The Path to 1 Trillion X Intelligence!

Twelve learnings, observations, and quasi-predictions from building at the intersection of AI, energy, and the physical world (plus lots of podcasts, videos, and articles).

Energy → Tokens → Intelligence → Inventions → Products → Abundance

That sequence is how I think about the world and the future.

My twelve learnings, observations, and quasi-predictions:

1. Intelligence is the product of several multipliers. A million, billion, trillion, etc. X increase is inevitable.

It sounds crazy, but we are going to get 1,000x intelligence. Then 1,000,000x. Eventually 1 quadrillion X (15 zeroes). Not from one breakthrough. From stacking gains that are already happening. In the same way that Moore's Law has taken us to an unfathomable number of transistors. Experts estimate that roughly 30 sextillion (30,000,000,000,000,000,000,000) transistors have been manufactured globally since they were invented in 1947.

Bigger models. Model efficiency. Falling token cost. Parallel agents. Better memory and context. Better algorithms. Bigger data centers. More data. Synthetic data. Intelligence here means useful ability to do work, and every one of those raises it.

Each is worth a multiple on its own. Multiply all of them and 1,000x is conservative. Perhaps per year before long! Multiply again and a million is a number you can say out loud without embarrassing yourself. The math is the easy part. Figuring out what to do with it is where things get fun!

Meanwhile everybody is arguing about whether the new model is four percent better on some eval. Who cares! The marginal price of intelligence is → $0 and the increases are compounding and accelerating.

2. Agents are the great catalyst and there will be trillions of them accelerating innovation.

Loops, graphs, and a goal. The modular unit of the scientific process. Agents actually do the work.

I run Claude Code and ChatGPT Codex every day. I spent last weekend inside Grok Bot. OpenClaw and Hermes Agent come at it from different angles. Now there is Muse, Instinct, and dots. Soon you use agents off a shelf for every task, or you wire them together yourself with LLMs, MCPs, and APIs. To me, that is manufacturing engineering and systems engineering. Less time writing code, more time designing the line.

Code replicates and runs 24/7. A few lines of it take output and thinking up 1,000x. Efficiency and productivity are great. That is how costs come down, and costs coming down is what frees everybody up to go invent. Automation is the on-ramp.

3. We're going to invent 1,000x amazing things and the impacts are unfathomable.

First chat. Then code. Then reasoning models. Now automation, agents, and loops. Next is invention, and invention will be 1,000x automation (and take 1,000x the compute).

Automation makes what we already have cheaper. Invention makes things that did not exist before. Both matter. The second one is much bigger and more impactful.

Most of the new stuff will be combinatorial, which is exactly what these models are already good at. Take what exists, search the space of combinations, test faster than any human team could. Drugs. Materials. Recipes. Movies. Products.

The wildcard: when can we automate the invention process itself? When does physical AI (robots) merge with digital AI? Just imagine automated labs with robots. Run the loop. Generate candidates. Test them. Keep what works. Repeat.

Zuckerberg argues a version of this in The Future Is For Everyone. He wants personal superintelligence in individual hands instead of concentrated inside a few institutions. I believe that everyone can become a builder.

With each invention, the flywheel accelerates and gains compound.

4. The core technologies are already good enough for a decade of inventions and products.

The "Product Overhang" is enormous. And it will be for years to come. The models are good enough. Go look at Artificial Analysis, Hugging Face, or OpenRouter. Thousands of models across dozens of labs. Gemini, Claude, ChatGPT, Qwen, xAI, the whole run of Chinese models. Open and closed. Small and cheap, large and expensive. Any flavor you want.

And almost nothing sits on top of them yet. As Nat Friedman argues, chat and coding tools and agents are barely the start.

If every lab stopped training tomorrow we would still have a decade of products to build. Anybody telling you the models are not ready is describing their own imagination. The gap between what these things can already do and what anyone has shipped is the biggest opportunity on the board.

5. We are going to build 1,000x more software.

Because we can and because it is so simple and cheap. Software exists to solve a problem. When the marginal cost hits zero you build the solution and then you delete it, or you just forget about it. Software will be built with single prompts by one user. Perfectly customized. Run locally. Spun up instantly. Never reviewed, never shared. It does not matter.

Minimal bells and whistles. Not meant for production. Not meant to persist. All the complexity and the calcs abstracted away so nobody using it ever sees them.

It may not even look like software. Excel is software with a UI so good the coding disappeared, and the formulas sitting on top of the database are very real. Everyone is already a builder. World models allow you to play games forever as each frame is spun up in real time.

Everyone can become a builder and it is hard to overstate the potential impacts.

