EXPONENTIALS
The data behind accelerating change
Humans think linearly, but technology advances exponentially. These charts track the trends that underpin the Singularity thesis, from compute and AI to biology, energy, and data. Most of these curves have held for decades and show no sign of slowing.
1026
FLOP (2025)
5T
Parameters
336B
Transistors
$200
Genome Cost
MOORES
RAY KURZWEILMoore's Law
Transistors per microprocessor doubling approximately every 2 years
TAKEAWAY: Hardware capability keeps compounding over multiple architecture eras.
RANGE: 1971-2026UNIT: Transistors per chipSOURCE: Kurzweil trend datasets, public model/system disclosuresUPDATED: 2026-02-06
PRICE-PERF
RAY KURZWEILPrice-Performance
Calculations per second per $1,000 - exponential improvement over decades
TAKEAWAY: The same budget purchases radically more compute every decade.
RANGE: 1956-2050UNIT: MFLOPS per $1,000SOURCE: Kurzweil trend datasets, public model/system disclosuresUPDATED: 2026-02-06
GENOME
RAY KURZWEILGenome Sequencing Cost
Cost to sequence a human genome falling faster than Moore's Law
TAKEAWAY: Biotech cost curves are approaching software-like economics.
RANGE: 2001-2030UNIT: USD per genomeSOURCE: NIH historical sequencing cost trend + projectionsUPDATED: 2026-02-06
SOLAR
RAY KURZWEILSolar Power Cost
Cost per watt of solar photovoltaic panels - exponential decline
TAKEAWAY: Energy abundance is increasingly driven by relentless cost compression.
RANGE: 1976-2030UNIT: USD per wattSOURCE: Historical PV module pricing + projectionsUPDATED: 2026-02-06
STORAGE
RAY KURZWEILStorage Cost
Cost per gigabyte of digital storage - dramatic decline over decades
TAKEAWAY: Cheap memory makes richer models and data-heavy products economically viable.
RANGE: 1980-2032UNIT: USD per GBSOURCE: Storage industry historical pricing + projectionsUPDATED: 2026-02-06
WIRELESS
RAY KURZWEILWireless Data Rates
Peak mobile network speeds from 2G to 6G
TAKEAWAY: Network throughput improvements unlock continuously richer AI interfaces.
RANGE: 1991-2032UNIT: Mbps/GbpsSOURCE: Mobile network standard peak throughput dataUPDATED: 2026-02-06
DATA
RAY KURZWEILGlobal Data Created
Zettabytes of data created per year - exponential growth
TAKEAWAY: Model training supply keeps rising as digital exhaust compounds.
RANGE: 2010-2030UNIT: Zettabytes/yearSOURCE: IDC-style global data creation trend + projectionsUPDATED: 2026-02-06
ROBOTS
RAY KURZWEILIndustrial Robot Population
Millions of operational industrial robots worldwide
TAKEAWAY: Automation capacity is expanding alongside AI capabilities.
RANGE: 2000-2030UNIT: Millions of robotsSOURCE: IFR-style installed robot base trend + projectionsUPDATED: 2026-02-06
SINGULARITY
RAY KURZWEILCountdown to Singularity
Kurzweil's chart showing accelerating paradigm shifts throughout history
TAKEAWAY: Major civilizational transitions arrive at progressively shorter intervals.
RANGE: 10B years ago to 2045UNIT: Years between paradigm shiftsSOURCE: Kurzweil milestone framingUPDATED: 2026-02-06
COMPUTE
Training Compute (FLOP)
Exponential growth in computational resources used to train AI models
TAKEAWAY: Frontier training runs keep scaling despite rising absolute cost.
RANGE: 2012-2028UNIT: FLOP per training runSOURCE: Kurzweil trend datasets, public model/system disclosuresUPDATED: 2026-02-06
PARAMS
Model Parameters (Billions)
The number of learnable parameters in frontier models
TAKEAWAY: Model capacity continues climbing, with inference efficiency improving in parallel.
RANGE: 2012-2028UNIT: Billions of parametersSOURCE: Kurzweil trend datasets, public model/system disclosuresUPDATED: 2026-02-06
INFERENCE
Inference Cost
Cost per million tokens for frontier model APIs - dramatic decline
TAKEAWAY: Cheaper inference expands viable use cases faster than capability alone.
RANGE: 2020-2027UNIT: USD per 1M tokensSOURCE: Public API pricing snapshots + projectionsUPDATED: 2026-02-06
GPU
GPU Performance
NVIDIA data center GPU TFLOPS (FP16) - exponential scaling
TAKEAWAY: Hardware throughput gains are a key multiplier on model progress.
RANGE: 2016-2028UNIT: Peak TFLOPS (FP16/BF16)SOURCE: NVIDIA data center GPU disclosures + projectionsUPDATED: 2026-02-06
BENCHMARK
Benchmark Performance
AI performance on standard reasoning and capability benchmarks
TAKEAWAY: Frontier systems are converging toward or beyond human baselines on many tasks.
RANGE: 2018-2027UNIT: Composite benchmark score (%)SOURCE: Benchmark aggregation from model reportsUPDATED: 2026-02-06
HISTORICAL DATA
PROJECTED
FUTURE ZONE
"
"
An analysis of the history of technology shows that technological change is exponential, contrary to the common-sense 'intuitive linear' view. So we won't experience 100 years of progress in the 21st century. It will be more like 20,000 years of progress.RAY KURZWEILTHE SINGULARITY IS NEAR (2005)