Assurd Techlabs

India's Refurbished Market Is Formalising. GPUs Are the Hard Case.

|By Assurd Engineering Lab|Market Signals

New devices are getting more expensive, and Indian buyers are turning to used ones. The phone market is already reorganising around that shift. Graphics cards are under the same pressure, but they lack what makes a used phone easier to trust: a way for the buyer to see what they are buying. This post follows that line from the published research, to the gap it leaves for a used GPU, to how an Assurd certificate is designed to close it.

Cited Research

Phones: price pressure is formalising resale

As of Counterpoint Research's H1 2026 release on India's refurbished smartphone market (published 1 September 2026), refurbished smartphone volumes in India grew 13 % year on year in H1 2026, while new smartphone volumes declined by around 11 %.[1] Counterpoint ties the shift to price pressure on new phones, strongest in the entry and mid-range segments, and its senior analyst names rising memory prices as a driver of new-device costs.

The market is changing shape as it grows. Counterpoint estimates that the organised channel holds about 20 % of India's refurbished smartphone market and the unorganised channel about 80 %, which it calls “significant headroom for formalization”. It credits Samsung's Certified Re-Newed initiative and other OEM-led programmes with bringing structure and credibility to the category, and it describes investment pushing the ecosystem toward stronger quality standards, transparency and consumer trust.

Graphics cards: the same pressure, a growing installed base

Graphics cards are feeling the same forces. As of Jon Peddie Research's Q2 2026 add-in-board release (published 9 September 2026), global shipments of PC graphics add-in boards reached 12.5 million units in Q2 2026, in a quarter that JPR says is traditionally down from the one before.[2] Jon Peddie names war, a memory shortage and shifting tariffs as the pressures on the market. He reads the quarter's spike in high-end sales as buyers getting ahead of price rises: “consumers rushed to buy AIBs before the prices went any higher”. JPR forecasts an installed base of 193 million add-in boards by the end of its 2025–2030 forecast period.

A Market Buyers Cannot See Into

Counterpoint's subject is phones, where certified programmes are credited with making used devices credible. For graphics cards, the research shows the price pressure and an installed base that will feed the next used market. But none of it measures the resale of graphics cards, and we found no published source that does. The used GPU market is hard to see in aggregate, and each card in it is hard to see on its own.

A typical listing gives the buyer photos, a GPU-Z screenshot and perhaps a benchmark result. None of them shows the conditions that decide whether a used card is a good buy:

  • Power delivery. A shorted power stage looks the same in a photo as a healthy one, and powering the card to find out can destroy the evidence.
  • Firmware and identity. A screenshot cannot show whether the firmware on the flash chip is a factory image, or whether the reported device matches the board in the box.
  • Thermal interface. One benchmark run is one moment. Dried-out paste or pads show up under sustained load, as a hotspot running far ahead of the core.
  • Memory. Most consumer cards expose no ECC error counters, so memory health is the hardest condition to read from software.
  • The report itself. A screenshot or PDF can be edited, and the buyer cannot tell who produced it.

Economists have studied what happens to a market like this. George Akerlof's 1970 paper on quality uncertainty[3] is summarised by the Royal Swedish Academy of Sciences, in its announcement of the 2001 Prize in Economic Sciences (10 October 2001), as showing how a market where sellers know more than buyers about quality can “contract into an adverse selection of low-quality products”.[4] For used GPUs, as we read it, that means a well-kept card sells at the price of a doubtful one, and the owner of a good card has little reason to sell through a channel that cannot prove the difference.

Turning a Card's Condition Into Evidence

The same 2001 announcement credits Michael Spence with the other half of the story. The better-informed side of a market can take a costly action that passes information credibly to the other side, which the announcement calls signalling. An Assurd certificate is designed to be that action for a used GPU. Each condition above maps to a step of SOP-GPU-001 v1.1, and the linked Playbook posts explain each method in full.

  • Power delivery is checked before power is applied. Phase 1 diode-checks the 12 V and 3.3 V inputs for a dead short and reads each primary rail's resistance to ground with a bench multimeter (Steps 1.5.1 and 1.5.2). A dead short quarantines the card before it is ever switched on, and the reading also catches gross partial shorts. How the rail check works.
  • Firmware is read from the chip, not from a screenshot. Phase 2 reads the VBIOS straight from the card's SPI flash. The server hashes the dump and checks it against a registry of factory images for that device and board (Step 2.2), and the certificate records the verdict. GPU-Z's fake-card flag or a critical spec mismatch stops certification (Step 2.3). What a registry match proves.
  • Thermals are measured over 30 minutes, not one run. Phase 3 logs telemetry every second across 1800 s of sustained load and takes the hotspot delta over the last 300 s (Steps 3.2 and 3.3). A degraded thermal interface is disclosed on the certificate. Why the trace runs 1800 s and what the hotspot delta shows.
  • Memory errors are recorded wherever the card reports them. On cards that expose lifetime ECC counters, the certificate records the uncorrectable-error count (Step 4.4). On cards that don't, it says so plainly rather than showing a zero. What the counters tell you.
  • The certificate proves where it came from. Every certificate is signed with an ES256 key held in Google Cloud KMS (Step 5.2). Anyone can verify the signature, in a browser or offline with OpenSSL, without taking the seller's word for it. Verify one yourself.

With a certificate, a listing can carry evidence instead of claims: a signed record of how the card's power stages, firmware, thermals and memory measured on the bench. Phones got that credibility from certified programmes. A certificate is designed to bring the same thing to the component whose condition is hardest to see.

Limits of This Analysis

  • The evidence is adjacent to the claim. Counterpoint measures phones, and JPR measures new add-in-board shipments. No source here measures used-GPU volumes or prices in India, and a primary source that does would test this reading directly.
  • Single periods. Each figure covers one period: H1 2026 for phones, Q2 2026 for graphics cards. Counterpoint does not state the unit or period of its organised / unorganised split.
  • What would change the reading. The reading would weaken if used-GPU buyers already had another reliable way to see a card's condition, such as warranties that transfer to a second owner.

Sources

  1. India Refurbished Smartphone Volumes Grew 13% YoY in H1 2026 as Rising Component Costs Boost Demand. Counterpoint Research. Published ; accessed . ↩
  2. Jon Peddie. Q2’26 PC graphics AIB shipments increased 10% from last quarter to 12 million units. Jon Peddie Research. Published ; accessed . Archived copy. ↩
  3. George A. Akerlof. The Market for "Lemons": Quality Uncertainty and the Market Mechanism. The Quarterly Journal of Economics, 84(3), Oxford University Press. Published ; accessed . Archived copy. ↩
  4. The Prize in Economic Sciences 2001 – Press release. The Royal Swedish Academy of Sciences, via NobelPrize.org. Published ; accessed . Archived copy. ↩