The Tao Pod

There are two primary bottlenecks to today's AI. And solving these bottlenecks is where the money is right now. A) DATA - the big models have ALL the data already. All the texts, videos, reddit posts, tweets, etc. How can they get more data? Other than the natural daily creation of new data the biggest repository of UNUSED data is behind the corporate firewall. Whoever can unlock the data behind the firewall will create a model bigger than anything OpenAI can do. The key is this: corporate data does not want to be free (competition reasons or even legal reasons in the case of healthcare) So the method has is decentralized learning techniques where AI trains locally inside the firewall, and then shares its weights (but not the data that created those weights) to an AI model that stitches this together. This is where $TAO is the best in the world and at a $2.5B value is cheap. B) CONNECTIVITY = POWER. Everyone is complaining about power. But people don't realize that the power problem is really a connectivity problem. 100,000 GPUs in a data center means a million or more copper / electricity interconnects connecting them all. Change that from electricity to MicroLEDs and you wll use 1/30 of the power. Problem solved. Who is doing this? $MRVL $ALAB $FABC $KOPN

THE NEW WAY TO VALUE A BUSINESS Price per "intelligence" is going straight down. Claude, OpenAI, and the open weight models (Kimi, Qwen, Deepseek, GLM, etc) are in a fight to zero. The infrastructure will still exist (people will NEED AI) but these models are a giant question mark to me. That said, how do you value companies when intelligence becomes an abundant commodity? You can no longer say, "My 50 developers spent 10 years working on this so that's my moat!" And intelligence is not just coding. Its a Super Bowl commercial (goodby ad agencies) . Its a legal contract. Its management consulting and banking. Its supply chain logistics. Its some of medical care. The new moats are: TRUST - e.g. I trust $HRB or $INTU to do my taxes. I'm not going to say to Kimi K3 "do my taxes" DISTRIBUTION - e.g. this is where $IBM or $ORCL might end up living. The average S&P 500 co doesn't want to hire a bunch of AI engineers to figure out how to streamline their business using AI. IBM already has the business and tech relationships with everyone. They will keep those and place the right AI products in companies. DATA - this is the true bottleneck of AI. AI already has ALL OF THE DATA. Except for: - data behind a corporate firewall (banks, pharma, logistics, etc. - data in a hospital (HIPAA laws) - hard-to-get data (the twitter feed, all YouTube videos) Data is why HuggingFace was bought for $12.9B last night. How does data get solved? Decentralized AI techniques (so as to train frontier level LLMs while keeping the data private behind the corporate firewall. Only two companies for this - Prime Agent (where NVDA is invested) and $TAO, a crypto. ($TAOX the public entity holding TAO The "next OpenAI" is whoever combines decentralized learning with building corporate consortiums (give companies equity pro-rata based on the data they contribute) to build a "Federated AI". What might be dead - new game studios, $ADBE (eventually anyone can edit their videos, photos, etc), $CRM (although they can argue they have distribution). Note: Physical AI, AI infrastructure, are not "intelligence companies" but picks and shovels of intelligence and they will be valuable. $NVDA $MRVL $ALAB $CLS $AVGO etc etc will be valuable for a long time.

At $TAOX, the largest public holder of $TAO tokens, we are super bullish. A) $TAO is the largest decentralized AI platform. B) $TAO is like an ETF of 128 AI state-of-the-art subnets. C)@const and the other developers on TAO have immense integrity. D) $TAO (with 128 valuable companies) has only a $3B market cap vs companies like OpenAI with $1.5Trillion in value. E) $TAO is evolving and will become only better with time, through governance and advances in tech. $TAO is not a crypto but an AI platform that is succeeding and under-noticed. $TAOX is the public vehicle to invest in TAO. We are invested in $TAO , the subnets, and own the largest media outlet about TAO, the @taodaily_io as well as public-facing front-end into TAO inference : BlueTAO.ai

