“Assembled. Component by component.”
Harpreet Singh Bedi · Rocket Singh: Salesman of the Year, 2009

Rocket Singh Computing Index

The changing economics of building a computer.

Reconstructed Quote unverified

Film: Rocket Singh: Salesman of the Year (2009)
Base year: 2009
Indexed unit: One business machine, assembled from ten constituents
Geography: India
Currency: INR

One computer. Ten constituents. Seventeen years of technological disruption.

Value at 2009

₹38,000

Value at last verified date (2025)

₹85,000

Rocket Singh removed the corporate markup. AI moved it into memory and silicon.

Nominal change: +123.7%
Compound annual growth: +5.2% a year
2009 value in 2025 money: ₹1,00,149
Real appreciation: −15.1%
Months of income, 2009: 9.9 months
Months of income, 2025: 5.0 months
Latest observation: Provisional

Base year — Release year of Rocket Singh, and the year the index is based at 100. The film never prices a machine, so the base value is the 2009 business build reconstructed constituent by constituent.

01Index overview

The machine as an index

Ten constituents, weighted at their 2009 shares. Every figure below is computed from the constituent series — none of it is typed in, so repricing one part moves the whole page.

Index, 2025: 223.7; 2009 = 100
Total change: +123.7%; Nominal rupees
Inflation adjusted: −15.1%; Net of Indian CPI
Capability index: 2,214; 22.1× the 2009 machine
Price per capability: −89.9%; Bottomed in 2024
Largest constituent: Processor; By share of the 2025 business build
Fastest riser: Cooling; By percentage, and from the smallest base
Largest fall: Graphics; Absorbed into the processor, not discounted

The index level is derived, not published. It has not been validated against an independent historical dataset, and the confidence label on each constituent is the honest read on how much weight any one of these figures can carry.

02The series

What it was worth, each year

Nominal local currency by default. Switch modes to strip out inflation, express the figure as earning time, or read it as quantity and percentage change.

Event markers

Value in the currency of the day, unadjusted.

Focus the chart and use ← → to walk the series, Home and End for the ends, Escape to clear.

Business machine, IndiaInterpolatedProvisional

One business machine, assembled from ten constituents — Business machine, India: ₹38,000 in 2009, ₹85,000 in 2025 — an increase of 123.7% across 17 annual observations.

Did it outrun the benchmarks?

India CPI, all groupsLagged
2.2× vs 2.6×
The only benchmark that shows what actually happened. The price barely moved. Everything around it got more expensive.
Per-capita net national incomeLagged
2.2× vs 4.5×
How many weeks of earnings a machine costs now against then.

03Two machines

Same job, and what people now mean by compute

Two reconstructions, priced at the snapshot. The first does the work Rocket Singh's customers actually did. The second is the machine the phrase has come to describe.

Business

Complete machine, including display and peripherals.

Current market

₹85,000

The modern equivalent of the machine Rocket Singh sold to a small office. Documents, spreadsheets, browser tabs, video calls.

Processor₹20,000
Six-core APU with integrated graphics23.5%
Motherboard₹11,000
Entry current-socket, DDR5, Wi-Fi 612.9%
Memory₹12,000
16 GB DDR5-5600, two modules14.1%
The single most volatile line on this page.
Storage₹6,500
1 TB NVMe Gen47.6%
GraphicsIncluded
Integrated in the processor
No separate purchase. In 2009 this tier bought a ₹3,200 card.
Power supply₹5,500
550 W, 80 Plus Bronze6.5%
Cooling₹2,500
Stock cooler, three case fans2.9%
Cabinet₹4,500
Steel mid-tower, mesh front5.3%
Display₹15,000
27-inch IPS, 2560 × 144017.6%
Networking and peripherals₹8,000
Wi-Fi 6, keyboard, mouse, webcam9.4%

Largest constituent: Processor at ₹20,000 — 23.5% of the build, 32.3% of the tower.

Compute

Complete machine, including display and peripherals.

Current market

₹1,96,000

Rendering, local model work and anything that treats the graphics card as the processor. The machine most people now mean when they say compute.

