🇨🇭 Global One SI - Prompt
$PLTR/USD H2 #Indicator #Signal #Trading #GlobalOneAI Globalone.ch #Prompt #AI
This media is not supported in your browser
VIEW IN TELEGRAM
Palantir $PLTR just posted this video of CEO Alex Karp:
“We’re hoping that you’ll go to a large language model company and learn that they don’t care about you at all.”
“We’re in the business of giving you the ability to solve those problems for yourself and own the means of production.”
“We’re hoping that you’ll go to a large language model company and learn that they don’t care about you at all.”
“We’re in the business of giving you the ability to solve those problems for yourself and own the means of production.”
🇨🇭 Global One SI - Prompt
It’s getting REAL 🔥🚀
Morgan Stanley expects $SPCX to reach $3.4T in annual revenue by 2040.
$3.4 trillion in annual revenue. $2.7 trillion in adjusted EBITDA. By 2040.
Morgan Stanley — one of the lead underwriters of the SpaceX IPO, the bank that has seen every number in the S-1 — just published those projections with their name on them.
Let that sink in.
$2.7 trillion in EBITDA is more than the entire current GDP of France. More than every Fortune 500 company’s profit combined. From a single company. In 14 years.
Now remember what SpaceX is actually building toward.
Launch. Reusability. Connectivity. Compute. Models. Mars.
Point-to-point Earth travel. Orbital data centers — Anthropic paying $1.25 billion a month, Google in active talks. Asteroid mining. Lunar manufacturing. Mars colonies — with Elon’s pay literally tied to getting there.
$28.5 trillion total addressable market across all businesses per the S-1.
Morgan Stanley is modeling $3.4 trillion in revenue by 2040 — roughly 10% of that TAM. If SpaceX captures even half of what it’s building infrastructure for, $3.4 trillion looks conservative.
$SPCX starts trading June 12th at $1.75 trillion.
Morgan Stanley just told you where they think it goes.
The rails are being built. The trains haven’t started running yet.
And the bank that led the deal just said the destination is $3.4 trillion.
$3.4 trillion in annual revenue. $2.7 trillion in adjusted EBITDA. By 2040.
Morgan Stanley — one of the lead underwriters of the SpaceX IPO, the bank that has seen every number in the S-1 — just published those projections with their name on them.
Let that sink in.
$2.7 trillion in EBITDA is more than the entire current GDP of France. More than every Fortune 500 company’s profit combined. From a single company. In 14 years.
Now remember what SpaceX is actually building toward.
Launch. Reusability. Connectivity. Compute. Models. Mars.
Point-to-point Earth travel. Orbital data centers — Anthropic paying $1.25 billion a month, Google in active talks. Asteroid mining. Lunar manufacturing. Mars colonies — with Elon’s pay literally tied to getting there.
$28.5 trillion total addressable market across all businesses per the S-1.
Morgan Stanley is modeling $3.4 trillion in revenue by 2040 — roughly 10% of that TAM. If SpaceX captures even half of what it’s building infrastructure for, $3.4 trillion looks conservative.
$SPCX starts trading June 12th at $1.75 trillion.
Morgan Stanley just told you where they think it goes.
The rails are being built. The trains haven’t started running yet.
And the bank that led the deal just said the destination is $3.4 trillion.
✍1
🇨🇭 Global One SI - Prompt
Jim Simons, founder of Renaissance Technologies, on why the most successful fund in history was built on collaboration, not finance: Simons opens with a confession: "I'm not your typical businessman." What follows is a formula for building a team, one he…
This 17 page pdf reveals the same technique Hedge Funds like Jim Simons' Renaissance Technologies use to find signal through noise.
Stanford released the complete Hidden Markov Model framework for everyone to use it.
Bookmark it before someone takes it down:
web.stanford.edu/~jurafsky/slp3/A.pdf
Stanford released the complete Hidden Markov Model framework for everyone to use it.
Bookmark it before someone takes it down:
web.stanford.edu/~jurafsky/slp3/A.pdf
✍1
🇨🇭 Global One SI - Prompt
This 17 page pdf reveals the same technique Hedge Funds like Jim Simons' Renaissance Technologies use to find signal through noise. Stanford released the complete Hidden Markov Model framework for everyone to use it. Bookmark it before someone takes it…
The secret of Hedge Funds is revealed in a 6 page PDF.
