Timeline · 2006 → 2030
How we got here.
Every milestone that shaped the AI data center industry, from Amazon's first EC2 instance to Meta's Hyperion coming online. Announcements, groundbreakings, silicon drops, nuclear deals — all on one spine.
2006
Infrastructure
AWS EC2 launches
Public cloud compute is born. First EC2 instance = 1 vCPU. Foundational — every modern AI lab grows on top of this.
2009
Infrastructure
Facebook Open Compute Project begins (internal)
Facebook designs its own servers + opens the specs in 2011, forcing server industry toward vendor-neutral, efficient designs.
2012
Silicon
AlexNet wins ImageNet on 2 GPUs
Two NVIDIA GTX 580s beat CPU-based approaches by ~10 percentage points. The moment GPU + deep learning fuse.
2016
Silicon
Google TPU v1 in production
First custom AI ASIC at hyperscale. Powers Google Search ranking + Translate.
2017
Silicon
Attention Is All You Need — Transformer paper
Architecture that every modern LLM uses, published by Google researchers. Changes what compute workloads look like.
2020-05
Silicon
GPT-3 released — 175B params
First publicly-visible sign that scale alone produces capability jumps. Training cost estimated at $4-12M (cheap in retrospect).
2022-03
Silicon
NVIDIA H100 announced
3.35 TB/s memory bandwidth, 700W TDP. Becomes the workhorse of the 2023-2024 training era.
2022-11
Policy & Energy
ChatGPT launches
100M users in 2 months. Triggers the AI infrastructure arms race everyone is still in.
2023-06
Silicon
AMD Instinct MI300X announced
First credible non-NVIDIA competitor at hyperscale. 192 GB HBM3. Microsoft, Meta, Oracle place orders.
2023-08
Finance
CoreWeave $2.3B debt raise
Debt-financed GPUs at massive scale. Proves the neocloud model — GPUs as bond collateral.
2023-11
Policy & Energy
Google + Fervo geothermal PPA signed
First large hyperscaler geothermal deal. Fervo enhanced geothermal in Nevada.
2024-03
Silicon
NVIDIA Blackwell (B200) announced
Dual-die GPU, 192 GB HBM3e, 1000W. First rack-scale system (NVL72) with 72 GPUs in one NVLink domain.
2024-03
Policy & Energy
AWS + Talen Energy $650M deallive
AWS acquires Cumulus data centers adjacent to Susquehanna nuclear plant — behind-the-meter nuclear power.
2024-07
Infrastructure
xAI Colossus 1 online — 100K H100 in 122 dayslive
Fastest gigawatt-class build ever. Broke records on site prep, substation build, and rack install. Powered partly by on-site gas turbines.
Read the case study →2024-09
Policy & Energy
Microsoft + Three Mile Island dealplanned
20-year PPA to restart TMI Unit 1 reactor. Targeted online 2028. First nuclear restart explicitly for AI load.
2024-09
Infrastructure
Stargate Abilene — construction beginsbuilding
Crusoe breaks ground on the 1,000-acre Lancium Clean Campus. First 2 buildings live by Sept 2025.
2024-10
Policy & Energy
Google + Kairos Power SMR dealplanned
First small modular reactor deal at hyperscale. Target: ~500 MW by 2030 from Kairos SMRs.
2024-11
Silicon
NVIDIA H200 shipslive
Drop-in H100 replacement with 141 GB HBM3e (vs 80 GB). Same TDP, 43% more memory bandwidth.
2024-12
Infrastructure
Meta Hyperion Louisiana announced (originally $10B)building
Richland Parish 2,250-acre campus announced. Entergy Louisiana builds 3 gas plants. Becomes 'the Hyperion problem' for state ratepayers.
Read the case study →2025-01
Infrastructure
Stargate Project announced on national TV
Trump White House press conference. $500B by 2029. 10 GW target. Biggest AI infra commitment in history.
Read the case study →2025-03
Finance
CoreWeave IPO at $23B
Raises $1.5B at $23B valuation. Validates the neocloud model at public-markets scale.
2025-06
Silicon
NVIDIA acquires Run.AI for $700M
GPU fractioning + scheduling for K8s. Tightens the NVIDIA vertical stack.
2025-09
Infrastructure
Stargate Abilene Phase 1 livelive
First 2 of 8 buildings operational on Oracle Cloud Infrastructure. 980,000 sq ft.
2025-10
Infrastructure
Project Rainier activated — 500K Trainium2live
World's largest non-NVIDIA AI cluster. 7 of 30 buildings online near New Carlisle, Indiana. Built in 7 months from groundbreaking.
Read the case study →2025-11
Silicon
Anthropic announces Google TPU deal
Multi-cloud AI factory — Anthropic commits to up to 1M TPUs + 1+ GW of new Google-operated compute.
2026-01
Infrastructure
xAI Colossus 2 online (Southaven MS)live
Second campus with ~27 gas turbines. Targeting combined 2 GW across Memphis + Southaven by end-2026.
2026-01
Policy & Energy
EPA closes 'portable turbines' loophole
New Source Performance Standard revised. Gas turbines require permits even for temporary operations. Directly responds to Colossus controversy.
2026-03
Infrastructure
Stargate Abilene capped at 1.2 GW
Grid constraints prevent 2 GW expansion. Microsoft leases adjacent 900 MW at Lancium. Neighbors, different strategies.
Read the case study →2026-03
Infrastructure
Stargate adds 5 new sites (~7 GW total)
Five new US sites announced — $400B+ of the $500B 10 GW commitment. Pivot to NVIDIA's Vera Rubin platform for new capacity.
2026-04
Policy & Energy
Entergy LA approves 7 more Meta gas plants
Louisiana PSC greenlights expansion. 10 gas plants total for Hyperion. Louisiana Energy Users Group protests.
Read the case study →2027
Policy & Energy
xAI gas turbine permits expire (Memphis)planned
Plan: replaced by two MLGW substations + 500-acre solar farm + expanded Tesla Megapack storage.
2028
Policy & Energy
Microsoft-TMI Unit 1 target onlineplanned
Three Mile Island restart completes. First grid-tied nuclear specifically contracted to AI loads.
2029
Infrastructure
Stargate $500B commitment full deploymentplanned
Public target for completion of $500B commitment across all Stargate sites.
2030
Infrastructure
Meta Hyperion fully operationalplanned
Richland Parish campus at full design capacity. First nuclear MOU due for development.
2030
Policy & Energy
Microsoft net-zero deadlineplanned
Self-imposed carbon-negative + water-positive + zero-waste target. 2024 report showed emissions UP since 2020 due to AI.
Want to go deeper?
See the full case studies of the named projects, or read Lesson 11 · Economics for the finance mechanics behind the 2024–2027 capex supercycle.