Processor₹20,500
Six-core, no integrated graphics10.5%
Motherboard₹13,000
B650-class, DDR5, PCIe 4.06.6%
Memory₹34,000
32 GB DDR5-600017.3%
Twice the capacity of the business tier at nearly three times the price.
Storage₹15,000
1 TB NVMe Gen4, high endurance7.7%
Graphics₹80,000
16 GB discrete accelerator40.8%
The largest single constituent, by a wide margin.
Power supply₹7,500
750 W, 80 Plus Gold3.8%
Cooling₹3,500
Large air tower cooler1.8%
Cabinet₹6,000
Mid-tower, high airflow3.1%
Display₹12,500
27-inch IPS, 2560 × 14406.4%
Networking and peripherals₹4,000
Wi-Fi 6, keyboard, mouse2.0%

Largest constituent: Graphics at ₹80,000 — 40.8% of the build, 44.6% of the tower.

The first machine does the same job. The second is what people increasingly mean when they say compute — and the difference between them is almost entirely two constituents.

Where the money sits, by tier

Shares are computed from the prices, so they move whenever a constituent is repriced. Nothing here is a fixed weight.

Business₹85,000

Largest: Processor, 23.5% of the build.

Performance₹1,16,000

Largest: Graphics, 29.3% of the build.

Compute₹1,96,000

Largest: Graphics, 40.8% of the build.

Processor
Motherboard
Memory
Storage
Graphics
Power supply
Cooling
Cabinet
Display
Networking and peripherals

04Constituents

Ten markets in a steel box

Each constituent has its own price history, its own capability history and its own set of markets that decide it. Open one to see what set the price.

Processor

Does the work. Sets what the rest of the platform has to be.

₹8,500₹20,000+135.3%index 235

Dual core, 2.8 GHz, 65 W, 45 nm Six cores, twelve threads, integrated graphics, 4 nm

22.4% of the 2009 build 23.5% of the 2025

What the money bought

Processor: 1,100 Multi-thread score 27,000 Multi-thread score24.5×

What sets this price

Foundry process transitions and the cost of each new node
Core-count competition between the two x86 vendors
Integrated graphics absorbing a separate component
Rupee depreciation on an entirely imported part

Processor throughput rose about twenty-five times while the price rose about two and a third. Per unit of work it is the strongest deflation story in the machine — and it is no longer where the money goes.

The processor became faster. The expensive part moved elsewhere.

Reconstructed

Motherboard

Connects everything, and decides what everything else can be.

₹4,300₹11,000+155.8%index 256

Entry chipset, 2 DIMM slots, PCIe 2.0, 100 Mbit ethernet Current-socket B-series, DDR5, PCIe 4.0, Wi-Fi 6, 2.5 Gbit ethernet

11.3% of the 2009 build 12.9% of the 2025

What sets this price

Power-delivery requirements rising with processor draw
Each PCIe and memory generation adding trace and validation cost
Networking and storage controllers moving onto the board
Shorter socket lifespans forcing more frequent platform purchases

The motherboard is the quiet riser: up more than two and a half times, and almost none of it visible to the buyer as a feature. It is also the constituent that locks in the cost of the next two — a board decides which processor and which memory generation you are allowed to buy.

The motherboard connects everything, which is also why everything can make it more expensive.

Reconstructed

Memory

Holds what the processor is working on right now.

₹2,100₹12,000+471.4%index 571

2 GB DDR2-800, 12.8 GB/s 16 GB DDR5-5600, 89.6 GB/s

5.5% of the 2009 build 14.1% of the 2025

Priced by capacity — ₹ per GB

₹1,050₹750−28.6%2 GB → 16 GB installed

Cheapest, 2023: ₹184
Fall to 2023: −82.4%
Since 2023: +306.8%

What a business build shipped with: 2 GB in 2009, 4 GB from 2011, 8 GB from 2015, 16 GB from 2021. Capacity moves in steps, so price per gigabyte drops sharply in a step year for reasons that have nothing to do with the memory market.

