I'd like to begin discussing my analysis of research in general, as it is not a universally accepted concept. Research is primarily applied to R&D companies within the biotech and AI spheres. But we are seeing the boom of multiple sectors within the tech territory, and I think the three that I should focus on are:
- AI
- PSI
- Biotech
First, what is the biggest sector right now? Currently, the most mature industry is biotech ($1.6T), comprising numerous large companies that dominate the market. They don't do R&D; they acquire the results of 1% of smaller, successful R&D startups. That's why the pharma sector sub-sector of biotech is both highly successful and dull at the same time. Nothing changes within them, but we are seeing a lot of new companies (the LBCs) emerge, and this is due to the rise of generative AI.
People think generative AI is going to solve biology, that they're going to solve everything. Currently, it remains a fairytale; therefore, the generative AI subsector within biotech remains relatively small (with even less narrative cohesion across the board). It is not as booming as the PSI subsector (within the robotics industry) and the AI industry I'm about to talk about.
The second largest (a close second) is approximately $700B. And with the kind of attention/care from the US government, due to this being a tech arms race, it's going to rise to the top if AND ONLY IF the bubble doesn't burst. The third industry I'm focusing on is robotics, which is currently not highly valued but is growing rapidly. As of December 2025, it's roughly $70B, and it's growing, albeit rapidly.
Returning to the main discussion, we are seeing a pattern that was something not many people discussed. There's a great transition from pure business, which we often associate with entrepreneurship, to pure research.
Before, we had Apple, Samsung, Sony, Nvidia, etc. - their valuations are based on the products they sell to consumers. All B2C. Microsoft is another example. With Windows and Office, everything is consumer-driven, and I'd still consider their enterprise customers to be borderline B2C. That's how they're valued so highly. Google is no different. They dominate the online advertising industry. That was their main product. That's why they are valued so highly. That was the old tech sector.
The significant transition occurred in late 2022 with the release of GPT-3. Ever since then, research has become the new pedigree for large tech. They are no longer thinking about how to sell. If you look at OpenAI today, they are losing money every day; for every dollar they earn, they lose three dollars. This is common sense. Why are they doing this? How is their valuation compared to the estimated private valuation of around $500B, and if they do an IPO, it will be approximately $1 trillion. This valuation doesn't come from something extrinsic. It's not entirely dictated by the number of active users they have, but rather by how well the model performs and how closely it approximates this hypothetical AGI.
The valuation comes from its expected potential. The potential they have to reach this level of AGI monopoly.
Of course, we still see many consumer applications that are doomed to fail, except for a few notable exceptions (e.g., Lovable, Manus, and Cursor - perhaps not Cursor). Most consumer products that are wrappers built on top of foundational AI lack relevance purely because the foundational models and the companies behind them are moving at such a rapid pace that the consumer products cannot really keep up. Those that have AI as the core of their product, I think, are not going to make it—most of them. The large foundation companies don't really concern themselves with products at this point, and OpenAI self-adjusted itself recently by initiating Code Red, purely because, even though they were trying to run ads and jump into the adult industry, Google was full-throttling research and was thoroughly caught up, surpassing everyone else in terms of the capabilities of the models that. As a result, OpenAI is refocusing its efforts on research and has launched GPT 5.2. Very prudent move. They will continue this race toward AGI alongside other companies, such as Anthropic, xAI, and Google DeepMind, among others.
All these large VC firms or investors they're betting they're betting on this binary outcome, and there are two end scenarios:
- It's not a bubble. It's gonna be AGI and it's gonna unlock human prosperity for centuries by enabling everything downstream
- They are not going to make it, and the bubble bursts
This is the same bet that PSI companies are making, especially the software players, including full-stack ones like Tesla. We're seeing DeepMind trying to build their own OS for robotics. We're seeing startups like π. We're seeing Tesla manufacturing Optimus with their own data and own modeling. The race toward PSI is also equally binary.
Different companies take different approaches, though. Build's strategy centers on data scaling; they believe more data translates to more human-like behavior. π takes the opposite approach: not quantity but quality, not egocentric first-person data at scale, but data variety and extreme modeling for max versatility. They're not building specialized systems for different manufacturing tasks; they're building a universal system for all robots.
Hardware is less of a bet and more of an engineering iteration problem. Improvements are inevitable. Unitree is already there and is much better than Boston Dynamics in terms of actuation and build quality. They're going to make it. But software is the main bottleneck. Hardware iterates gradually; it's not a bet, just steady improvement. Software (the mind of the robot) is where the fundamental uncertainty lies.
Meta is capturing first-person egocentric data. Google is doing something, though I'm not sure what exactly. Everyone's taking different methodologies, but with software, it's still a one-versus-zero play. No in-between. If they make it, robotics goes sky-high. If whoever reaches PSI first, they dominate everything. If Tesla wins, they own the entire industry. If software companies like π win, they own the Windows of robotics and need to start worrying about their hardware customers.
Everyone is transitioning to research. Founders, CEOs, boards, investors, governments—everyone's betting on pure research. There are predictions, but no end in sight. We predict AGI in two years, but are we really close? Is that the exact, precise timeline? This dynamic will compound in biotech as well. Most drug startups will fail. The ones that succeed will win big—acquisition, IPO, whatever. But it's rare. And biotech faces an additional challenge: there are so many different diseases, and no magic drug cures everything. It's divergence instead of convergence. Companies like π, Tesla, and OpenAI are trying to build universal models. It's the opposite direction in biotech (drug discovery-wise, not necessarily computationally - I will go more in depth later on).
The game they are playing isn't about generating revenue. It's a race against time—specifically, the duration between reaching investor milestones and investor patience. You'll never run out of money if you make significant progress every year. But you'll lose if you don't deliver what you promise. It's a game of tripling down. If it reaches zero instead of one, you collapse; if no one can reach one, America dies.
I hope AI and robotics succeed because I want to depend on them as foundational tooling/enablers. I use AI to build AI. I'm constantly exploring what automation can apply to biomanufacturing. Both generalist and specialist models are beneficial to me.
Additionally, my approach carries low risk because my R&D commitment is minimal. We're investing in tooling and personnel while establishing a groundbreaking paradigm within the drug discovery sphere.
It's not the time yet for biology to fully embrace Generative AI and expect it cures everything. It won't cure everything. What we need now is assistance-level tooling, and that's what AI and robotics automation will be for biology over the next five to ten years. What I'm doing is a launchpad to pursue R&D down the line. Break into an industry with low risk, build revenue, accumulate market share and influence, then start exploring R&D. The digital twin concept will emerge then. But I still want to frame it as tooling, not as some ultimatum for humanity. I'm playing in the most mature market as a shovel seller selling the newest type of shovels no one has ever seen (with immediate revenue and establishment of credibility).
Long-term goal: computational biology twins that map out all molecular interactions within the human body, simulating diseases related to oncology, regenerative medicine, aging, longevity, mitochondrial degeneration. All of that will be enabled down the line. But right now, the best thing to do is find a revenue stream. Heavy research isn't worth it yet.
People say startups require risk-taking. True - but the best startups take calculated risks. Generate revenue before burning investor money on pure research. Your lifeline depends on investors; if you lack a good narrative or fall behind schedule, it's detrimental. The best approach is what Google did. That's why they'll win as soon as they integrate Titans architecture + MIRAS into Gemini 4, which I expect will happen as soon as next year.