Stanford released the complete LSTM neural network framework for trading that quants at firms like Citadel & 2Sigma are known to use & released it for free.
Bookmark it before someone takes it down:
cs230.stanford.edu/projects_fall_2020/reports/55614857.pdf
Stanford released the complete LSTM neural network framework for trading that quants at firms like Citadel & 2Sigma are known to use & released it for free.
Bookmark it before someone takes it down:
cs230.stanford.edu/projects_fall_2020/reports/55614857.pdf
✍1
This media is not supported in your browser
VIEW IN TELEGRAM
Superhuman-level🤯
a Unitree G1 carrying a box up onto a high table.
(Perhaps Amazon could consider using it for deliveries? G1 + Rivian)
a Unitree G1 carrying a box up onto a high table.
(Perhaps Amazon could consider using it for deliveries? G1 + Rivian)
🇨🇭 Global One SI - Prompt
Elon Musk on Terafab: "It's worth noting that there's not a single high volume computer memory fab in America right now, zero. There's one being built by Micron, but that will not reach volume production until I believe 2028 and there's something built in…
D3 Chips Don’t Just Tolerate Space They’re Built For It
Engineered to run hot, they shed heat straight into the vacuum through passive radiators, eliminating the energy tax of Earth cooling. Paired with uninterrupted solar arrays, the result is 10× FLOPS per watt.
This is how xAI-scale intelligence moves off-planet. Not a workaround. The native architecture.
Engineered to run hot, they shed heat straight into the vacuum through passive radiators, eliminating the energy tax of Earth cooling. Paired with uninterrupted solar arrays, the result is 10× FLOPS per watt.
This is how xAI-scale intelligence moves off-planet. Not a workaround. The native architecture.
🇨🇭 Global One SI - Prompt
D3 Chips Don’t Just Tolerate Space They’re Built For It Engineered to run hot, they shed heat straight into the vacuum through passive radiators, eliminating the energy tax of Earth cooling. Paired with uninterrupted solar arrays, the result is 10× FLOPS…
D3 chips achieve 10× FLOPS/watt in orbit by turning vacuum physics into an advantage: hotter junctions, passive radiation, constant solar.
The deeper question is what happens to civilization when the marginal cost of intelligence falls this sharply off-Earth.
Does AI abundance accelerate the Mars transition or does it change what “abundance” even means?
Curious to hear builders’ takes.
The deeper question is what happens to civilization when the marginal cost of intelligence falls this sharply off-Earth.
Does AI abundance accelerate the Mars transition or does it change what “abundance” even means?
Curious to hear builders’ takes.
🇨🇭 Global One SI - Prompt
D3 chips achieve 10× FLOPS/watt in orbit by turning vacuum physics into an advantage: hotter junctions, passive radiation, constant solar. The deeper question is what happens to civilization when the marginal cost of intelligence falls this sharply off…
This media is not supported in your browser
VIEW IN TELEGRAM
Galactic Civilization Is Forcing a Complete Revolution in Chip Design
Earth’s energy limits make terawatt-scale AI impossible on the surface. That’s why Terafab is deploying massive orbital compute clusters powered by constant solar energy, while redesigning silicon from the ground up for vacuum, radiation, and fully autonomous off-world labor.
This Drives Four Non-Negotiable Architectural Choices:
>>Radiation-hardened D3 family
80% of output is destined for orbit. The D3 isn’t a commercial chip with hardening added later—it’s built from the transistor up with triple-modular redundancy, finFET structural tweaks, and advanced ECC to survive ~10¹⁵ cosmic ray hits per year at 99.999% uptime.
>> High-temperature vacuum optimization
No air, no liquid cooling. Heat must radiate into space. By deliberately engineering the D3 to run safely at much higher junction temperatures, radiator mass and size drop dramatically—delivering roughly 10× better FLOPS/watt in orbit than any ground-based system.
>> Massive on-chip SRAM for edge autonomy (AI5/AI6)
Mars propellant plants and orbital construction can’t rely on Earth comms. Optimus robots need real-time bipedal control and reasoning entirely on-device. That’s why half the accelerator area in the AI5/AI6 lines is dedicated to enormous SRAM—shattering the memory wall and boosting effective bandwidth by an order of magnitude.
