Sovereign cloud is pushing compute back toward localized, controlled environments.
That means less uniform infrastructure and far more variability in how software performs.
The interesting question is what happens when optimization has to adapt to every environment, not just one?
New Category: SelfAware Compute. Software that understands how it runs and optimizes itself.
New category:
SelfAware Compute
Software that understands how it runs and optimizes itself accordingly.
#NewCategory
Everyone talks like AI limits are just compute. They’re not.
Energy. bandwidth. memory. physics. It’s how these resources get extracted, built and deployed.
Then it’s how we adapt software around those constraints… or work around them entirely.
youtube.com/shorts/wvzJH...
The next wave of compute isn’t just more infrastructure.
It’s smarter execution on top of it.
That’s where the real gains are.
#Infrastructure
In the future, software won’t just run.
It will:
• observe
• predict
• adapt
Every execution becomes intentional.
#AI #Systems
This pattern shows up a lot. “Just one more year” has a way of piling up over time.
At some point it stops being about the tools and starts being about how hard the system is to actually change.
Billions in data center projects are being delayed or blocked as communities push back on energy, cost, and impact.
When those same constraints keep showing up, it starts to feel like infrastructure alone isn’t the whole answer.
Probably means the software layer has to change too.
The future is coming... Code that understands itself.
We believe the future of software is:
Not faster code
Not bigger models
But code that understands itself.
#Future
We're building the engine to make AI run more efficiently - from energy to apps.
#SelfAwareCompute #SustainableComputing
#ComputeEfficiency
youtube.com/shorts/SxoTO...
Stop throwing hardware at inefficient execution.
#HotTake
The Paradigm: Optimization isn't static. Execution is dynamic. selfawarecompute.com
Parallelization was about doing more at once.
SelfAware Compute is about doing only what’s necessary.
That’s a very different idea.
#Software
This isn’t:
• a new programming language
• a new chip
• a new framework
Self-Aware Compute works on existing systems.
It changes how software executes.
#Tech
We’ve been scaling hardware for decades.
More cores
More GPUs
More power
What if instead…
we made software smarter about how it runs?
#FutureOfCompute
Compute cost = product viability.
Get it right → scalable margins
Get it wrong → lose money per user
Execution efficiency is now a business problem.
#Startups #AI
New starter pack: Energy + Compute Efficiency
Compute is now constrained by energy.
And inefficiency shows up as cost.
These are the people thinking about how systems actually scale under real-world limits.
go.bsky.app/Ui89jmu
The Economics: Every inefficiency = cost. Execution determines outcomes. selfawarecompute.com
AI isn’t just a model problem.
It’s an energy problem.
Inference is now the majority of compute—and it runs constantly.
Every inefficiency = real cost.
#AI #Energy
Software runs blindly.
Execution isn’t static.
That’s about to change.
#SelfAwareCompute
The Reality: Same Code. Same Hardware. Completely different execution. selfawarecompute.com
Same code. Same hardware. Different inputs.
→ Completely different execution behavior
But today, software treats them the same.
That’s inefficiency at scale.
#AI #Systems
Most optimization today is static.
Compilers optimize once.
Then execution is fixed.
Self-Aware Compute is dynamic.
It adapts every time software runs.
#Engineering
Self-Aware Compute means software can:
• understand execution pathways
• predict time + memory before running
• optimize based on goals (energy, speed, cost)
Not guess. Not approximate. Understand.
#Tech
Fascinating - “The companies doing everything ‘right’… came in second.”
There’s a balance: core functionality has to work end-to-end but optimizing everything too early slows you down.
Thanks for sharing your research.
Wild that we can:
• train trillion-parameter models
• simulate physics
• generate video
…but can’t answer,
“How long will this code take to run?”
#DeveloperThoughts
Linocut print ‘Feynman Diagram Bauhaus’ is a hand-carved and printed variable edition on cream-coloured Japanese washi paper, 8" by 10" with a jumble of Feynman diagrams and coloured shapes with a Bauhaus feel. Shapes include circles, rings, triangles, stripes, semicircles and irregular semicircles under wiggly lines. Colours are tints of fuchsia, blue-green, blue, yellow and purple with Feynman diagrams in black lines.
April 14 has been designated World Quantum Day in honour of Planck’s Constant which can be rounded to h~ 4.14×10−15 eV·s (and some folks write April 14 as 4/14*). 🧪🐡🎢 Planck’s constant comes up a lot in quantum mechanics; for instance a photon’s energy is h times its frequency). 🧵
🤣
The Constraint - Compute is now limited by energy. Execution is the missing layer. selfawarecompute.com
Data centers used ~460 TWh of electricity last year.
And it’s accelerating fast.
We don’t have a hardware problem.
We have an execution problem.
#Energy #AI
Who else should be on this?
Looking for people working on:
• Execution and runtime behavior
• Performance and systems
• Energy and efficiency in compute
Drop names 👇
Here’s a starter pack for people thinking about this shift:
go.bsky.app/5gP1ifo
These are people focused on how software actually runs, not just how it’s written.
Future of software execution starter pack:
• Execution ≠ static
• Inputs change everything
• Efficiency is the constraint
• Energy = cost
• Runtime is where it matters
Most software still runs blind.
That’s about to change.
#SelfAwareCompute
Software today is “compile and pray.”
You compile once
→ assume performance
→ hope it works across all inputs
But:
Inputs change
Systems change
Workloads change
Execution doesn’t adapt.
#SoftwareEngineering
Compute is becoming the most constrained resource in the world.
Not storage. Not bandwidth.
Compute.
And we’re still running software blindly.
That’s the problem.
#AI #Infrastructure
