Mark Fidelman (@markfidelman) header
Mark Fidelman (@markfidelman) profile picture

Exascale Labs marketing leader | AI infrastructure, data centers, chips and robotics | Author of Agent Shock™

Los Angeles, CA
Joined March 2009
Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)Photo by Mark Fidelman (@markfidelman)
“AI is taking over the world” but nobody’s talking about the two massive problems that could kill the whole revolution before it even starts. Let’s unpack it. 🧵
6
6
1
45
24,002
Tau Robotics, an MIT startup is sending humanoids to clean San Francisco homes. $100 for a one-time visit. But Humans are driving them by remote. MIT Tech Review's reporter watched one wobble on stairs and knock over olive oil. Let’s call it like it is, a robot training data-collection business with a mop.
1
133
Xiaomi says its humanoid worked an eight-hour shift sorting parts at its EV factory. It gave no part counts, no success rate, no word on how often a human stepped in. So the eight hours sounds like a PR pitch, let’s see the video
3
1
2
828
Buff Billionaires?
🤖 Made with AI
180
Agility Robotics is on Nasdaq. TechCrunch: $174K revenue in the first half of 2026, mostly from related parties. $362M loss. $300M in "planned contracts" from 30 unnamed customers. Show me the invoices.
1
1
150
A bitcoin miner just signed a $28 billion deal with an AI lab. The lab is bringing the chips. So what does the miner have that’s worth a 15-year lease? That’s one question in Issue 20 of AgentShock. I also get into why Goldman took coding agents away from its bankers, why Uber is negotiating with AI assistants after they booked just four rides, and how Amazon lost a data center vote by nearly four to one. The numbers tell a stranger story than the headlines. Read Issue 20: [link in the comments]
2
1
2
9,908
Google switched Gemini on for many K-12 students in August. Now Gemini and Search are off by default for student accounts until schools opt in! 150 million Classroom users. Then a rug pull reversal 😁 #AI #EdTech
1
2
3
523
Since Sep 2: 16,794 MW of AI data centers announced. But only 81.4 MW have been energized. Same board: 16.3T tokens a day through the top ten models, 84% of them Chinese labs. The agents are compounding. The grid is falling behind. AgentShock 019 is out. #AIAgents #DataCenters
2
2
5
271
A person with ALS just communicated at 75 words per minute using nothing but their brain. No keyboard. No speaking. Just thoughts they converted into words. Neuralink says its implant achieved this in the first week, with only a 5.8% error rate across 125,000 words. The results haven’t been independently verified yet, but imagine what this could mean for people who’ve lost their ability to speak. Pretty incredible.
4
8
588
Mark Fidelman retweeted
Tesla is celebrating Halloween this year by having a Cybercab autonomously drive around Austin with skeletons inside haha 🎃
847
4,590
591
43,201
1,165,611
Google's best business isn't one we use every day. * Cloud revenue: $30.7B, backlog $561B * Search growth slowed to 12% * Capex $51.4B, free cash flow negative * Reddit: over a million fewer weekly Google referrals * The stock dropped about 10% after hours My take: Search made Google rich. TPUs and cloud are paying for what comes next. Their Gemini AI answer box is taking away the clicks, and the people (like me) who made the content are feeling it. If your traffic came from Google, what's your plan for when it doesn't? #Google #GoogleCloud
2
2
411
You Said Cheapest.
🤖 Made with AI
1
1
201
Mark Fidelman retweeted
Speaking for Exascale Labs, I think Elon is right that some neoclouds won't survive tighter financing. But I wouldn't write off the whole group. I'd look at who's paying for the compute and whether the business can cover its power bills and debt if a big customer leaves.
Replying to @chamath
Pretty much. As of the past few weeks, debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers. SpaceX closed $40B of debt for GPUs, but we’re investment grade and have Starlink & launch revenue. That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many.