>> Recursive design-manufacturing loop
At 100–200 billion chips per year, 12–18 month cycles are obsolete. Terafab integrates design, lithography, fabrication, and orbital simulation under one roof. Iterate overnight, test in radiation chambers, revise masks, and reprint compressing development from quarters to days.
This isn’t incremental progress. Every transistor decision is subordinated to making humanity multi-planetary and eventually multi-stellar.
The chips being taped out right now are the literal substrate for the next branch of human civilization.
Earth’s energy limits make terawatt-scale AI impossible on the surface. That’s why Terafab is deploying massive orbital compute clusters powered by constant solar energy, while redesigning silicon from the ground up for vacuum, radiation, and fully autonomous off-world labor.
This Drives Four Non-Negotiable Architectural Choices:
>>Radiation-hardened D3 family
80% of output is destined for orbit. The D3 isn’t a commercial chip with hardening added later—it’s built from the transistor up with triple-modular redundancy, finFET structural tweaks, and advanced ECC to survive ~10¹⁵ cosmic ray hits per year at 99.999% uptime.
>> High-temperature vacuum optimization
No air, no liquid cooling. Heat must radiate into space. By deliberately engineering the D3 to run safely at much higher junction temperatures, radiator mass and size drop dramatically—delivering roughly 10× better FLOPS/watt in orbit than any ground-based system.
>> Massive on-chip SRAM for edge autonomy (AI5/AI6)
Mars propellant plants and orbital construction can’t rely on Earth comms. Optimus robots need real-time bipedal control and reasoning entirely on-device. That’s why half the accelerator area in the AI5/AI6 lines is dedicated to enormous SRAM—shattering the memory wall and boosting effective bandwidth by an order of magnitude.
>> Recursive design-manufacturing loop
At 100–200 billion chips per year, 12–18 month cycles are obsolete. Terafab integrates design, lithography, fabrication, and orbital simulation under one roof. Iterate overnight, test in radiation chambers, revise masks, and reprint compressing development from quarters to days.
This isn’t incremental progress. Every transistor decision is subordinated to making humanity multi-planetary and eventually multi-stellar.
The chips being taped out right now are the literal substrate for the next branch of human civilization.
✍1
🇨🇭 Global One SI - Prompt
3 Simple & Effective Bias Correction Methods for Your Polymarket Weather Trading Bot (Decaying Average + MOS + Kalman Filter) If you're building a trading bot for weather markets, one of the most common mistakes is jumping straight into U-Net, XGBoost, and…
Media is too big
VIEW IN TELEGRAM
Stanford just dropped the exact linear regression math that prints millions on Polymarket weather markets
1 hour 48 minutes. free. straight from Stanford
bookmark & watch - this is the most honest “how real weather models actually work” lecture ever published
forget the "AI weather bot" YouTube grifters. this is the classical foundation meteoblue (swiss weather service) uses: probabilities instead of point forecasts, decades of bias correction, and why the market stays mispriced at 6¢
exactly why your decaying average + MOS + Kalman stack actually works. no hype, just edge
then start building your own bot
1 hour 48 minutes. free. straight from Stanford
bookmark & watch - this is the most honest “how real weather models actually work” lecture ever published
forget the "AI weather bot" YouTube grifters. this is the classical foundation meteoblue (swiss weather service) uses: probabilities instead of point forecasts, decades of bias correction, and why the market stays mispriced at 6¢
exactly why your decaying average + MOS + Kalman stack actually works. no hype, just edge
then start building your own bot
✍1
🇨🇭 Global One SI - Prompt
Anthropic CEO: “At Anthropic, we essentially have Claude designing the next version of Claude itself.” Meanwhile, Anthropic says every new model is given code that trains a small AI model and asked to make it faster. Claude Opus 4 achieved a 3x speedup.…
Anthropic just called for a global way to slow frontier AI because its own models may be approaching recursive self-improvement, where a system helps build a stronger version of itself without direct human control.
Future models will become so good at research, experiments, debugging, and training design that humans will stop being the main bottleneck.
Once that loop starts, progress could shift from human-paced engineering to machine-assisted improvement, which makes every safety test, law, and lab policy feel late by default.
