The number that traveled was 0.7%. India's onshore crypto exchanges, which once handled roughly 7% of the country's centralized exchange inflows, now capture 0.7% of them. Brazil's onshore venues moved the opposite way — from about 1.5% to 12.5% across a comparable stretch. Both figures come from a Chainalysis report, and both are repeated far more confidently than the report itself permits.
That gap is the real story. Not that India's exchanges are dying, but that a methodologically careful dataset told us something narrower than the headlines extracted from it. I have spent enough time inside on-chain attribution to know a share metric is a model, not a measurement. This model has three moving parts, and only one of them is the tax.
India is, by volume, the largest crypto market in Central and Southern Asia and Oceania — roughly $88.4 billion in centralized exchange inflows over the measured year. The user base is not small and the activity is not fake. What changed is where the value lands. Onshore platforms — CoinSwitch, CoinDCX, and the rest — watched their share of that flow collapse, with the sharpest drop concentrated around mid-2022.
Two policy instruments shadow that date. Section 393 of India's Income Tax Act, which carried forward the earlier Section 194S, imposes a 1% tax deducted at source on virtual digital asset transfers. It is not an annual filing obligation. It is a withholding at the point of transaction. The threshold sits at ₹50,000 per year for specified individuals and HUFs, ₹10,000 for other payers, with the obligation exempting payers under ₹1 crore of turnover or ₹50 lakh of professional receipts. Overpayment is recoverable through credit and refund. That architecture matters more than the rate.
The responsibility allocation is messier than the rate suggests. The original Section 194S framework distinguished between buyer liability in direct peer-to-peer trades and exchange liability in settled transactions. Different transaction structures produce different withholding agents. Compliance, in practice, is a routing problem before it is a tax problem.
The timing rule is the mechanical heart of it. Withholding triggers at the earlier of credit or payment — not at withdrawal, not at settlement of a position, not at year end. The cash is extracted the instant the transfer is recorded. For an active trader recycling capital intraday, the effective cost compounds with frequency. A 1% levy applied across ten round trips is not 1% of a position. It is a tax on velocity.
Brazil is the control group the report reaches for. Its onshore share climbed from 1.5% to 12.5%, which reads as proof that friendly tax treatment revives local venues. Read it more slowly and the comparison weakens: the samples are not matched, the observation dates are not aligned, and the denominators are not identical. A control group that isn't controlled is a suggestion, not a finding.
Start with the attribution method, because everything downstream inherits its errors. Chainalysis assigns service activity to a country using website traffic, adjusted by the square root of per-capita GDP. It is not node-level attribution. It is not KYC-level attribution. It is a proxy assembled from browsing behavior and an economic correction factor, and the firm says so.
I audited crowd-sale contracts in 2017 and submitted integer-overflow findings to core developers who answered with automated tickets. Nobody peer-reviewed the methodology. The same structural gap applies here: a commercial firm publishes a geographic attribution model, and the market treats it as a census. Transparency is a feature, not a default state — Chainalysis discloses its limitations, which is more than most peers do, but disclosure does not remove bias.
Algorithmic fairness assumes fair inputs. A traffic-based model rewards platforms with large, brand-direct browsing bases. Binance and its peers generate enormous identifiable India traffic. Smaller onshore venues, whose users arrive through app deep links and referral paths, are harder to see. If the model underweights them, 0.7% is not a floor. It is a number with an unknown amount of downward pressure already baked in.
The contamination problem is familiar. In 2026 I audited the oracle feeds feeding autonomous trading agents and found that 40% of the training data was synthetic transaction history generated by rival protocols. Garbage in, garbage out, and the output still looked authoritative. Chainalysis admits that removing VPN and bot traffic is imperfect and that the uncertainty is embedded in the comparison. That admission is the honest part. The dishonest part is the market quoting the result as a hard fact.
Here is the construction that matters most: an Indian user who reaches an offshore venue through a VPN is still counted inside India's measured market, but is not counted inside India's onshore exchange share. The numerator and the denominator are drawn from different populations. That does not merely describe the shift. It amplifies it.
Now the tax. Section 393 does not wait for the fiscal year. On a ₹100,000 transfer, ₹1,000 is withheld at the moment of settlement, and that ₹1,000 is not available for the next purchase. Ashish Singhal of CoinSwitch made the point directly: the cash leaves the user's working balance immediately. For a high-frequency retail trader, that is not a rounding error. It is a permanent drag on turnover, applied at the worst point in the cash cycle.
This is where the design shows its hand. Most jurisdictions tax realized gains after the fact. India taxes the transaction itself. The logic held; the incentives were broken. A regime intending to capture revenue instead changes the venue where the transaction happens, and collects nothing from the venue it pushed users toward.