What the money bought

Memory capacity: 2 GB 16 GB8.0×
Memory bandwidth: 13 GB/s 90 GB/s7.0×

What sets this price

The DRAM production cycle, which has always overshot in both directions
High-bandwidth memory for AI accelerators competing for the same wafers
Server and hyperscale contracts taking priority allocation
Manufacturers shifting capacity toward higher-margin enterprise parts

Memory fell about eighty per cent per gigabyte between 2009 and 2023, which is roughly what the industry promises. Then it reversed harder than at any point in the series. The reason is not a shortage of factories. It is that personal computers stopped being the most persuasive customer in the room.

In 2009, memory made the computer faster. In 2026, AI made the memory expensive.

Current market

Storage

Holds everything else. Now also decides how fast the machine feels.

₹2,800₹6,500+132.1%index 232

500 GB, 7200 rpm, 90 MB/s, SATA 1 TB NVMe Gen4, 7,000 MB/s, PCIe

7.4% of the 2009 build 7.6% of the 2025

Priced by capacity — ₹ per GB

₹5.60₹6.50+16.1%500 GB → 1 TB installed

Cheapest, 2015: ₹2.65
Fall to 2015: −52.7%
Since 2015: +145.3%

The 2020 drop is a technology substitution, not a shortage: the build moved from a 1 TB spinning disk to a 512 GB solid-state drive, giving up half the capacity for roughly a hundredfold improvement in seek time. Price per gigabyte rises when that happens and the machine still got better.

What the money bought

Storage speed: 90 MB/s read 7,000 MB/s read77.8×
Storage capacity: 500 GB 1,000 GB2.0×

What sets this price

The 2011 Thailand floods, which took out a large share of disk assembly
NAND oversupply cycles and the periodic production cuts that end them
The substitution from spinning disk to solid state
Enterprise and data-centre storage demand for training data

Per gigabyte, storage costs more than it did in 2009. Every bit of the gain went into speed instead: the drive reads about seventy-eight times faster and the gigabyte is 16% dearer. That is a real improvement, and it is not the improvement anyone was promised.

The disk stopped spinning. The pricing cycle did not.

Current market

Graphics

Drew the interface. Now runs the workload.

₹3,200Included−100.0%index 0

Entry discrete card, 512 MB, display output Integrated in the processor — no separate purchase

8.4% of the 2009 build 0.0% of the 2025

What the money bought

Graphics: 250 Benchmark score 3,500 Benchmark score14.0×

What sets this price

Integrated graphics becoming sufficient for office work
Two mining cycles distorting consumer pricing
AI accelerator demand reshaping the entire product stack
Enterprise pricing setting the ceiling for consumer parts

In a business machine this line went to zero: from 2016 the processor did the job and the card stopped being bought. In a compute machine it went the other way entirely and is now the largest single constituent. The same component became free in one tier and dominant in another.

The component that drew the spreadsheet now costs more than the spreadsheet machine.

Reconstructed

Power supply

Feeds everything, and has to be sized for the worst part in the box.

₹2,200₹5,500+150.0%index 250

400 W, unrated 550 W, 80 Plus Bronze

5.8% of the 2009 build 6.5% of the 2025

What sets this price

Processor and graphics power draw setting the required headroom
Efficiency certification becoming a purchase requirement
Copper and capacitor input costs
Electricity tariffs making efficiency worth paying for

The business machine barely needs more power than it did — this tier draws under 150 W. The compute machine needs four times that, and the power supply is sized for the graphics card rather than for the work.

Compute became faster. The electricity meter noticed.

Reconstructed

Cooling

Removes the heat that performance creates.

₹400₹2,500+525.0%index 625

Stock cooler, one case fan Stock cooler, three case fans

1.1% of the 2009 build 2.9% of the 2025

What sets this price

Thermal output rising with sustained boost behaviour
Cooling becoming a design decision rather than an afterthought
An entire product category forming around the problem

The largest percentage rise of any constituent in this tier, from the smallest base. Cooling went from a part that came in the processor box to a line that gets specified. It is still under three per cent of the machine.

Performance created heat. Heat created another product category.

Estimated

Cabinet

Holds it together. The one part with no supply chain drama.