KIMI K3 vs OPENAI / ANTHROPIC KIMI is disappointingly slow. I'm going to describe why and what stocks will benefit because - guess what,- Kimi, DeepSeek, etc are all about to get A LOT better.  Here's the key difference: Kimi K3 is an extreme "mixture of experts." It's a 2.8 trillion parameter model (almost as big as Claude), but for each token it activates just 16 of its 896 experts — about 104 billion parameters. So it's "playing" on a much smaller (but hopefully just as accurate) slice of the model. Those 896 experts are spread across many GPUs. And every token can call a different set of 16 — so the GPUs are constantly passing pieces of the answer back and forth. This hammers the Interconnects.  Now assume the closed models — Claude, GPT — run dense or close to it. A dense model knows exactly where its weights live. The data flows through the same GPUs in the same order, every prompt, all planned in advance. Claude pays for that with more raw compute per token — but it barely touches the slow wires between machines. THIS IS THE BOTTLENECK OF KIMI. And for the 1000s of companies that will adopt Kimi because its cheaper. The wires between machines are the slow part. Inside a GPU, data moves at roughly 8 TB/s today. Between GPUs could be as low as 0.1 TB/s. Around 1% of the speed. It's like Kimi is riding a Ferrari and has to stop and take a bicycle every mile or so — while Claude drives one car on one highway the whole way.That's a big part of why Kimi K3 feels slow on most providers right now, while Claude feels instant. Kimi's architecture is arguably smarter per dollar of compute it's chatter runs over the slowest links in the data center. Which is why interconnects ARE THE TRADE underneath the entire open-model story: as interconnects improve, extremely sparse open-weight models are the workloads that benefit the most. Fix the plumbing and the open models' can/will surpass the closed models. The "plumbing" for 40 years was CPUs. Then GPUs. Now...Interconnects.  Who's in the race: $CRDO — the leader in copper-based electrical interconnects (active electrical cables). The incumbent technology, and the one hitting physical limits first. But copper is hitting physical limits. Too. slow.  $NVDA — just put $2 billion each into $COHR and $LITE to lock up laser-based optical interconnects. $MRVL — led the latest funding round of private Mojo Vision and is jointly developing MicroLED-based optical interconnects with them. Read that again: Marvell is betting on MicroLEDs. But no product. No MicoLED.  Avicena — private, well funded, has demoed MicroLED links but nothing scaled yet. Then there's $FABC, the highest reward play in my opinion and I'm a shareholder, which is jointly developing a high-speed, energy-efficient MicroLED interconnect (Neural I/o) with $KOPN — Big advantage: KOPN has years of MicroLED manufacturing experience, including for the military — and $FABC will own the IP with $KOPN on these interconnects. The company says it's targeting a demonstration of the platform by the end of 2026.The tell that MicroLED interconnects are real: Marvell, a $172+ billion company, just made the same bet and Jensen Huang then stated, "MRVL is the next trillion dollar company."  Kimi is painfully slow today. That's not a model problem(!) — it's a plumbing problem. And the plumbing is where the money is going.

WHY IS TECH GOING DOWN AND WHAT IS GOING TO HAPPEN? A lot of things, added together, are making people nervous about AI/tech/data centers, etc. Hence this selloff. I'm going to address them one by one: A) NVDA backstopping OpenAI $250B data center buildout. Meaning: NVDA will back the loans OpenAI is taking out to buy $NVDA chips. Feels a little circular. People forget that this is normal in every industry. You help your customers pay for your goods (e.g. car companies own car financing companeis) so that your customers can grow and then become bigger customers. Its just weird to see this at a 100 billion dollar+ level. But it is more noise than signal. B) $MU . Chinese competitor $CXMT IPOing. Are they competitive and cheaper? Answer: of course not! 1) MU is BOOKED OUT through 2028 anyway. And even with that there is still a shortage. 2) MU Trades now at 5x earnings. For 30 years it hasn't traded lower than 6. 3) Does anyone really think America is going to store all of its memory on Chinese memory chips? 4) MU can just use its own cash to buy back the company in a couple of years. Thats insane. C) $ASML. A trillion dollar monopoly with China just announcing they can compete. Yes, with 20 year old DUVs. Nobody has the ASML EUVs. ASML has a 20 year head start on this. There is no competition. D) Kimi K3. This is a big infrastructure issue. If open-weight Kimi K3 (valued at $ 30 billion ) is as good as Opus 5, as people are suggesting, then why is Anthropic valued at $1 trillion. And will this affect all the data center buildouts that Anthropic, OpenAI, Google, etc are planning. The answer is NO, but its complicated. 1 ) Kimi K3, lets face it, sucks compared to Claude. I don't care what the benchmarks say. I am using Kimi K3 hosted by @chutes_ai . For simple app development K3 is taking 20x as long as Opus 5. 20x! Its means nothing if the results are great if I can't even get the results (my app requests are relatively simple compared with what an enterprise woudl request). 2) Let's even say Kimi K3 is as good (it isn't). Then MORE data centers will be built because you still need massive infrastructure to support mega-demands on open weight models like Kimi. So this is a positive for $NVDA (GPUs) $MRVL (connectivity) $INTC (CPUs) $FABC (interconnects) $ALAB (interconnects) . 3) Well, what if Kimi K4 is as good (it won't be). Then I'll use Claude Opus 8 whenever that comes out. But if it is as good then great, see #2. Here's what's happening. All of the above news together has created uncertainty. I get it. People sell first, ask questions later. But each one of these issues turns out not to be an issue. Here's the REAL issue: Claude Opus 5 and ChatGPT 5.6 ROCK! But what that means for stocks requires another post.