Anthropic says this has not happened yet, but warns that the jump may arrive before governments, companies, and researchers have a trusted way to measure or restrain it.
The hard part is verification, because a huge AI training run is easier to hide than a weapons site, and any lab that secretly keeps training while others pause could gain the lead.
Anthropic is now ~$1T, may reach $50B annualized revenue, and competes fiercely with OpenAI, so every safety claim also lands inside a giant business fight.
Future models will become so good at research, experiments, debugging, and training design that humans will stop being the main bottleneck.
Once that loop starts, progress could shift from human-paced engineering to machine-assisted improvement, which makes every safety test, law, and lab policy feel late by default.
Anthropic says this has not happened yet, but warns that the jump may arrive before governments, companies, and researchers have a trusted way to measure or restrain it.
The hard part is verification, because a huge AI training run is easier to hide than a weapons site, and any lab that secretly keeps training while others pause could gain the lead.
Anthropic is now ~$1T, may reach $50B annualized revenue, and competes fiercely with OpenAI, so every safety claim also lands inside a giant business fight.
🇨🇭 Global One SI - Prompt
⏰ Oh yeah super excited about this! 🚀 A new local 12B dense model from Google! You have to try this out. 💥 Google just dropped Gemma 4 12B → a powerful multimodal open model that actually runs well on laptops (and cards). Run on 👉 ✦ RTX-4070 TI ✦ RTX-5060…
vLLM major update supports all the DGX Spark +Blackwell optimizations, plus now with support for NVFP4 kv cache get 2x-4x boost in model context capacity!
If using NVFP4 KV Cache you will have to opt for MTP instead of DFlash but worth it for large Agent swarms.
github.com/AEON-7/vllm-ultimate-dgx-spark/pkgs/container/aeon-vllm-ultimate
If using NVFP4 KV Cache you will have to opt for MTP instead of DFlash but worth it for large Agent swarms.
github.com/AEON-7/vllm-ultimate-dgx-spark/pkgs/container/aeon-vllm-ultimate
✍1
This media is not supported in your browser
VIEW IN TELEGRAM
ETH Zürich Unveils HELIOS: A Four-Armed Humanoid #Robot Built for Space Operations
🇨🇭 Global One SI - Prompt
🇺🇸 The made-in-USA "picks & shovels" micro-cap ($115M market cap) riding three revolutions at once: $AMPG. Amplitech sells the "ears" of three secular waves at the same time, space, quantum computing, and AI-RAN $NOK $NVDA. The market still prices it like…
🇺🇸 The made-in-USA "picks & shovels" of the space race that almost nobody owns: $AMPG.
A ~$130M micro-cap with NASA as customer is selling the "ears" of the satellites going up $ASTS $MNTS.
The market still prices it like a sleepy parts maker.
1/ The setup.
The sky is about to get crowded.
Starlink is pushing toward tens of thousands of satellites.
Amazon's Kuiper is launching its own constellation.
SpaceX's anticipated $SPCX listing has the entire space trade buzzing and names like $MNTS are part of the same wave.
Every satellite that goes up has the same problem.
2/ The problem AMPG solves.
A signal from 300 miles up is impossibly faint by the time it reaches a dish.
You have to amplify it, without adding noise, or the data is garbage.
That's a Low-Noise Amplifier.
The literal "ears" of a satellite.
AmpliTech has spent 20 years building the best ones: Ku, Ka and X-band LNAs for SATCOM, designed and manufactured in the USA.
3/ The customers.
This isn't a startup praying for its first order.
AmpliTech already lists NASA among its customers, alongside Lockheed Martin, Northrop Grumman, Boeing, Raytheon and L3Harris.
Two decades supplying the people who actually build America's space and defense hardware.
4/ The "made in USA" moat.
As space turns into a national-security priority, the U.S. government and defense primes increasingly demand a domestic supply chain, no foreign-sourced critical components.
AmpliTech designs and builds its LNAs on American soil. A "safe," sovereign supplier exactly when the demand for one is exploding.
5/ The picks-and-shovels logic.
AMPG doesn't need to win the satellite race.
It doesn't need to pick Starlink vs Kuiper vs anyone.
It just sells the ears to whoever launches.
Every constellation, every ground station, on every continent, all of it needs low-noise amplification.
AMPG sells the same critical piece to all of them.
6/ The thesis in one line.
