Hey all, Happy Wednesday! Crypto’s weekly gains held up despite Tuesday’s pullback as oil approached $100 and Strategy paused BTC purchases again. We also look at Solana’s Alpenglow upgrade, what is coming to mainnet and what it could mean for SOL.

US markets traded lower as they returned from the long Labor Day weekend, while crypto posted relatively modest losses. Brent approached $100 amid Tuesday’s renewed fighting, with US forces reporting the destruction of five Iranian tankers and Iran responding with missile launches toward US targets in Jordan.

BTC fell 0.7% to around $79,000 following Monday’s 1.7% decline, while the S&P 500 lost 0.6%, the Nasdaq 100 fell 0.8% and gold declined 2.1%. Meme (3.9%), Crypto Miners (2.8%) and Privacy (1.7%) led, while Modular (-4.1%), Crypto Equities (-2.5%) and Lending (-2.4%) lagged, with 15 of 25 indices finishing lower. 

Despite Tuesday’s broad pullback, all 25 sector indices remain higher over the past seven days, including 13 with double-digit gains. As we noted over the past week, Privacy has gained nearly 30% alongside ZEC’s rally, while Crypto Miners are up 26% as AI demand continues to shape the sector’s performance. Those weekly gains suggest appetite for risk has extended well beyond BTC. 

Notably, Strategy opened the week by disclosing no further BTC purchases, just one week after buying 4,603 BTC to end a 10-week pause. Instead, the company spent $176.3M repurchasing STRC and doubled its digital credit buyback authorization to $2B. With MSTR trading at 0.87x NAV, Strategy prioritized reducing preferred dividend obligations over adding more BTC. That marks a change from the previous week, when it raised $602.8M through stock sales, using $369.7M to buy BTC and $151.8M to repurchase STRC. 

Crypto enters this week’s inflation releases with most of its recent gains intact following Tuesday’s pullback. Attention now turns to Thursday’s PPI and Friday’s CPI, the final inflation readings before the Fed’s Sept. 16 FOMC decision.

— Joe

Crypto's premier institutional event is arriving in Asia this October 7.

Join us for a day of top-tier conversations and networking and hear CEO & Founder voices from the top protocols in the space.

Alpenglow Isn’t Finished

Having passed validator governance in September 2025, Alpenglow is now targeted for mainnet activation later this year, retiring the two mechanisms that have defined Solana since launch, Proof of History and TowerBFT. The upgrade promises a reduction in finality from roughly 12.8 seconds to a target of 100 to 150 milliseconds, though the version reaching mainnet is narrower than that headline suggests.

While the activation proposal includes Votor, the new voting and finality layer, it excludes Rotor, the propagation redesign that does the heavy lifting for sub-second finality, which has no published timetable. Given the 100 to 150 millisecond target assumes both components are live, Turbine will continue carrying blocks across the network at activation, meaning Votor's gains are capped by whatever Turbine delivers until Rotor arrives under its own proposal.

Where TowerBFT reaches finality by stacking 32 successive votes, which is where the 12.8 seconds comes from, Votor gets to the same place in a single round when 80% of stake agrees, or two rounds when 60% does, running both paths concurrently and finalizing on whichever completes first. Votes also cease to be transactions altogether, as validators exchange them directly and anchor a compact certificate onchain, removing the roughly three quarters of Solana's transactions that are currently just validators voting.

Although the new Validator Admission Ticket replaces roughly 2.16 SOL per epoch of vote fees with a flat 1.6, a 26% cut to a validator's breakeven cost, the saving is less stable than it appears. Because an epoch is a fixed count of slots rather than a fixed span of time, epochs come around more often as slot times fall toward 200 milliseconds, meaning the daily outlay roughly doubles even as the per-epoch fee holds. The more consequential shift is for SOL itself: where vote fees today are only half burned, the ticket burns in full, lifting burn per validator by about 50% and pushing consensus burn past what the entire network currently burns from all transactions combined. Against daily issuance of around 68,800 SOL this remains trivial, but the direction is right.

While votes may comprise most of Solana's transactions, they only represent about 5% of its compute, meaning the freed blockspace is modest, though it does remove a floor of consensus traffic that scaled with the validator set rather than with demand. 

Whether any of that capacity accrues to SOL was supposed to be answered by SIMD 553, which would have burned a resource-based fee on compute-heavy transactions but failed after application teams pushed back on the cost. Until a resource-based fee passes in some form, Alpenglow makes Solana faster and cheaper to secure while leaving the value accrual question open.

— Toma

Pretraining progress is mostly coming from data

Dwarkesh Patel and Jerry Han find that improvements in training data contributed far more to pretraining progress between 2019 and 2025 than changes to model architecture. At a fixed compute budget, newer datasets produced a 12x gain in compute efficiency compared with 3.7x from newer model recipes, with most of the gains occurring independently. However, model research still played a crucial role by making larger training runs stable and practical. The results come from relatively small models, so the balance may differ at the frontier. If data remains the main driver of pretraining progress, the key question is whether synthetic data can offset the finite supply of high-quality material available online.

Stellar Q2 Token Holder Report

Blockworks finds that Stellar had its strongest quarter of institutional adoption to date, with tokenized RWAs doubling to $3B, payment volume rising 68.8% to $16.43B and DTCC selecting the network for tokenized securities trials. Stellar is also evolving beyond payments, with Soroban smart contracts now generating 94.4% of network fees and 82.4% of onchain trading volume. However, network adoption does not translate directly into value accrual for XLM. Fees remain minimal, are neither burned nor distributed, and circulating supply increased 2.8% during the quarter. Growth also remains concentrated, with four issuers holding 92.5% of tokenized assets, making issuer diversification and DTCC’s implementation key milestones to watch.