Exabits retweeted
What is really happening in AI
1/ The real bottleneck in AI isn’t just GPUs.
It’s power, cooling, and data center architecture.
Hyperscalers own the entire stack.
Neo clouds mostly rent space + GPUs.
We’re building the layer in-between:
GPU-as-a-Service + vertically integrated power & cooling.
1/ The real bottleneck in AI isn’t just GPUs.
It’s power, cooling, and data center architecture.
Hyperscalers own the entire stack.
Neo clouds mostly rent space + GPUs.
We’re building the layer in-between:
GPU-as-a-Service + vertically integrated power & cooling.
2
3
11
2,809
Not all AI compute is urgent.
Instead of "compute as fast as possible," we're asking: How do we meet grid constraints while still delivering on performance SLAs?
Result? Power capping, intelligent workload shifting, and data centers acting as flexible reserves for the grid.
AI have to fight the grid- it can help stabilize it.
Instead of "compute as fast as possible," we're asking: How do we meet grid constraints while still delivering on performance SLAs?
Result? Power capping, intelligent workload shifting, and data centers acting as flexible reserves for the grid.
AI have to fight the grid- it can help stabilize it.
4
0
6
2,715
In the last 10 years, Moore's Law gave us approximately 100x compute. We scaled 1,000,000x.
We'll keep doing it. Energy efficiency directly drives factory revenues and token costs.
Hardware prices rise, but token generation gets so much faster- token costs drop an order of The future of AI isn't just more power. It's radically smarter power.
We'll keep doing it. Energy efficiency directly drives factory revenues and token costs.
Hardware prices rise, but token generation gets so much faster- token costs drop an order of The future of AI isn't just more power. It's radically smarter power.
2
1
9
1,918
Why do HVDC + Solid-State Transformers (SSTs) matter for next-gen AI?
Everyone focuses on GPUs, but as racks race toward 1MW with Blackwell & Vera Rubin, the real bottleneck is delivering massive, clean power efficiently. NVIDIA’s 800 VDC architecture boosts efficiency up to 5%, cuts maintenance 70%, and slashes TCO 30%. Fewer losses, less copper, lower cooling. Traditional AC worked for old data centers. AI needs HVDC for low-loss transmission + SSTs as smart precision regulators.
GPUs are the muscle. This is the circulatory system.
exascalelabs.ai/blog/why-hvdc-…
Everyone focuses on GPUs, but as racks race toward 1MW with Blackwell & Vera Rubin, the real bottleneck is delivering massive, clean power efficiently. NVIDIA’s 800 VDC architecture boosts efficiency up to 5%, cuts maintenance 70%, and slashes TCO 30%. Fewer losses, less copper, lower cooling. Traditional AC worked for old data centers. AI needs HVDC for low-loss transmission + SSTs as smart precision regulators.
GPUs are the muscle. This is the circulatory system.
exascalelabs.ai/blog/why-hvdc-…
1
1
7
1,293
F.03, the first humanoid robot to enter the White House alongside Melania Trump, just created history.
Serious AI compute powers advanced robotics from research to reality. Elite compute is needed to train, optimize, and operate these intelligent systems.
Exabits can power the oday. Visit exabits.ai to learn more.
Serious AI compute powers advanced robotics from research to reality. Elite compute is needed to train, optimize, and operate these intelligent systems.
Exabits can power the oday. Visit exabits.ai to learn more.
1
0
7
1,280
Power grid is overbuilt for rare peak days. 99% of the time it has massive excess capacity sitting idle. We should design data centers to flex: throttle down or shift load when the grid needs it most. Use the idle power instead of forcing endless new buildouts.
Why isn’t this happening?
Why isn’t this happening?
Here's my conversation with Jensen Huang, CEO of NVIDIA, the most valuable & one of the most influential companies in the history of human civilization. It is the engine powering the AI revolution.
This was a fascinating & inspiring conversation, in parts super-technical on engineere AI stack, memory, power, supply chain (TSMC, ASML, etc), in parts about leadership & psychology, and in parts personal & philosophical about life, consciousness, mortality, and human nature.
It's here on X in full and is up everywhere else (see comment).
Timestamps:
0:00 - Introduction
0:33 - Extreme co-design and rack-scale engineering
3:18 - How Jensen runs NVIDIA
22:40 - AI scaling laws
37:40 - Biggest blockers to AI scaling laws
39:23 - Supply chain
41:18 - Memory
47:24 - Power
52:43 - Elon and Colossus
56:11 - Jensen's approach to engineering and leadership
1:01:37 - China
1:09:50 - TSMC and Taiwan
1:15:04 - NVIDIA's moat
1:20:41 - AI data centers in space
1:24:30 - Will NVIDIA be worth $10 trillion?
1:34:39 - Leadership under pressure
1:48:25 - Video games
1:55:16 - AGI timeline
1:57:29 - Future of programming
2:11:01 - Consciousness
2:17:22 - Mortality
0
2
12
1,611
NVIDIA's Isaac GR00T & Cosmos models redefining robotics dexterity at GTC2026. Simulate, train, deploy faster with Exabits' global GPU marketplace. Accessible compute, lower costs, and no vendor lock-in.
Ready to build the next humanoid? Visit exabits.ai
Ready to build the next humanoid? Visit exabits.ai
1
0
7
835
Live from #GTC2026 in San Jose, Exabits Co-Founder Dr. Hoansoo Lee <@hoansoo>, CIO Jonathan Jaranilla <@JJcosino@JJcosinoO Mark Fidelman <@markfidelman> @markfidelmand witnessing the next AI era unfold- agentic systems, physical AI, massive Blackwell/Rubin scale, DLSS 5, robotics!
