the general-purpose runtime where simulated worlds become products and systems of record. It already has the editor, the scene graph, the input and UI stack, the networking, the asset pipeline, and a deployment story that reaches billions of devices and more than 20 platforms. It already ingests CAD, BIM and ROS, already drives smart factories, airports, bases, medical and industrial training sims. Unity Studio pushes this further by moving the authoring layer into the browser and putting it in the hands of process engineers, city planners, logistics managers and AI researchers, who can pull in CAD models, Omniverse USD scenes or, in time, layouts born from world models, then wire agents, robots and data feeds into live applications without ever touching engine code.

Stripe wants to increase the GDP of the internet. Unity’s logical endgame is to increase the GDP of simulated worlds. The pot of gold at the end of the rainbow is in the trillions. That is the scale of the Unity wager.

Subscribe now

Disclaimer:

• a) At the time of publishing this article, I own shares in Unity Software Inc. (U).

• b) This article should not be misconstrued as financial advice. Do your own research.

• c) I put a lot of effort into this article. If you like it, please help me grow my publication by subscribing: it’s free. I publish weekly, bringing you a glimpse from the extropian future, and the ideas and markets wiring it all together. And if you also decide that the article is worth sharing with someone in your life, that’d mean the world to me.

Subscribe now

Share

I like non consensus plays. After all, why pick stocks if not to beat the S&P 500? Sure, you could go the Joseph Carlson way and pick the best compounding machines the market has to offer, and rip the rewards. Or, you could go after momentum plays and chase the OKLOs and the Palantirs of the world, hoping the time bomb doesn’t explode right while you are holding it. Unity is neither of those. You are a lion running after a gazelle. You don’t follow the gazelle in its footsteps. You try to triangulate it and be there before it is. Unity is the gazelle. Right now, you can’t see it being valued at one trillion dollars, but that’s just because you’re following the market in its footsteps. Instead, use your imagination. After all, what is beating the stock market if not an exercise in imagination? You need to picture the future well before it takes shape. And in the future, 5, maybe 7 years from now, Unity is worth one trillion dollars.

This is why.

#1 Unity is hated - very much so

Nobody likes Unity. And why should anyone? The company is losing money hand over fist, and even tried to scam game developers with the runtime fee thing back in 2023. Which makes for a perfect setup for a moonshot type of return. Think about it. Moonshots are made by sifting through shit in the gutter. No one is gonna give you the perfect stock on a silver platter. It’s gonna be ugly, and you ain’t gonna like it. Needless to say, all that glitters is not gold, and sometimes, often times, almost always in fact, shit is just shit. So what makes Unity not shit?

First of all, management has been upended. Since Matt Bromberg became CEO in 2024, interim CEO Jim Whitehurst has moved to executive chair of the board, and Unity has fully killed the runtime fee, reverting to a simpler, seat-based subscription model. Workforce has been reduced by 25%, and the company has exited film/VFX bets like the Wētā professional services piece. A new CFO has been appointed (Jarrod Yahes (ex-Shutterstock)). The CTO has resigned, which is not good news per se, but we’ll see what the future holds. And, most importantly, Vector (more on this later).

Basically, it’s a whole new company. Sure, you could say that “that which we call a rose, by any other name would smell as sweet”. And to some extent, you would be right. Not everything has changed. In fact, what was good about Unity hasn’t changed. That is…

#2 Unity owns gaming (mostly of, at least)

Well, this might be a stretch. Unity has competitors: Unreal Engine in its Create segment, and AppLovin in its Grow segment. But it still owns 70% of mobile gaming and 51% of gaming on Steam, which makes investing in Unity right now the same type of bet you could do by investing in Netflix in 2022, when hating Netflix was popular. My point is, you don’t wanna be the new Bill Ackman and sell Unity at 45 dollars. Unity is the Netflix of gaming, kinda. It’s got a quasi-monopoly in the sector and why should you ever bet against a monopoly?

#3 Unity is a turnaround play

As I said, Unity is hated, which means it’s most likely mispriced. Remember Peter Lynch: people always misprice hated stocks. Since nobody likes them, nobody takes the time to price them correctly. You can do it. And not because you’re a phd in maths or anything like that. You don’t need to work at the Medallion Fund to recognize an opportunity. Temperament, like Buffet says, is the most important quality of a stock picker. It’s guts. It’s like what they say about streets and blood, the time to buy is when there’s blood in the streets. Right now, there is. The street is called Unity. And you know what’s good about this street? They’re cleaning it.

Not only has Bromberg gone back on the runtime fee shit, it also has admitted the company’s mistakes, which, if you ask me, makes for a good leader, and promised no further shenanigans in monetization schemes. Since Bromberg is at the helm, the company is “no longer at war with its customers” and has been shipping consistently. Unity 6, with mixed-to-improving sentiment from developers. And Vector, Unity’s new AI ad platform, which let me tell you, is a major deal: it’s boosted the ad network revenue 15% sequentially, with guidance for continued double-digit growth in Q3 - all while bringing 15-20% growth in the number of app installations and in-app purchase value. This is a turnaround in the making.

#4 Unity is a secret AI play

So I just told you why the company is a good investment: everyone hates its guts, it’s a quasi-monopoly, and it’s a turnaround play. What I didn’t tell you is, Unity is a secret AI play. Vector is just the start (now, bear with me, because this requires some imagination).