The biggest theme is definitely the speed of inference in AI. The main bottleneck is power usage. But this bottleneck is not solved by providing more electricity or power. Its solved by making the interconnects that connect XPUs to each other, faster. This is what will surprise people from an investing standpoint.

TOMORROW - from 330-430 ET . Q&A . ask me about: stocks, entrepreneurship, economy, relationships, writing, crappy people, crypto, fun, depression, marriage, divorce, anything at all.

MARKETS ARE HAVING A 'KIMI MOMENT' The markets are having a "Kimi moment", similar to when DeepSeek was released in early 2025. Markets are selling off thinking that big AI companies won't need to build out as much infrastructure . But this isn't true. Kimi K3 is the latest open source AI model out of China (interestingly, it first came out the box saying, "Hi, I'm Claude", suggesting, as many people think, that it trains on Claude data it steals). Kimi is good. Very good. And so people are worried: will there be the same demand for data center buildouts if open source models are just grabbing the trophies? They had the same worry in January 2025 when DeepSeek came out. The answer is: demand will be greater than ever. Any selloff due to fears of no buildout will be reversed. A) JEVON's PARADOX. When a technological commodity becomes more efficient, demand for that commodity actually goes up. An example is DeepSeek, the first open source AI model out of China. The same selloff happened. Guess what? Capex went up every quarter since then. When a resource becomes cheaper, people think of more applications for it, and demand continues to rise. B) THE MODEL IS DIFFERENT FROM THE INFERENCE - just because someone trained a model more efficiently, doesn't mean demand for inference goes down. If demand goes up (as suggested in "A"), then the need for GPUs, CPUs, interconnects, etc go up. C) THE SELLERS OF COMPUTE DON'T CARE. Oracle, Microsoft, Coreweave, Facebook, etc - anyone selling compute and inference, they don't care what models they are hosting - they just want to povide inference for the people. If anything, their margins will go up if they use cheaper models. D) AGENTS - Last week was the first week that human traffic on the internet was exceeded by agent traffic (traffic being done by AI-created bots). Agents require 50x the amount of "tokens" (the units that AI is prices in) than humans do. This does not change with new models being released. If anything, the better the model, the more agents will be made with it. E) FRONTIER MODELS TAKE ON THE COMPETITION Anthropic proved last February that many of the Chinese open source models are stealing Claude's inference to train their own models. Does this mean Anthropic and OpenAI give up? Of course not. They will just make better models. Even since February, Anthropic has released Fable 5 and OpenAI has released ChatGPT 5.6, which are still the best two models. These trillion dollar companies won't rollover. They will build faster than ever. F) ITS ALREADY DONE Land purchases, power purchase agreements, GPU sales, Memory sales, Interconnect sales, etc etc are already contracted out through 2029. Let's talk then but chances are the rise of new models will continue to create new demand for inference and the buildouts will continue. But, for sure, the next three years trillions of dollars will be spent. G) THE LINUX EXAMPLE. Another example of Jevon's Paradox. Linus was an open source free operating system in the 90s. A cheaper operating system allowed MORE software, data centers, operating systems, etc to be created, resulting in a huge data center buildout in the 2000s. This Kimi Moment will mirror the DeepSeek moment of early 2025. Everyone will reiterate capex. Money will be exchanged. Chips will be bought. Stocks will flourish.

I've been on Twitter / X for 20 years. I've loved every minute of it. Social media has saved my life. When determining whether to make a post and if the post will create value, I like the Rotary International "4 Way Test": - Is it TRUE - Is it FAIR (i.e. it doesn't harm anyone) - Will it build GOODWILL and BETTER FRIENDSHIPS - Will it be BENEFICIAL to all concerned. Just following this test I've made so many friends here, tested ideas, and have even built businesses and written books because of content created here. I try to never skip this test before posting. Maybe this is corny. But I just wanted to say this. Thank you, X.

Product idea: as soon as I say "Bye" on a Zoom call, it should disconnect me. Else there's always those 1-5 seconds where both people are looking around for the "leave" button. Awkward. This would be the next quadrillion dollar company