Code does not lie, but it can be misled, and so can a statute. The withholding regime assumes the taxable event stays inside the jurisdiction. That assumption is the load-bearing wall, and it is the first thing a VPN removes.
This is the same failure mode I keep finding in governance design. "Code is law" collapses the moment an obligation attaches to a legal person rather than a contract. Section 393 does not ask a smart contract to withhold; it asks an entity to. The enforcement surface is an office in Bengaluru, not a state machine. Every compliance regime eventually reveals that the upgrade path runs through humans with signing keys.
The asymmetry is structural. Onshore exchanges are designated withholding agents. They must deduct, report, and remit. An offshore platform without an Indian presence may simply not deduct. Singhal said as much, and the report repeats it as an operator's framing rather than a proven fact — correctly, because the offshore compliance picture is genuinely unclear. But the competitive geometry is not unclear: the compliant venue carries a cost the non-compliant venue may not, and the user pays the difference in friction.
I traced the hash to the wallet, and the wallet was offshore. The report notes that executed trades, revenue, and customer counts measure a different part of the business than received value does. That is the quiet admission: onshore platforms may still have users. What they lost was the high-value flow — the traders whose volume justified the venue. Retail headcount and captured value decoupled, and the decoupling is the actual damage.
The damage does not stop at the exchange's own P&L. Onshore venues anchor a perimeter — payment rails, custody providers, compliance vendors, local market makers. When the anchor loses its flow, the perimeter loses its reason to exist. I have watched this in Layer 2 fragmentation: splitting the same liquidity across more venues does not scale it, it thins it. Here the thinning runs the other direction. Liquidity consolidates offshore, and the onshore perimeter starves.
In 2020 I spent hundreds of hours tracing Compound's incentive flows and found the headline yield was largely inflationary emissions, not organic revenue. The pattern repeats at the exchange level. Onshore share is supported by a policy environment, and when that environment turns hostile, the share does not decline gradually. It reprices.
Here I have to be careful about causation, because the report is. The onshore collapse clustered around mid-2022. The TDS took effect in July 2022. The same months contained the collapse of Terra and the broader deleveraging that erased more than $2 trillion from the market. Three variables, one window. When I modeled the Luna burn loop in 2022, the discipline that saved the analysis was refusing to attribute a single output to a single input. The report states plainly that the extent to which the withholding tax drove the share decline remains unmeasured. That sentence is the most important one in the document, and it is the one the headlines dropped.
There is a data gap the report also flags but does not resolve: the specific observation date for the 0.7% reading is not pinned down, and sample consistency across the comparison is not disclosed. If the reading was taken in a particular month, it is a snapshot, not a window average. Synchronized year-over-year share comparisons require matched dates and matched samples. Neither is confirmed.
One more temporal wrinkle: the report's publication window sits well after the activity it observes. The dataset captures flows from before October, released near the end of September. In a market that reprices on policy announcements, a four-to-six week lag can mean the reading describes a regime that has already shifted. Readers treating the figure as current are reading a rearview mirror.
So what did the bulls get right? Two things, and they matter more than the bearish read. First, India's activity is intact. $88.4 billion in inflows is the largest regional figure in the dataset. Capital did not leave crypto; it left a jurisdiction's onshore venues. That is the difference between a market dying and a market routing around a toll. Anyone reading 0.7% as "India rejected crypto" has misread the numerator.
Second, onshore share is not a one-way ratchet. Brazil climbed from 1.5% to 12.5%. Whatever combination of tax environment and local execution produced that, it proves reversibility. A share that fell can rise. The bearish case for Indian exchanges is a policy case, not a structural verdict, and policy is the one input that can change inside a single budget cycle.
The falsification test is simple. If onshore share keeps falling while the broader market recovers, tax friction is doing the work. If it stabilizes as volatility normalizes, the mid-2022 collapse was mostly the deleveraging wearing a tax costume. That is the experiment nobody has run yet, and it is the one that would settle the argument.
There is also a paradox the bears skip: a withholding regime that pushes the tax base offshore can collect less than it would have collected onshore. If users migrate to venues beyond the reach of Section 393, India has not taxed the activity — it has relocated it. Tax regimes that drive their base across a border are not enforcement victories. They are measurement failures wearing a revenue label.
The 0.7% figure will be cited for years, and most citations will treat it as a fact rather than a model output with a known bias. Watch three signals instead. Whether the next report matches samples and dates, which would make the Brazil comparison real rather than rhetorical. Whether CBDT imposes registration or withholding obligations on offshore platforms, which would close the arbitrage the onshore venues are complaining about. And whether onshore share stabilizes before the next policy cycle, which would tell us how much of the collapse was the tax and how much was the market.
A share is not a verdict. It is a reading taken with a known instrument error, published by people honest enough to name the error, and consumed by people who did not read that far. The number that matters is not 0.7%. It is the distance between what the data supports and what the market decided it said.