₹2,000₹4,500+125.0%index 225

Steel micro-ATX, one fan mount Steel mid-tower, mesh front, cable routing

5.3% of the 2009 build 5.3% of the 2025

What sets this price

Steel and shipping costs
Airflow becoming a design requirement
Nothing else, which is the point

The control group. The cabinet tracks ordinary manufacturing inflation and almost nothing else, which makes it the best available measure of how much of every other line is a supply story rather than a cost story.

Every index needs one constituent that just gets slightly more expensive.

Reconstructed

Display

The part the user actually looks at.

₹10,000₹15,000+50.0%index 150

18.5-inch TN, 1366 × 768, 1.05 megapixels 27-inch IPS, 2560 × 1440, 3.69 megapixels

26.3% of the 2009 build 17.6% of the 2025

What the money bought

Display resolution: 1.05 Megapixels 3.69 Megapixels3.5×

What sets this price

Panel fabrication capacity, which has been in surplus for most of the period
Panel technology moving from TN to IPS as the default
Size and resolution expectations rising faster than price

Display is the counterweight to everything else on this page. Half again the price for three and a half times the pixels, a better panel type and a larger screen. The deflation people assume happened to computers did happen — it happened to the monitor.

The monitor became flatter. The pricing story became more complicated.

Reconstructed

Networking and peripherals

Connects the machine to the work, which is increasingly not on the machine.

₹2,500₹8,000+220.0%index 320

100 Mbit ethernet, keyboard, mouse Wi-Fi 6, gigabit ethernet, keyboard, mouse, webcam

6.6% of the 2009 build 9.4% of the 2025

What the money bought

Network: 100 Mbit/s 1,000 Mbit/s10.0×

What sets this price

Remote work making a camera and a microphone mandatory
Wi-Fi generations turning a free port into a purchased radio
Cloud and SaaS moving the real cost outside the cabinet entirely

This line more than tripled, and it still understates the change. A 2009 machine worked without a network. A 2025 one is a client: the files, the software and increasingly the compute are somewhere else, on a subscription that never appears on this invoice.

The computer became more connected. The monthly costs moved outside the cabinet.

Estimated

05Price versus capability

What it costs against what it can do

A price index on its own says nothing about a technology product. Two indices on the same base, and the ratio between them, say almost everything.

Price index

223.7

What the machine costs. 2009 = 100.

Capability index

2,214

What it can do. 22.1× the base machine.

Price per capability

10.1

−89.9% since 2009.

CapabilityPriceLog scale, 2009 = 100

The finding

Price per unit of capability fell every year from 2009 to 2024. Then it rose +26.9%.

Computing capability grew far faster than price for most of two decades, which is the whole reason a computer stopped being a capital purchase. The current shift is different in kind: consumer machines now compete with enterprise infrastructure for their two most valuable constituents, and lose. That is not a technology slowing down. It is a buyer being outbid.

What goes into the capability index

Metric	Supplied by	Weight	2009	2025	Change
ProcessorMulti-thread score	Processor	25%	1,100	27,000	24.5×
GraphicsBenchmark score	Graphics	18%	250	3,500	14.0×
Storage speedMB/s read	Storage	14%	90	7,000	77.8×
Memory capacityGB	Memory	12%	2	16	8.0×
Display resolutionMegapixels	Display	9%	1.05	3.69	3.5×
Memory bandwidthGB/s	Memory	8%	13	90	7.0×
Storage capacityGB	Storage	8%	500	1,000	2.0×
NetworkMbit/s	Networking and peripherals	6%	100	1,000	10.0×

Weights are published rather than tuned, and they are a judgement. A capability index nobody can audit is a number with a shrug attached. Benchmark suites change between hardware generations, so read a twenty-two-fold gain as an order of magnitude, not a precise multiple.

06Why the index moved

Six markets, none of them the computer market

Every force below sets prices somewhere else and arrives here as a consequence. The cabinet is in most of these lists precisely because nothing touches it.

