In May the Hyperliquid thesis rested on three proofs: revenue, the buyback, and the Circle deal. Three months later two of them have landed and one has not. The buyback engine runs daily and now provably burns — 49M HYPE cumulative, 4.9% of max supply, at an average price paid of ~$28. AQAv2 went from announcement to cash: activated on 26 August, it delivered a first payment of $14.58M on 3 October against a ~$5B USDC base, of which $10.15M had already bought and burned 112,600 HYPE. Revenue is the clause that missed. $429M through 15 September is first on CoinGecko's 2026 ranking excluding stablecoins and ~$106M ahead of Pump.fun in second — but it is ~$50M a month, an annual pace near $600M against $844M in 2025, and that is below the bear case in the May report. Volume did not fall; the take rate did. Meanwhile HYPE went from ~$45 to a record ~$98 on 23 September. The engines are real and auditable; what has changed is what you pay for them.
Volume rises, revenue doesn't follow
The May report set three tests, and this addendum exists to mark them. Revenue is the one that missed, and it missed in an instructive way: not by activity falling, but by the business earning less from the same activity. Hyperliquid booked $429M through 15 September, first on CoinGecko's 2026 ranking excluding stablecoins and ~$106M clear of Pump.fun in second — a genuinely dominant position. Yet that is ~$50M a month, an annual pace near $600M against the $844M recorded in 2025, and on the 30 days to 6 October revenue was $58M, about $0.7B a year in fees. Add AQAv2 and the buyback fuel comes to ~$0.87B, against the ~$1.03B the May model assumed. That is below the report's own bear case. Volume did not cooperate with the pessimism — perp volume went from ~$180B over 30 days in April to more than $200B by 16 September, with open interest at $14.7B, the highest since October 2025 — which is exactly what makes the shortfall diagnostic rather than cyclical. The effective fee per dollar traded fell. The HIP-3 builder share and staking discounts are the natural suspects, and the number to watch is revenue per dollar of volume: if it does not stabilise, more volume stops turning into more buyback.
Still not just hype — but price ran ahead
The other two tests passed, and they passed on-chain where anyone can check them. The Assistance Fund buys daily with ~97–99% of fees, typically $1–2M a day, and what it buys is burned — with the fund's historic stock recognised as burned through governance, which turns a holding into a retirement. Cumulative burn stands at ~49M HYPE, 4.9% of max supply, at an average price paid of ~$28. The second engine is the one that matters more for durability, because it is countercyclical: AQAv2 was announced on 14 May, activated on 26 August with Coinbase as treasury deployer and Circle as technical deployer, and routes ~90% of net reserve yield to the fund. Its first payment arrived on 3 October — $14.58M on a ~$5B USDC base, $10.15M of which had already bought and burned 112,600 HYPE. Annualised, ~$175M against a May model of $190M: ~8% light, and tied to deposits rather than to trading volume. Mechanism confirmed, thesis corroborated.
So the honest read is that this is still not only hype — but the price ran ahead of it. Both engines exist, both are auditable, and the USDC one now pays in cash. What changed in five months is the relationship between price and fundamental: HYPE doubled from ~$45 to a record ~$98 on 23 September while annualised revenue fell ~15%. The multiple more than doubled on both rulers — market cap over revenue from ~10× to ~24×, FDV over revenue from ~44× to ~106× — against the CME at ~15×. In May the circulating market cap was cheaper than the CME per dollar of revenue; today it costs ~1.6× more, and ~2.4× more on FDV. One consolation sits inside the same arithmetic: the share of unlocks the buyback can absorb through November 2029 improved from the May bear case of 15% to ~6% of what remains, because the fuel is still flowing even as it buys fewer tokens. What the market is paying for today is optionality — US entry through the pending Payward/Bitnomial HIP-3 markets, RWAs, ETFs — rather than the cash flow in front of it.
Key findings
- Volume rose and revenue did not follow — the one clause that missed. On the same ruler (DefiLlama), 30-day perp volume went from ~$180B in April to more than $200B by 16 September, open interest reached $14.7B (the highest since October 2025) and DEX leadership is intact. But $429M in 8.5 months is ~$50M a month, an annual pace near $600M against $844M in 2025. More volume with less revenue means a lower effective fee per dollar traded; the HIP-3 builder share and staking discounts are the natural suspects.
- The buyback engine runs — and now it burns. The Assistance Fund buys every day with ~97–99% of fees, typically $1–2M a day, and the HYPE it buys is burned; the fund's historic stock was recognised as burned through governance. Cumulative burn is ~49M HYPE, 4.9% of max supply, at an average price paid of ~$28. The caveat is arithmetic: with HYPE near $93, every buyback dollar retires half the tokens it retired at $45.
- AQAv2 went from announcement to cash, and it is the countercyclical engine. Announced on 14 May and activated on 26 August, it has Coinbase as treasury deployer and Circle as technical deployer for issuance, redemption and CCTP, with ~90% of net reserve yield routed to the Assistance Fund. The first payment landed on 3 October: $14.58M on a ~$5B USDC base, of which $10.15M had already become a buy-and-burn of 112,600 HYPE, with the balance still in the fund. Annualised that is ~$175M against the May model's $190M — about 8% below — and it is tied to deposits rather than to volume.
- The price ran ahead of the fundamentals, and the multiple is where you see it. In five months HYPE went from ~$45 to a record ~$98 on 23 September while annualised revenue fell ~15%, from ~$1.03B to ~$0.87B. Market cap over revenue went from ~10× to ~24× and FDV over revenue from ~44× to ~106×, against the CME at ~15×. In May the circulating market cap was cheaper than the CME per dollar of revenue; today it costs about 1.6× more, and on FDV the jump is ~2.4×. What the market is paying for now is optionality — US entry, RWAs, ETFs — rather than cash flow.
- One number decides whether the thesis holds: revenue per dollar of volume. If the take rate does not stabilise, more volume simply does not become more buyback, and the whole flywheel loses its input. Three other things are worth watching alongside it. Payward (Kraken) announced permissioned HIP-3 markets for US clients on 16 September with Bitnomial as deployer and clearing, still pending CFTC approval. Labs unstaked 3.75M HYPE (~$329M) — October's team tranche — sold OTC to a single institution with no disclosed lock-up. And AQAv2's second payment is due in early November, where the question is whether the USDC base clears $5B toward the base case's $7B.