3
7
327
The biggest Nvidia customer just hedged: * Oracle expands Vera Rubin to multi-gigawatt scale * Google TPUs offered on Oracle cloud for the first time * Cerebras and SambaNova inference chips coming too * Oracle’s own Acceleron chips replace some Nvidia networking * Capex target still about $90–95B My take: Training is Nvidia’s. Inference is up for grabs, and buyers know it. The multi-vendor shelf is the new moat for clouds. If you rent compute, are you locked to one chip, or can you move workloads?
1
2
3
315
The humanoid market’s first real volume is coming: *Agile Robots: 1,000–2,000 compact humanoids next year *Production starts end of 2026, ships early 2027 *Runs Nvidia Isaac GR00T on Jetson Thor *Target buyers: universities and corporate labs *Meanwhile ANYmal X is inspecting oil and gas sites for Equinor My take: Labs buy hardware to build the software that makes factories viable. That’s how this works. It’s slower than the demos, but at least it's real #Robotics #Humanoids
1
2
3
317
That’s a pretty big disagreement to build a career around. Jeff Bezos thinks AI is more likely to help save us than destroy us. Stephen Hawking warned it could become the worst thing to happen to humanity. I pulled together 15 predictions from 2016 through 2026, with the names, dates, and timelines attached. Same technology. Very different futures. Which prediction do you buy? And which one do you think will age terribly? 👇 The infographic puts them side by side.
2
2
4
217
Whoa... OpenAI says it’s building a humanoid robot. Demo targeted for 2027. I assumed the LLMs would stay in the cloud. Apparently, the brain wants a body, and Tesla just got some real competition (albeit well behind). #Robotics #OpenAI
3
1
2
577
Mark Fidelman retweeted
If you want to understand what Elon Musk is really fighting, and who is really fighting him, this is the clearest map I can give you :
🚨 I spend an enormous amount of time trying to understand the chaos we're living through. Not the headlines. The machine behind them. 🚨 If you want to understand what Elon Musk is really fighting, and who is really fighting him, this is the clearest map I can give you. Expensive houses. Expensive schools. Expensive healthcare. Taxes that rise every year while everything gets worse. Borders nobody voted to open. Speech rules nobody voted for. Most people look at this and blame capitalism. The market failed, they say, so the state has to step in and fix it. They have it exactly backwards. What they want to fix it with is what broke it. This isn't chaos, and it isn't capitalism. It has a direction, a method, and a hundred-year history. Let me reverse-engineer it. In 1926, Mussolini's regime threw Antonio Gramsci, co-founder of the Italian Communist Party, into prison. From his cell he asked the question that would shape the next century: why did the revolution win in Russia but fail in the West, where Marx said it was inevitable? His answer: in Russia, the state was everything, so seizing it was enough. In the West, the state was only the outer wall. Behind it stood the real fortress: the church, the family, the school, the press, the culture. Western workers never revolted because the culture held them. So the conclusion was simple. Don't storm the palace. Capture the culture. Win the institutions and you never have to win the argument, because you decide which arguments are allowed. In 1967, the German student leader Rudi Dutschke gave it a slogan: the long march through the institutions. Paris supplied the intellectual software, and I've already apologized for my country's part in that. Then came the most patient operation in modern history. The students of the 60s became the professors of the 80s. Their students became the journalists, judges, civil servants and HR directors of the 2000s. Their students now run the foundations, the NGOs and the agencies. No single step looked like a revolution. That's exactly why it worked. The method was communist. The ambition became global. And it grew like a cancer. A cancer never attacks the body head-on. It hijacks the body's own cells and feeds on the body's own resources. This system never had to defeat the West. It just had to make the West pay for it. In America, the machine found a political home: the Democratic Party. Not the millions of decent people who vote for it. The apparatus. The donor networks, the activist groups, the consultants, the captured agencies that keep moving in the same direction whoever wins. In Europe, it found something even better. The EU's founders, Schuman, Adenauer, De Gasperi, were Catholic anti-communists who wanted to make war impossible. But they built exactly what the long march dreams of: power that never has to face a voter. A Commission nobody directly elects holds the right