Semiconductor manufacturing
Throughout
Every part in the machine except the cabinet ultimately comes off a wafer. As nodes shrank, the cost of building a fab and the cost of each new process step rose faster than the transistors got smaller — and the number of companies able to build one fell to three.
- Process-node transitions
- Foundry capacity and yield
- Advanced packaging
- Geographic concentration
- Export restrictions
- Processor — Each node costs more to develop and amortise.
- Graphics — Largest dies, so most exposed to yield.
- Memory — Capacity decisions made years ahead of demand.
- Cabinet — Steel is not a semiconductor.
Memory cycles
2009 onward, repeatedly
DRAM has always overshot. Three manufacturers add capacity together, prices collapse, they cut production together, prices spike. The 2017–18 supercycle and the 2019–20 collapse are both visible in this index. What is different now is who the marginal buyer is.
- Oversupply and coordinated production cuts
- Server versus consumer allocation
- High-bandwidth memory demand
- Contract versus spot pricing
- Memory — The constituent this driver is about.
- Motherboard — Each memory generation raises board cost.
- Storage — NAND follows a similar cycle on its own clock.
Data centres
2020 onward
Hyperscale build-out changed which customer matters. A cloud operator signs multi-year contracts for memory and storage at volumes no consumer market can match, and takes priority allocation when supply tightens. The desktop buyer is now the residual claimant.
- Hyperscale cloud expansion
- Enterprise SSD demand
- Server-memory contracts
- Nearline drive shortages
- Power infrastructure
- Memory — Server DIMMs outbid consumer modules.
- Storage — Nearline and enterprise drives sold out ahead.
- Networking and peripherals — The work moved to the network, so the network became a purchase.
Artificial intelligence
2023 onward
Training and inference demand turned the graphics card into strategic infrastructure and high-bandwidth memory into the scarcest thing in the supply chain. Both are made on capacity that used to serve desktops. This is the largest single reason the 2025 column looks the way it does.
- Accelerator scarcity
- High-bandwidth memory allocation
- Local model workstations
- AI-capable consumer hardware
- Graphics — Consumer cards priced against enterprise demand.
- Memory — HBM competes for the same fabrication capacity.
- Storage — Training data has to live somewhere.
- Power supply — Higher draw needs more headroom.
- Cabinet — Unmoved, as always.
Crypto
2017–18 and 2020–21
Two mining cycles bid consumer graphics cards away from the people they were designed for, then returned them to the second-hand market at once when mining stopped being profitable. A rehearsal for what AI demand would do at a much larger scale.
- Mining profitability
- Secondary-market pricing
- Post-mining resale supply
- Graphics — Two visible spikes, both fully reversed.
- Power supply — Mining rigs pulled high-wattage units out of stock.
Global disruptions
2011, 2020–22
The Thailand floods took out disk assembly for a year. The pandemic did the same to logistics, then added a demand shock on top of it. Neither had anything to do with the technology, and both are clearly visible in the series.
- Natural disasters
- Pandemic demand and logistics
- Factory shutdowns
- Trade restrictions
- Currency movement
- Storage — The 2012 spike is the Thailand floods.
- Networking and peripherals — A webcam became mandatory in 2020.
- Display — Everyone bought a second monitor at once.

07Same money

What could Rocket Singh build with the same money?

Hold the rupees still and let the year move. For fifteen years this table only ever got better.

The sum is held fixed. The year moves. Capability is the same composite used above, indexed so the 2009 business machine is 100.

What ₹85,000 bought in each year, by constituent.
Year	Processor	Memory	Storage	Graphics	Display	Capability
2006Data confidence: Estimated	Dual core, 3.0 GHz	2 GB DDR2	250 GB disk	Mid discrete	19-inch LCD	96
2009Data confidence: Reconstructed	Quad core, 2.8 GHz	4 GB DDR2	1 TB disk	Mid discrete	22-inch LCD	178
2016Data confidence: Reconstructed	Quad core, 4.0 GHz	16 GB DDR4	2 TB disk + 250 GB SSD	Mid discrete	24-inch LED	546
2020Data confidence: Reconstructed	Six core, 3.7 GHz	16 GB DDR4	1 TB NVMe	Mid discrete	27-inch IPS	1,394
2023Data confidence: Reconstructed	Eight core, 4.5 GHz	32 GB DDR5	2 TB NVMe	Mid discrete	27-inch QHD	2,318
2025Data confidence: Current market	Six core APU	16 GB DDR5	1 TB NVMe	Integrated	27-inch QHD	2,214

2006Before the index begins.
2025This is the Business tier. The same money that bought a discrete card in 2023 does not this year.