While td, we're already provisioning the GPUs to run it: enterprise-grade clusters ready now.
Witness → Build.
While td, we're already provisioning the GPUs to run it: enterprise-grade clusters ready now.
Witness → Build.
2
1
15
1,388
Where's the power for exploding AI data centers coming from?
âś…Solar/wind only half the time + batteries? Sketchy & pricey for 24/7 gigawatt loads.
âś…Small modular nuclear: clean, steady, reliable-seems like the winner.
@markfidelman asks- any OTHER clean solution (no gas/oil) thatDrop your thoughts 👇
At Exabits, our AIOps-optimized GPU clusters slash waste & boost efficiency- no new plant needed.
âś…Solar/wind only half the time + batteries? Sketchy & pricey for 24/7 gigawatt loads.
âś…Small modular nuclear: clean, steady, reliable-seems like the winner.
@markfidelman asks- any OTHER clean solution (no gas/oil) thatDrop your thoughts 👇
At Exabits, our AIOps-optimized GPU clusters slash waste & boost efficiency- no new plant needed.
0
1
15
918
Next week, don’t treat NVIDIA GTC (March 16-19) like “just another AI event.” That’s crazy.
Mark <@markfidelman>, Exabits CMO will be there- look him up!
Key themes already signaling: inference, agentic AI, physical AI (robotics), open models & AI factories.
Predictions:
1. Keynote wot’s about lowest-cost, massive-scale intelligence.
2. Jensen leans into economics: drive token costs down, boost throughput, make long-context reasoning profitable. (Rubin teases up to 10x inference cost reduction vs Blackwell 🔥)
3. Biggest announcements? Not just GPUs—networking, photonics, storage, power, memory. Bottlenecks shifting to data movement for agent reasoning at scale.
4. Expect heavier push on open models + physical AI. NVIDIA wants the full stack (models, software, tooling, robotics, deployment).
When everyone recaps the keynote, watch the real signals: who gets infra edge, factory advantage & cheap inference at scale. That’s where the money moves.
Mark <@markfidelman>, Exabits CMO will be there- look him up!
Key themes already signaling: inference, agentic AI, physical AI (robotics), open models & AI factories.
Predictions:
1. Keynote wot’s about lowest-cost, massive-scale intelligence.
2. Jensen leans into economics: drive token costs down, boost throughput, make long-context reasoning profitable. (Rubin teases up to 10x inference cost reduction vs Blackwell 🔥)
3. Biggest announcements? Not just GPUs—networking, photonics, storage, power, memory. Bottlenecks shifting to data movement for agent reasoning at scale.
4. Expect heavier push on open models + physical AI. NVIDIA wants the full stack (models, software, tooling, robotics, deployment).
When everyone recaps the keynote, watch the real signals: who gets infra edge, factory advantage & cheap inference at scale. That’s where the money moves.
0
1
15
1,107
George sums it up perfectly: Organizations remain stuck until the entire factory is redesigned for Institutional Intelligence, but individual AI is like switching from steam to electric motors- huge personal gains.
However, without massive, dependable compute at the base layer, none of that redesign is possible.
Exabits: the mothership driving tomorrow's agents, coordination, and spontaneous systems, just fuel for genuine institutional upside- no slop.
However, without massive, dependable compute at the base layer, none of that redesign is possible.
Exabits: the mothership driving tomorrow's agents, coordination, and spontaneous systems, just fuel for genuine institutional upside- no slop.
2
1
7
975
Big Tech clouds charge more because they're racing to build the next ChatGPT themselves.
Exabits? We're just in the business of selling compute- no distractions, no self-dealing. We're different: pure compute providers. No internal AI races. No scarcity games. Just high-end power at fair prices.
Exabits? We're just in the business of selling compute- no distractions, no self-dealing. We're different: pure compute providers. No internal AI races. No scarcity games. Just high-end power at fair prices.
1
1
10
1,015
Kevin O'Leary on AI right now: If he were 25 today, he'd go ALL IN on implementation/execution for small businesses desperate to adopt AI (not just consulting) + data center development. Demand is insane- only 5GW under construction vs 30GW needed!
At Exabits, we're building the centers. The road is tough, needs capital, but it's the niche that wins big. Who's ready? Hear it from @markfidelman, Exabits' CMO.
At Exabits, we're building the centers. The road is tough, needs capital, but it's the niche that wins big. Who's ready? Hear it from @markfidelman, Exabits' CMO.
1
2
14
1,529
Exabits retweeted
@markfidelman alpha dropped: @exa_bits going public on NASDAQ in 3-6 months. service + customer advantage over AWS/Google/Microsoft. "good luck especially if you're smaller org" disrespectfully dominated. IPO szn. bussin.
1
2
5
736
Minimize infrastructure risk with Exabits:
âś… Distributed across multiple enterprise AI data centers worldwide- no single point of failure
âś…Built-in global redundancy for consistent reliability âś… High uptime supports uninterrupted training and inference
Deploy with ely.
âś… Distributed across multiple enterprise AI data centers worldwide- no single point of failure
âś…Built-in global redundancy for consistent reliability âś… High uptime supports uninterrupted training and inference
Deploy with ely.
0
1
12
620
Exabits retweeted
🧠Decentralized AI Model Training panel at BASS Denver 2026
@benfielding @gensynai @markfidelman @exa_bits @shayebackus @castorhat @PrismaXai @jonathankingvc @cbventures
youtube.com/watch?v=2btTir…
the highlights 👇
@benfielding @gensynai @markfidelman @exa_bits @shayebackus @castorhat @PrismaXai @jonathankingvc @cbventures
youtube.com/watch?v=2btTir…
the highlights 👇
1
2
8
8,094