A ~$130M, debt-free, 🇺🇸 U.S.-made micro-cap supplying the hardware the entire LEO buildout runs on.
Priced like a niche parts maker, positioned like an arms dealer in the space race.
An the best part?
It's still valuable outside space:
- AI-RAN with NVIDIA. The first and only one Open6G.
- Quantum with IBM. The only supllying a critical bottleneck in USA 🇺🇸.
A ~$130M micro-cap with NASA as customer is selling the "ears" of the satellites going up $ASTS $MNTS.
The market still prices it like a sleepy parts maker.
1/ The setup.
The sky is about to get crowded.
Starlink is pushing toward tens of thousands of satellites.
Amazon's Kuiper is launching its own constellation.
SpaceX's anticipated $SPCX listing has the entire space trade buzzing and names like $MNTS are part of the same wave.
Every satellite that goes up has the same problem.
2/ The problem AMPG solves.
A signal from 300 miles up is impossibly faint by the time it reaches a dish.
You have to amplify it, without adding noise, or the data is garbage.
That's a Low-Noise Amplifier.
The literal "ears" of a satellite.
AmpliTech has spent 20 years building the best ones: Ku, Ka and X-band LNAs for SATCOM, designed and manufactured in the USA.
3/ The customers.
This isn't a startup praying for its first order.
AmpliTech already lists NASA among its customers, alongside Lockheed Martin, Northrop Grumman, Boeing, Raytheon and L3Harris.
Two decades supplying the people who actually build America's space and defense hardware.
4/ The "made in USA" moat.
As space turns into a national-security priority, the U.S. government and defense primes increasingly demand a domestic supply chain, no foreign-sourced critical components.
AmpliTech designs and builds its LNAs on American soil. A "safe," sovereign supplier exactly when the demand for one is exploding.
5/ The picks-and-shovels logic.
AMPG doesn't need to win the satellite race.
It doesn't need to pick Starlink vs Kuiper vs anyone.
It just sells the ears to whoever launches.
Every constellation, every ground station, on every continent, all of it needs low-noise amplification.
AMPG sells the same critical piece to all of them.
6/ The thesis in one line.
A ~$130M, debt-free, 🇺🇸 U.S.-made micro-cap supplying the hardware the entire LEO buildout runs on.
Priced like a niche parts maker, positioned like an arms dealer in the space race.
An the best part?
It's still valuable outside space:
- AI-RAN with NVIDIA. The first and only one Open6G.
- Quantum with IBM. The only supllying a critical bottleneck in USA 🇺🇸.
🇨🇭 Global One SI - Prompt
$FJET Last few days, I talked about the technical setup. Now, the story is becoming bigger. The market is finally starting to pay attention to what Starfighters actually owns. Most traders see: A tiny micro-cap. What they may be missing: → Space infrastructure…
$FJET
One of the more interesting Space & Defense setups on my watchlist.
After several months of consolidation, price is now pushing against a key downtrend line while volume continues to improve.
Watching closely to see whether momentum can continue building from here.
One of the more interesting Space & Defense setups on my watchlist.
After several months of consolidation, price is now pushing against a key downtrend line while volume continues to improve.
Watching closely to see whether momentum can continue building from here.
🇨🇭 Global One SI - Prompt
$OUST "To perceive the world in full context requires a combination of structure and colour, and Rev8 is the first sensor to unify both." And Ouster was also the first of the three companies that announced colour LiDAR to actually bring it to market! This…
$OUST
Ouster's native colour LiDAR is really on another level.
And for anyone not too familiar with what it means that Ouster offers 256 channels in its sensor, here's a visual (from a competitor ;)) showing how the number of channels can impact overall LiDAR performance.
Ouster's native colour LiDAR is really on another level.
And for anyone not too familiar with what it means that Ouster offers 256 channels in its sensor, here's a visual (from a competitor ;)) showing how the number of channels can impact overall LiDAR performance.
🇨🇭 Global One SI - Prompt
$OUST Ouster's native colour LiDAR is really on another level. And for anyone not too familiar with what it means that Ouster offers 256 channels in its sensor, here's a visual (from a competitor ;)) showing how the number of channels can impact overall…
This media is not supported in your browser
VIEW IN TELEGRAM
256 vertical channels enable two times the angular sampling density of a standard OS1. That means twice as many laser returns are stacked into the same vertical field of view. In practice, that's the difference between sampling a mural with broad horizontal stripes and resolving the figures inside it.