08Calculation

The complete equation

Not a footnote. Every step, every number, the dataset it came from and the rounding rule applied to it.

processor + motherboard + memory + storage + graphics
equals the compute half
Five constituents that carry most of the price and almost all of the capability.
Source: Retail price survey
+ power + cooling + cabinet + display + peripherals
equals the rest of the machine
Five more. In 2009 they were 45% of the bill; the display alone was a quarter of it.
Source: Retail price survey
= ₹38,000
equals a business machine in 2009 — index 100
The base year, and the year the film is set in.
Source: Retail price survey
constituent price ÷ base price × 100
equals each constituent's own index
Every constituent gets its own base-100 series, so they can be compared to each other rather than only to the total.
Source: Retail price survey
Σ (constituent index × base-year weight)
equals the weighted computing index
Base-year weights, so the index measures price movement and not a change of shopping list.
Source: Retail price survey
Σ (metric ÷ base metric × 100 × weight)
equals the capability index
Eight specification and benchmark metrics with published weights.
Source: Spec sheets / benchmark suites
price index ÷ capability index × 100
equals price per unit of capability
The measure that matters. It fell for fifteen straight years and then turned.
Source: Spec sheets / benchmark suites

Conversion — One build, one price, summed from ten constituents. “Business” is held to a constant market position rather than a constant specification — pricing a 2009 machine in 2025 parts is not a comparison anyone would make. Tiers are never spliced into one series: that would measure the change in ambition, not the change in price.

09Affordability

Price is only the first question

How much it cost, how much the underlying thing appreciated, and how hard it was for a person to afford — three different questions with three different answers.

Time on the clock

How long an average income has to run to cover it.

20099.9 months

20255.0 months

Easier to afford: 0.5× the income burden it carried in 2009 — income grew faster than the price.

What the old money buys

What the 2009 outlay of ₹38,000 would buy at 2025 prices.

2009 prices1 builds

2025 prices0.45 builds

The same money buys 44.7% of what it did — the rest is price.

Inflation-adjusted value

What the 2009 price is worth in 2025 money, against what the thing actually costs.

2009 price, restated₹1,00,149

2025 price, actual₹85,000

Real change of −15.1%. Adjusted for inflation it is cheaper than it was, however the nominal price reads.

10Interpretation

What drove the change

The index sits at 224 against a base of 100. Indian consumer prices rose faster, so in real terms the machine is about 15% cheaper than it was — which is the least interesting true thing that can be said about it.

The capability index sits at 2,214. The same tier of machine is about twenty-two times more capable than the one Rocket Singh sold: twenty-five times the processor throughput, eight times the memory, seventy-five times the storage read speed, three and a half times the pixels. Price per unit of capability fell about ninety per cent.

It fell every single year from 2009 to 2024. Then it rose. That reversal is the finding on this page, and it is small in absolute terms and very large in what it implies — a fifteen-year trend in the price of computing stopped, and not because computing got harder to make.

Take the machine apart and the reason is legible. The display is up 50% for three and a half times the pixels. The processor is up 135% for twenty-five times the work. The cabinet tracks ordinary inflation, as it always has. Memory is up 471%, and almost all of that arrived in the last two years.

Memory and storage are made on capacity that a data centre also wants, and a data centre signs multi-year contracts at volumes a desktop buyer cannot approach. High-bandwidth memory for accelerators competes for the same wafers as the module in this machine and earns considerably more of them. The graphics card, meanwhile, became the machine: in the business tier it costs nothing because the processor absorbed it, and in the compute tier it is 41% of the entire bill.