6. Business is evolving to value outcomes and impact over inputs and outputs.

The economy has been evolving this way for years. People will pay for performance. Results matter more than effort. The software underneath stops mattering, and so do the tokens. Especially when agents can create outputs.

A useful ladder:

• Inputs. Hours, employees, seats, tokens.
• Outputs. Reports, messages, actions, leads.
• Outcomes. Resolved problems, completed hires, saved energy, delivered capacity.
• Impacts. Increased profit, avoided cost, better health, faster time-to-power.

The farther down that ladder you go, the more valuable the offering becomes. Measurement, attribution, and risk-sharing get harder at the same rate. For now, humans are still needed to take responsibility for outputs and drive outcomes and impacts.

Why pay for inputs and outputs when you can buy the outcome? AI is what makes that work. It helps define success up front and then verify it was delivered.

7. We may all start vibe-living. Abstraction FTW!

This one also sounds crazy, but hear me out. You've probably heard of vibe coding: giving in to Claude Code or ChatGPT Codex to write software for you with single prompts (and then just speaking English to have it improve). Vibe-living is similar. The ultimate luxury, and the natural extension of number six. The richest of the rich, the people with access to the most abundance, just request things. In English.

You request an outcome and you name the spend. $X in, Y result out. Things get done, built, designed, and fixed.

Underneath it is an auto-combo. Software, robots, and capital assembled on the fly against a verifiable outcome, then taken apart once that outcome is met. A temporary assembly that existed to produce one result.

Vibe coding was the small version. The big version is that you stop specifying how any of it happens. Verification is the whole trick. Define done, prove done, put money against it, and let the system figure out the rest. Personal agents are a huge step here. Robotics is the next step.

Abundance abstracts away the effort and the inputs it takes to get an outcome.

8. Routing of [model, agent, tool, data center, robot] will be huge.

There is already a plethora of models, agents, and tools to keep up with. Nobody is hand-picking a model per task for much longer. Router companies, or the frontier labs themselves, will do it for us.

Route on capability, task complexity, cost, quality, reliability, data privacy, security, scalability, and repeatability. Tokens have different flavors and different costs. Workflows need different latencies, reliability, and uptime. Training jobs are nothing like inference jobs. Cheap and fast for the easy 80 percent, expensive and slow for the rest.

Tools and agents are proliferating the same way, into prepackaged workflows that solve one specific thing. Whoever owns that layer owns a lot.

It's all just abstracted away.

9. Voice finally works and you need to try it (again, I know!).

Voice frickin' works. Huzzah. Fast, accurate, and the old bugs are gone. No more fighting the transcript or cleaning up every third word. Cross-platform like Wispr Flow, built into ChatGPT, built into my Tesla. It all just works. Except Siri, of course.

I talked most of the first draft of this article. Why does that matter? Because typing has been a governor on thinking my entire working life. I think at conversation speed and I type at about a quarter of it, so most of what I thought never made it out of my head. That bottleneck is gone.

10. Speed to power (not "chips") is the limiter today.

Every link in that earlier sequence runs on electrons. The bottleneck has shifted (for now) from GPUs to watts (technically watt-hours). Energy is the rate limiter.

Abundance requires access to energy as the rawest input (and preferably affordable energy at that). Lack of energy in the transitioned form of electricity at nearly any price is the biggest point of friction today. Solar, wind, storage, geothermal, nuclear, and gas. We need everything today and far more than we can supply. Building more is also how prices come down, and prices coming down is what massively accelerates the sequence.

The grid has more buffer in it than people assume. Solar and batteries. Flexible compute. Virtual power plants. Demand response. Behind-the-meter gas. Repurposing commercial and industrial sites that already have interconnection. Modular data centers. Data centers overseas. Data centers in space. Data centers in homes (SPAN XFRA). Data centers at commercial sites and at corporate parks (Crusoe Spark). And even waiting in the interconnection queue to connect to the grid.

Energized megawatts are the binding constraint on the next decade. Bring all the gigawatts and terawatt-hours we can get!

11. There has never been a better time to start a company.

Everything above stacks into this one. Cheap intelligence. An embarrassing product overhang. Instantaneous, free software. Buyers who want outcomes. Routing that keeps the bill down. A million X this; a billion X that. WTF!!!!

One person with a stack of agents covers what used to take a whole team.

The floor to start has never been lower and the ceiling on a small team has never been higher. More companies. More things that exist because one person could finally make them. MORE!

12. There has never been a better time to be alive!

An abundant future does not arrive on its own. It needs to get built. Don't wait. Have agency. Get building. And have fun! The future is for everyone.