Matthew Collins
@matthewcollins.bsky.social
Archaeology, University of Cambridge | The Globe, University of Copenhagen
Ancient Proteins | Medieval Manuscripts | Proteomics and AI | 🇺🇦
Ancient Proteins | Medieval Manuscripts | Proteomics and AI | 🇺🇦
9/ The Lesson for the C-Suite: Archaeology teaches us that complexity is a subsidy of stability. If you build your entire strategy on a "Palatial" supply chain (Nvidia/TSMC), you are vulnerable to a 1200 BCE-style collapse.
February 15, 2026 at 8:35 AM
9/ The Lesson for the C-Suite: Archaeology teaches us that complexity is a subsidy of stability. If you build your entire strategy on a "Palatial" supply chain (Nvidia/TSMC), you are vulnerable to a 1200 BCE-style collapse.
8/ Innovation through Constraint.
Will we are seeing the same "accidental" toughening. Because we can't all afford $40k H100s, developers outside USA / China are forced to use "charcoal" methods:
A more resiliant 'steel AI' would impact markets and change workpatterns
Will we are seeing the same "accidental" toughening. Because we can't all afford $40k H100s, developers outside USA / China are forced to use "charcoal" methods:
A more resiliant 'steel AI' would impact markets and change workpatterns
February 15, 2026 at 8:35 AM
8/ Innovation through Constraint.
Will we are seeing the same "accidental" toughening. Because we can't all afford $40k H100s, developers outside USA / China are forced to use "charcoal" methods:
A more resiliant 'steel AI' would impact markets and change workpatterns
Will we are seeing the same "accidental" toughening. Because we can't all afford $40k H100s, developers outside USA / China are forced to use "charcoal" methods:
A more resiliant 'steel AI' would impact markets and change workpatterns
7/ The "Steel" of AI is a happy accident. Steel wasn't a deliberate invention—it was a byproduct of struggle. Smiths couldn't melt iron like bronze; they had to work it in charcoal fires. That charcoal "accidentally" introduced carbon, making the metal tougher and lighter than the "elite" bronze.
February 15, 2026 at 8:35 AM
7/ The "Steel" of AI is a happy accident. Steel wasn't a deliberate invention—it was a byproduct of struggle. Smiths couldn't melt iron like bronze; they had to work it in charcoal fires. That charcoal "accidentally" introduced carbon, making the metal tougher and lighter than the "elite" bronze.
6/ The "Steel" of AI.
Bronze was brittle. Iron (and eventually steel) was resilient. The "Steel" of AI isn't a massive, fragile LLM in a single data center—it’s a million small, locally-tuned models running on "gaming PCs."
Resilience beats Scale when the environment gets harsh.
Bronze was brittle. Iron (and eventually steel) was resilient. The "Steel" of AI isn't a massive, fragile LLM in a single data center—it’s a million small, locally-tuned models running on "gaming PCs."
Resilience beats Scale when the environment gets harsh.
February 15, 2026 at 8:35 AM
6/ The "Steel" of AI.
Bronze was brittle. Iron (and eventually steel) was resilient. The "Steel" of AI isn't a massive, fragile LLM in a single data center—it’s a million small, locally-tuned models running on "gaming PCs."
Resilience beats Scale when the environment gets harsh.
Bronze was brittle. Iron (and eventually steel) was resilient. The "Steel" of AI isn't a massive, fragile LLM in a single data center—it’s a million small, locally-tuned models running on "gaming PCs."
Resilience beats Scale when the environment gets harsh.
5/ Small Language Models (SLMs) are the new Iron.
If global trade or energy crises make 100,000-GPU clusters unsustainable, we move to the "Artisanal" era.
Quantisation: Making models smaller/faster.
LoRA: Efficient fine-tuning.
Edge Computing: AI running on ARM, Gaming PC not server farms.
If global trade or energy crises make 100,000-GPU clusters unsustainable, we move to the "Artisanal" era.
Quantisation: Making models smaller/faster.
LoRA: Efficient fine-tuning.