The structural change is not that computers got expensive. It is that the value inside a computer moved — from the processor to the graphics card, from local storage to somebody else's infrastructure, from consumer memory to server memory. Personal computers are no longer the most persuasive customer in the markets that supply them.

Semiconductor manufacturing

Every constituent except the cabinet comes off a wafer. Nodes got smaller and the cost of each new one got larger, while the number of firms able to build a leading-edge fab fell to three.

Memory cycles

DRAM overshoots in both directions and always has. The 2017–18 supercycle and the 2019–20 collapse are both in this series. What changed is who the marginal buyer is.

Data centres

Hyperscale build-out made cloud operators the priority customer for memory and enterprise storage. The desktop buyer became the residual claimant when supply tightens.

Artificial intelligence

Accelerator demand turned the graphics card into strategic infrastructure and high-bandwidth memory into the scarcest thing in the chain. Both are built on capacity that used to serve desktops.

Crypto

Two mining cycles bid consumer graphics away from consumers and then returned it all at once. A rehearsal for what AI demand would do at a far larger scale.

Global disruptions

The 2011 Thailand floods took out disk assembly for a year. The pandemic did it to logistics and added a demand shock on top. Neither had anything to do with the technology.

What this index does not claim

The line is a paraphrase of the film's premise, not a verified quote.
The constituent prices are a reconstruction. They are internally consistent — the ten lines sum to the headline in every year, and the index is computed from them rather than typed in — but each line is a representative retail price for its tier, not a specific quotation from a specific shop on a specific day.
“Business”, “Performance” and “Compute” are judgements about market position, not fixed specifications. Holding market position constant is the only honest way to price technology across seventeen years, and it is still a judgement.
The capability index is a weighted composite of eight metrics whose weights are published on the page and are a defensible choice rather than a derived one. Benchmark suites change between hardware generations, so a twenty-two-fold capability gain is an order of magnitude, not a precise multiple.
The 2025 memory, storage and graphics figures are current-market rather than settled. They were moving fast at the snapshot date and an annual figure understates where the year ended.
The three tiers are priced at the snapshot only. There is no seventeen-year history for Performance or Compute, because those categories did not mean the same thing in 2009 and splicing them would measure ambition rather than price.
The index prices the parts, not the labour, the warranty or the shop. Rocket Singh's business sold all three, and its margin lived in exactly the place this index cannot see.
Cloud, SaaS and connectivity costs are named in the peripherals constituent but are not in the series. A recurring monthly cost is not a component price and putting it in one would make the index dishonest in a different direction.

11Provenance

Source ledger

Every dataset behind this page, with the adjustments applied on the way in and how gaps are treated. Last verified 31 December 2025.