to propose laws for 450 million people. Chat Control. The digital euro. Speech rules written in Brussels. Nobody stormed anything. They just drafted it. Add the World Economic Forum, the UN agencies and the global foundation circuit, and you have the full map. You don't need a secret meeting when everyone went to the same schools, reads the same papers, funds the same NGOs and flies to the same conference in Davos every January. It's not a conspiracy. It's a class with shared incentives. That's harder to fight, because there's no headquarters to raid. And the billionaires? Most people who became the richest in the world got absorbed. They built something real, then traded it for an aura. Bill Gates is the perfect example: he built the software that ran the world, then became the bankroller of global health bureaucracies and the funder of newsrooms covering his own causes. Laurene Powell Jobs bought The Atlantic. Benioff bought Time. You're not buying a business. You're buying the right to be called good by the people who decide who is good. That's the corruption. Not suitcases of cash. Money for aura, grants for loyalty, NGOs for policy. All legal. All invisible. Here's the proof it's one system and not a hundred separate problems. Look at what happened to prices over the last 25 years. Everything exposed to markets and technology got cheaper: TVs, software, phones, clothing. Everything controlled by institutions exploded: college, healthcare, childcare. And housing, where you need a committee's permission to build anything. Same country, same economy, same decades. The difference isn't the product. It's who controls it. Taxes go up and services get worse, because the money doesn't reach the services. It feeds the layers. Immigration happens at a scale no electorate clearly chose, because for a system that lives on managing problems, more problems means more budget. And a two-income family today can't buy the house their parents bought on one salary. That's real poverty, manufactured by a machine that is rewarded for looking good, never for being good. MAGA isn't the cause of the chaos. It's the immune response. Brexit, the farmers' revolts, the votes shaking every European capital: same thing. The body finally noticed the tumor. So what we're living through is not a normal political cycle. It's a fight to the death between two models. One runs on the perception of goodness. The other runs on reality. And it had to be the richest man in the world. Anyone smaller would have been bought, sued or cancelled. Every billionaire before him took the deal. Elon walked out of the club and turned the lights on. He bought Twitter for $44 billion, opened the Twitter Files and gave the world back a public square where any narrative can be checked in real time. He went inside the government and helped shut down USAID. Now Politico announces that the Democrats' gavels are coming for him. His answer: he'll find out who is really pulling the strings and go after them, not just their puppets. He has one line he keeps coming back to: he cares about the reality of goodness, not the perception of it. That sentence is the whole war. Now the part that makes me certain of how it ends. A lie is expensive to maintain. It needs gatekeepers, censors, fact-checkers, friendly journalists, a budget, and constant repairs every time reality contradicts it. The truth costs nothing to maintain. It just keeps being true. AI is driving the cost of checking anything toward zero. Any budget, any study, any statistic, any promise, audited in seconds by anyone with a phone. Yes, AI can generate lies at scale too. But producing a lie is cheap. Defending it against a billion instant audits is impossible. That's what acceleration really means. It doesn't pick sides. It speeds up feedback loops. Systems aligned with reality compound. Systems built on lies hit the wall faster. And you cannot build rockets, design chips or cure diseases with a false model of the world. Physics doesn't read op-eds. The more intelligence we pour into the world, the more expensive every lie becomes. You cannot scale a lie. You can only scale what's true. We are close to the collapse of the old order and the birth of a new one. Not the one from the WEF's own 2016 video, "you'll own nothing and you'll be happy." The opposite. An order built on truth-seeking, efficiency, science, private property and free markets. You will own things. You will build things. And you'll be happy, because it will be real. Gramsci's long march took a century. Its collapse will take years. History will remember this moment. A man walked into the most powerful machine ever built, refused the deal everyone else took, and chose reality over perception. I have never been more optimistic.
102
899
105
2,030
125,148