This is a Rev8 OS1 Max driving Clarion Alley in San Francisco. Clarion has continuous murals from sidewalk to roofline, hand-painted text, and fine pattern detail at every height. The kind of scene where you need much higher resolution to see what's actually there.
This is a Rev8 OS1 Max driving Clarion Alley in San Francisco. Clarion has continuous murals from sidewalk to roofline, hand-painted text, and fine pattern detail at every height. The kind of scene where you need much higher resolution to see what's actually there.
🔥 US macroeconomic data:
▪️ Unemployment rate (May)
FACT: 4.3%
FORECAST: 4.3%
PREDICTION.: 4.3%
▪️ Change in the number of employed in the non-agricultural sector (May)
FACT: 172K
FORECAST: 85K
PREDICTION.: 179K
▪️ Average hourly wage (compared to the same period last year) (y/y) (May)
FACT: 3.4%
FORECAST: 3.4%
PREDICTION.: 3.6%
▪️ Average hourly wage (m/m) (May)
FACT: 0.3%
FORECAST: 0.3%
PREDICTION.: 0.2%
▪️ Unemployment rate (May)
FACT: 4.3%
FORECAST: 4.3%
PREDICTION.: 4.3%
▪️ Change in the number of employed in the non-agricultural sector (May)
FACT: 172K
FORECAST: 85K
PREDICTION.: 179K
▪️ Average hourly wage (compared to the same period last year) (y/y) (May)
FACT: 3.4%
FORECAST: 3.4%
PREDICTION.: 3.6%
▪️ Average hourly wage (m/m) (May)
FACT: 0.3%
FORECAST: 0.3%
PREDICTION.: 0.2%
Goldman Sachs just dropped the most precise map of where $7.6 trillion is going over the next five years and it tells you exactly which companies are standing in the middle of an unavoidable flood of capital (Save this).
The numbers are worth understanding precisely before talking about who benefits.
Goldman's baseline projects $765 billion in AI capital expenditure in 2026 alone, growing to $1.6 trillion annually by 2031.
Over the full 2026 to 2031 period, cumulative spend breaks down to $5.1 trillion in compute, $2.1 trillion in data centers, and $358 billion in power.
Nvidia is assumed to command 75% of all compute spend throughout the period, using the Rubin VR200 chip at $80,500 per GPU as the baseline.
The data center specification charts reveal how dramatically physical requirements are escalating.
A standard cloud data center runs 5–15 kW per rack while a transitional Blackwell era AI data center runs 130–200 kW per rack.
The AI factory of the future, running Rubin and Feynman silicon operates at 500+ kW per rack, at greater than 1 gigawatt scale, with liquid cooling only.
Traditional hyperscale data centers cost roughly $10 million per megawatt to build while the next generation AI data centers are being discussed at $15 to $20 million per megawatt.
Goldman identifies silicon useful life as the single biggest swing factor in the entire model.
At a 3-year replacement cycle, cumulative compute depreciation hits $3.99 trillion and at 7 years, it drops to $2.23 trillion, a $1.76 trillion difference on one assumption alone.
Power is only $358 billion of the total, but Goldman is explicit, it is the only component that can prevent the other 95% of the stack from deploying.
Amazon's Andy Jassy put it, "Our single biggest constraint is power." Connecting large data centers to the grid takes years.
Now here are the companies standing directly in the path of each layer of this capital.
Nvidia is still the most concentrated bet on the compute layer.
At 75% of $5.1 trillion in compute spend over six years, that is $3.8 trillion in cumulative revenue flowing through one company's products.
The 75% gross margin on data center GPUs is the reason every hyperscaler is trying to build custom silicon to escape it while simultaneously continuing to buy Nvidia because nothing else performs at the same level.
Vertiv is the direct infrastructure play on the data center upgrade cycle.
Every rack going from 40 kW to 500+ kW needs liquid cooling systems, power distribution, and thermal management infrastructure that simply did not exist at prior density levels.
Vertiv just deepened its liquid cooling capabilities through a strategic acquisition and was named a key partner on Hut 8's large AI-focused Texas data center campus.