Mid-range self-assembled desktop PC — total component cost, India, ₹Duku Design retail survey / PCPartPicker / Indian component retailersData confidence: Reconstructed2009–2025
Publisher
Duku Design retail survey / PCPartPicker / Indian component retailers
Date range
2009-01-01 to 2025-12-31
Geography
India
Frequency
annual
Original unit
₹ per component, Indian retail listings
Normalised unit
₹ for a complete mid-range build (eight priced lines)
Currency
INR
Adjustments
“Mid-range” is held to a constant market position, not a constant specification — the build that a mainstream buyer would choose in each year. That is the only way to price technology over sixteen years, and it is also why the nominal figure understates what actually changed.
Missing data
Indian listings are thin before 2012; those years are reconstructed from US retail converted at the year's average exchange rate plus a customs and margin uplift.
Last checked
2025-12-31
Licensing
Compiled from retail listings and price trackers, cited with attribution
Source
https://pcpartpicker.com/trends/
Representative business-machine capability — benchmark and specification metricsManufacturer specifications, PassMark and 3DMark score archives, storage review benchmarksData confidence: Reconstructed2009–2025
Publisher
Manufacturer specifications, PassMark and 3DMark score archives, storage review benchmarks
Date range
2009-01-01 to 2025-12-31
Geography
Global specifications, Indian retail availability
Frequency
annual
Original unit
Benchmark score, GB, GB/s, MB/s, megapixels, Mbit/s
Normalised unit
Index, 2009 = 100, weighted composite
Currency
n/a
Adjustments
Each metric is the figure for the representative part in that year, not a measurement of any individual machine. Metrics are rebased to 2009 = 100 and combined with the published weights shown in the capability section. Benchmark suites change between generations, so cross-decade scores are indicative of order of magnitude rather than exact ratios.
Missing data
No gaps. Every year has a representative configuration, because the tier is defined by market position rather than by a specific product.
Last checked
2025-12-31
Licensing
Cited with attribution; headline scores only
Source
https://www.cpubenchmark.net/
DRAM and NAND flash contract prices — industry benchmark, US$TrendForce DRAMeXchange contract and spot price reportingData confidence: Reconstructed2009–2025
Publisher
TrendForce DRAMeXchange contract and spot price reporting
Date range
2009-01-01 to 2025-12-31
Geography
Global
Frequency
monthly
Original unit
US$ per module and per gigabyte, monthly contract price
Normalised unit
Index, 2009 = 100, annual average
Currency
USD
Adjustments
Monthly contract prices averaged to an annual figure and rebased to 2009 = 100. Used here as context for the Indian retail component series, not as a substitute for it — an Indian buyer pays a retail price that includes freight, duty and margin.
Missing data
The 2025 average runs to the snapshot date and is provisional. Memory prices moved fast enough in the second half of 2025 that an annual average understates where the year ended.
Last checked
2025-12-31
Licensing
Cited with attribution; headline figures only
Source
https://www.trendforce.com/price
India consumer price index, all groups — annual average, rebased 2012 = 100MoSPI (CPI Combined) spliced to Labour Bureau (CPI-IW)Data confidence: Reconstructed1999–2025
Publisher
MoSPI (CPI Combined) spliced to Labour Bureau (CPI-IW)
Date range
1999-01-01 to 2025-12-31
Geography
India (combined rural + urban)
Frequency
annual
Original unit
Index, mixed bases
Normalised unit
Index, 2012 = 100
Currency
INR
Adjustments
CPI Combined begins in 2011. Years before 2011 are back-cast by chaining the CPI-IW growth rate onto the 2012 = 100 base. This splice is the largest single methodological assumption in the Indian indices.
Missing data
No gaps. Pre-2011 values are derived, not observed — flagged as such on the chart.
Last checked
2025-12-31
Licensing
Public statistics, cited with attribution
Source
https://mospi.gov.in/web/mospi/download-tables-data/-/reports/view/templateFour/16702
Per-capita net national income at current prices — monthly equivalent, ₹MoSPI, National Accounts StatisticsData confidence: Reconstructed1999–2025
Publisher
MoSPI, National Accounts Statistics
Date range
1999-01-01 to 2025-12-31
Geography
India
Frequency
annual
Original unit
₹ per capita per year, current prices
Normalised unit
₹ per capita per month
Currency
INR
Adjustments
Annual figure divided by 12. Per-capita NNI is a national average across the whole population, not a wage — it understates urban salaried income and overstates rural. Read the affordability rows as a direction of travel, not a payslip.
Missing data
Fiscal years mapped to their starting calendar year. The 2025 value is a provisional advance estimate.
Last checked
2025-12-31
Licensing
Public statistics, cited with attribution
Source
https://mospi.gov.in/web/mospi/national-accounts-statistics

Pop Indices is editorial and educational. Historical returns are described, never projected, and nothing here is investment advice. Film titles, characters and dialogue are referenced for commentary and remain the property of their rights holders.

The computer became more powerful.The market around it became more concentrated.

This is not a measure of computer inflation. It is a record of where the value inside a computer went.

From the processor to the graphics card
From local storage to somebody else's infrastructure
From consumer memory to server memory
From personal productivity to artificial intelligence

In 2009, Rocket Singh had to understand the customer. In 2026 he would also need to understand the data centre.

A Duku Design experiment in cultural memory, product economics and the changing price of compute.

Take it with you

The card carries its own provenance

A figure that travels without its source is the thing this section exists to avoid, so the export always includes the credit line.

1080 × 1080 · square posts everywhere