Edge Computing: AI running on ARM, Gaming PC not server farms.
February 15, 2026 at 8:35 AM
5/ Small Language Models (SLMs) are the new Iron.
If global trade or energy crises make 100,000-GPU clusters unsustainable, we move to the "Artisanal" era.
Quantisation: Making models smaller/faster.
LoRA: Efficient fine-tuning.
Edge Computing: AI running on ARM, Gaming PC not server farms.
If global trade or energy crises make 100,000-GPU clusters unsustainable, we move to the "Artisanal" era.
Quantisation: Making models smaller/faster.
LoRA: Efficient fine-tuning.
Edge Computing: AI running on ARM, Gaming PC not server farms.
4/ The "Iron Age" Pivot.
When the Bronze Age collapsed, the "Democratic" technology of Iron took over. Why? Because iron ore is everywhere. It didn't need a 2,000-mile trade route; it just needed a local blacksmith with the right "recipe."
When the Bronze Age collapsed, the "Democratic" technology of Iron took over. Why? Because iron ore is everywhere. It didn't need a 2,000-mile trade route; it just needed a local blacksmith with the right "recipe."
February 15, 2026 at 8:35 AM
4/ The "Iron Age" Pivot.
When the Bronze Age collapsed, the "Democratic" technology of Iron took over. Why? Because iron ore is everywhere. It didn't need a 2,000-mile trade route; it just needed a local blacksmith with the right "recipe."
When the Bronze Age collapsed, the "Democratic" technology of Iron took over. Why? Because iron ore is everywhere. It didn't need a 2,000-mile trade route; it just needed a local blacksmith with the right "recipe."
3/ Palatial vs. Artisanal Economies.
Bronze Age empires were "Palatial"—highly centralized, elite-controlled, and fragile.
We are currently in the Palatial AI era: Power is concentrated in a few "Hyperscalers" (Big Tech/Foundries). Only "Palaces" (the data center) have the tech.
Bronze Age empires were "Palatial"—highly centralized, elite-controlled, and fragile.
We are currently in the Palatial AI era: Power is concentrated in a few "Hyperscalers" (Big Tech/Foundries). Only "Palaces" (the data center) have the tech.
February 15, 2026 at 8:35 AM
3/ Palatial vs. Artisanal Economies.
Bronze Age empires were "Palatial"—highly centralized, elite-controlled, and fragile.
We are currently in the Palatial AI era: Power is concentrated in a few "Hyperscalers" (Big Tech/Foundries). Only "Palaces" (the data center) have the tech.
Bronze Age empires were "Palatial"—highly centralized, elite-controlled, and fragile.
We are currently in the Palatial AI era: Power is concentrated in a few "Hyperscalers" (Big Tech/Foundries). Only "Palaces" (the data center) have the tech.
2/ The Dependency Trap
Bronze wasn't a "local" tech. It required Copper (common) + Tin (rare). Tin traveled 1000s of km to reach the Mediterranean.
Today’s "Bronze" is Compute. High-end AI requires specialized GPUs + HBM memory. If that hyper-centralized supply chain snaps, the "empire" halts.
Bronze wasn't a "local" tech. It required Copper (common) + Tin (rare). Tin traveled 1000s of km to reach the Mediterranean.
Today’s "Bronze" is Compute. High-end AI requires specialized GPUs + HBM memory. If that hyper-centralized supply chain snaps, the "empire" halts.
February 15, 2026 at 8:35 AM
2/ The Dependency Trap
Bronze wasn't a "local" tech. It required Copper (common) + Tin (rare). Tin traveled 1000s of km to reach the Mediterranean.
Today’s "Bronze" is Compute. High-end AI requires specialized GPUs + HBM memory. If that hyper-centralized supply chain snaps, the "empire" halts.
Bronze wasn't a "local" tech. It required Copper (common) + Tin (rare). Tin traveled 1000s of km to reach the Mediterranean.
Today’s "Bronze" is Compute. High-end AI requires specialized GPUs + HBM memory. If that hyper-centralized supply chain snaps, the "empire" halts.