Everyone knows Unity as a gaming company, and as you know from Nvidia, gaming is AI’s secret cousin. Indeed, the largest pie of Unity’s future earnings won’t come from advertising, and neither from gaming: it will come from physical AI.

Physical AI - also called embodied AI - is a big pie. We’re talking high double-digit CAGR and a TAM as large as the entire physical labor market. There are several players in the space, the main ones being Tesla with Optimus, Figure with Figure, Apptronik with Apollo, 1X with Neo Gamma, and the Chinese company Unitree with a slew of humanoid and non-humanoid robots.

What is the opportunity? Perception. Synthetic data. Let me explain. Humanoids, like LLMs, need to be trained on massive datasets. Which begs the question: where do you get a big enough dataset to train these robots on common everyday tasks? I mean, there’s not plenty of POV videos of people doing the laundry on YouTube. But that hasn’t stopped Figure AI to train its humanoid to do just that. How did they pull it off? Human video collection. Which means: they used actual data coming from the real world.

Recently Figure CEO Brett Adcock announced Project Go-Big, the company's "push to build the world's largest humanoid pretraining dataset. We are scaling out human video collection and our F.02 humanoid can now learn directly from human video. This project is accelerated by our partnership with Brookfield, who owns over 100,000 residential units."

How scalable is this? Not much. First, usable videos of humans doing shit in first person is a scarce resource. Second, even if it were as big as the internet, it wouldn’t be enough. There’s been some chatter earlier this year about LLMs needing synthetic data after eating all the internet. The same thing will happen with robots. Unfortunately, LLMs and VLAMs (Vision-Language-Action-Models) are not yet able to generalize the concept of a dog after seeing a dog once, like humans do. I mean, my 4-yo nephew doesn't need to watch millions of dogs to understand what a dog is, but AIs do. They’re artificial intelligences, not artificial GENERAL intelligences, so to speak. My point is, it is OBVIOUS they’ll need synthetic data. Think about how big the human experience is. We can do millions of physical tasks, from petting a dog to folding clothes, and on and on. You can’t reasonably expect this humongous dataset to come from people labelling real-world videos. There aren’t enough videos to begin with.

And where will this treasure trove of synthetic data come from?

Digital twins - this is what you need. You need someone to build a virtual environment where digitized humanoids can roam freely and collect data about the world. This is the story of AlphaGo Zero. While tackling the challenge of building the world's strongest AI Go player, the genius people at Google's DeepMind taught their AI all the rules of Go and trained it with a dataset of expert human games. Then, they had a breakthrough. Why stop at that? Why not train the AI against itself? That's what they did. In October 2017, DeepMind unleashed AlphaGo Zero upon the world, a version created without using data from human games - without even teaching it the rules of Go! By playing only against itself, AlphaGo Zero was able to beat AlphaGo Lee in three days by winning 100 to 0, reached the level of AlphaGo Master in 21 days, and exceeded all previous versions in 40 days.

There is a lesson to be learned from this parable: to go superhuman you need synthetic data. Enters Unity. Unity is used to build interactive, physics-based environments where you can (a) simulate robots and sensors, (b) connect real robot stacks via ROS, and (c) generate data to train/validate physical AI before deployment. In short, Unity ships and maintains the core building blocks used in humanoid-training twins:

• Unity Robotics Hub (open-source tutorials and packages for ROS, URDF, pick-and-place, pose estimation, etc.)

• ROS-TCP Connector

• Perception (Unity’s official synthetic-data toolkit)

• Sensors/SystemGraph and docs for LiDAR/depth/IMU and authoring

Basically, Unity controls the entire software stack for humanoid training. And that's not all. Unity's digital twins deployments don't stop at humanoids. They're used at

• Hyundai Motor, for building a real-time twin of its plants

• Vancouver International Airport for staffing, passenger-flow, maintenance and decarbonization planning

• Orlando Regional, which includes a 40-sq-mile high-fidelity core, used for coordinating agencies and exploring infrastructure scenarios

• U.S. Air Force (Tyndall AFB) - we’re talking about a “largest-of-its-kind” twin used for design, resilience, day-to-day planning, and sustainment after Hurricane Michael

• Sitowise, which built a Unity-based Hong Kong 5G radio-propagation simulator for planners, and earlier a district-scale city twin (Hervanta) to test urban scenarios

• Bell Textron, where they built a program/turbine mockups & facility twin for design review.

• Prespective (Unit040), where they built a Unity-based industrial simulation suite used by OEMs to prototype workcells, run throughput studies, and de-risk commissioning

What these names tell us is the massive opportunity at the heart of Unity's digital twin endeavor. Not just humanoids: live IoT/IT data in a 3D ops view; unified “single pane” for mobility, 5G, zoning, climate resilience, and more. The metaverse will get very real very fast, although it won’t resemble anything of what Zuck envisioned. It won’t be a B2C metaverse. At first, the metaverse will be B2B, and Unity will be its dream machine.

Subscribe now

Disclaimer:

• a) At the time of publishing this article, I own shares in Unity Software Inc. (U).

• b) This article should not be misconstrued as financial advice. Do your own research.

• c) I put a lot of effort into this article. If you like it, please help me grow my publication by subscribing: it’s free. I publish weekly, bringing you a glimpse of the extropian future, and the ideas and markets wiring it all together. And if you also decide that the article is worth sharing with someone in your life, that’d mean the world to me.

  Subscribe now

  Share