The liquid cooling market is growing from $5.5 billion today to $15.75 billion by 2030, and Vertiv is the dominant provider in that market.
Vistra is the power thesis in its most direct form.
The $358 billion power segment is the critical path for the entire $7.6 trillion, and Vistra has spent the last 18 months locking up that critical path through long-term nuclear power purchase agreements.
Vistra secured a 20 year agreement with Meta for over 2,600 MW of nuclear energy, plus a separate deal with AWS from its Comanche nuclear facility.
Goldman Sachs and Jefferies both upgraded the stock after the Meta deal was announced.
The architecture of this trade is simple.
Goldman's model is not a prediction of whether AI spending happens but rather a model of the minimum physical capital required to deploy infrastructure that has already been contracted, already been announced, and is already under construction.
The compute layer requires the chips, data center layer requires cooling and power infrastructure and the power layer requires nuclear at scale on multi-decade contracts.
All three layers are being funded simultaneously, and all three have identifiable public companies sitting directly in the path of the capital.
The numbers are worth understanding precisely before talking about who benefits.
Goldman's baseline projects $765 billion in AI capital expenditure in 2026 alone, growing to $1.6 trillion annually by 2031.
Over the full 2026 to 2031 period, cumulative spend breaks down to $5.1 trillion in compute, $2.1 trillion in data centers, and $358 billion in power.
Nvidia is assumed to command 75% of all compute spend throughout the period, using the Rubin VR200 chip at $80,500 per GPU as the baseline.
The data center specification charts reveal how dramatically physical requirements are escalating.
A standard cloud data center runs 5–15 kW per rack while a transitional Blackwell era AI data center runs 130–200 kW per rack.
The AI factory of the future, running Rubin and Feynman silicon operates at 500+ kW per rack, at greater than 1 gigawatt scale, with liquid cooling only.
Traditional hyperscale data centers cost roughly $10 million per megawatt to build while the next generation AI data centers are being discussed at $15 to $20 million per megawatt.
Goldman identifies silicon useful life as the single biggest swing factor in the entire model.
At a 3-year replacement cycle, cumulative compute depreciation hits $3.99 trillion and at 7 years, it drops to $2.23 trillion, a $1.76 trillion difference on one assumption alone.
Power is only $358 billion of the total, but Goldman is explicit, it is the only component that can prevent the other 95% of the stack from deploying.
Amazon's Andy Jassy put it, "Our single biggest constraint is power." Connecting large data centers to the grid takes years.
Now here are the companies standing directly in the path of each layer of this capital.
Nvidia is still the most concentrated bet on the compute layer.
At 75% of $5.1 trillion in compute spend over six years, that is $3.8 trillion in cumulative revenue flowing through one company's products.
The 75% gross margin on data center GPUs is the reason every hyperscaler is trying to build custom silicon to escape it while simultaneously continuing to buy Nvidia because nothing else performs at the same level.
Vertiv is the direct infrastructure play on the data center upgrade cycle.
Every rack going from 40 kW to 500+ kW needs liquid cooling systems, power distribution, and thermal management infrastructure that simply did not exist at prior density levels.
Vertiv just deepened its liquid cooling capabilities through a strategic acquisition and was named a key partner on Hut 8's large AI-focused Texas data center campus.
The liquid cooling market is growing from $5.5 billion today to $15.75 billion by 2030, and Vertiv is the dominant provider in that market.
Vistra is the power thesis in its most direct form.
The $358 billion power segment is the critical path for the entire $7.6 trillion, and Vistra has spent the last 18 months locking up that critical path through long-term nuclear power purchase agreements.
Vistra secured a 20 year agreement with Meta for over 2,600 MW of nuclear energy, plus a separate deal with AWS from its Comanche nuclear facility.
Goldman Sachs and Jefferies both upgraded the stock after the Meta deal was announced.
The architecture of this trade is simple.
Goldman's model is not a prediction of whether AI spending happens but rather a model of the minimum physical capital required to deploy infrastructure that has already been contracted, already been announced, and is already under construction.
The compute layer requires the chips, data center layer requires cooling and power infrastructure and the power layer requires nuclear at scale on multi-decade contracts.
All three layers are being funded simultaneously, and all three have identifiable public companies sitting directly in the path of the capital.
✍1