Skip to content

Pulse - April 25

25 Apr 202620 min
Read by asset class: MacroCrypto
ContentsOverview

Records on the surface, stress underneath. Equities kept printing highs, oil pushed back into triple digits, and crypto got another reminder that fragile security still sits under the hood. Fresh highs can still hide uglier truths.

Key Takeaways

  • Oil is back in charge; triple-digit Brent is the macro tell.
  • Record highs in SPX and NDQ hide a narrower, more fragile market underneath.
  • Gold lost momentum even with geopolitics still supportive.
  • AI is widening both enterprise moats and attack surfaces.
  • Kelp’s exploit served as a reminder that DeFi is one dependency chain, not a set of isolated apps.
  • Prediction markets moving into perps is a theme worth tracking.

Macro: The Hormuz Risk Premium

Kicking off the week with both SPX and NDQ at fresh records, momentum hinged on the U.S.-Iran situation while stagflation pressure got louder. Oil retook triple-digits trading an unresolved energy shock. Iran still has leverage over the Strait of Hormuz, the U.S. blockade is still in place, and Hormuz normally handles about 20% of daily global oil supply. That’s the channel through which inflation, yields, and growth all get hit at once.

OIL — Oil is still the main macro variable. Brent topped $103 on Thursday and briefly pushed near $105, while WTI traded above $94. The market is not pricing apocalypse, but it is pricing ongoing disruption. Iran says it will not reopen the strait unless the U.S. lifts its blockade, while Washington says it has “total control.”

GOLD — Gold spent the week grinding lower and losing momentum after failing near recent highs. Geopolitics in the Strait of Hormuz remain tense, which is theoretically gold‑supportive via risk premium and higher energy/inflation expectations. But the stronger dollar and higher-for-longer rates narrative are leaning against it.

SPX — The S&P 500 ground sideways on the week. Monday and Tuesday were sold on renewed U.S.-Iran tension and higher oil, then Wednesday ripped to a record close as Trump extended the Iran ceasefire and earnings held up, and Thursday turned into digestion as mixed results hit sentiment and Brent stayed above $100.

NDQ — Leadership is still concentrated in AI, semiconductors, and mega-cap tech rather than broad cyclicals. Same weekly rhythm as SPX, just with more torque: early-week geopolitical de-risking, then a sharp Wednesday squeeze on strong earnings and another leg higher in tech, before Thursday’s pause as IBM and Tesla weighed on sentiment and fresh concern hit legacy software models exposed to AI disruption.

Crypto: Composability Cuts Both Ways

It’s one wake-up call after another, totaling almost $800M lost in 2026 already after this week’s Kelp $290M rsETH exploit. AI is the invisible hand behind the scenes, shedding light on careless and sloppy human practices. Those same practices sit underneath code securing millions in unprotected capital deployed onchain. Exploiting code has always been easier than hardening it. AI has made it effortless to push unvetted code as developers themselves take shortcuts, and the offense still has the edge over defense. That’s forcing a lot of investors to ask whether having money onchain is still worth the structural risk.

AI is the quiet presence in the room that everyone feels but few name. It tilts the odds without demanding attention. The underlying capabilities have been around since at least GPT‑5 (Summer 2025) and Opus 4.5 (November 2025); what’s changed is that they’re now being actively weaponized. On top of that, everyone has access to far more powerful models now.

Kelp’s rsETH exploit underscored again that DeFi isn’t a set of isolated protocols, but a chain of dependencies where composability cuts both ways: stress in one corner can cascade everywhere. Things get more fragile once you add liquid staking, with the same LSTs threaded through multiple lending venues. It gets even uglier when a 1:1 multisig is responsible for managing a seemingly “risk-free” asset (as in “backed by ETH, earning Ethereum’s staking rate”) across chains.

Indeed, the root cause of the exploit wasn’t staking, restaking, lending, or crosschain risk (Aave, Eigen, or LayerZero can’t be blamed). The issue was a forged crosschain message where a 1-of-1 security setup trusted the release of 116,500 rsETH without a matching source-chain burn. One side said “release”; the other side never actually paid. This wasn’t LayerZero’s fault, but rather an improper DVN setup, which is Kelp’s responsibility (since it’s their own security stack; LayerZero remains an immutable, permissionless, censorship-resistant protocol).

The attacker funded their wallet through Tornado Cash roughly 10 hours before the hit, and Kelp paused rsETH contracts roughly 46 minutes later, which blocked two follow-up attempts of 40,000 rsETH each. Good emergency response, but too late for the first hole. Nobody wants to be left without a chair when the music stops. ETH’s price appreciation could make things worse when it comes to estimating bad debt across lending markets, of which Aave carried over $200M of borrowed ETH/WETH against impaired collateral at the time of the hack.

AI: Infra Lock-in, Capacity Arms Race

  • Amazon / Anthropic — Compute lock-in, not just funding: Amazon will invest $5B now and up to $20B more into Anthropic, while Anthropic committed to spend more than $100B over ten years on AWS technologies; Anthropic also said the deal secures up to 5GW of capacity for Claude training and deployment. This is a decade-long cloud capacity prepayment dressed as strategic investment. Bullish AWS, Trainium, power, data centers, and AI infra suppliers. Less bullish for “model-margin purity” if labs keep pre-committing massive spend before profitability is clean.
  • Google Cloud Next — Agents become the monetization layer: Google used Cloud Next to unify AI products under Gemini Enterprise, with AI agents at the center. It also announced eighth-generation TPUs, split between TPU 8t for training and TPU 8i for low-latency inference. Google is not trying to win only with model benchmarks. It is pushing the full stack: chip, cloud, data, security, workflow. That can be an underappreciated enterprise moat.
  • Merck / Google — Vertical AI is where budgets clear: Drug maker Merck will partner with Google Cloud and could invest up to $1B over several years into AI infrastructure, Gemini Enterprise licensing, and embedded Google Cloud engineers across drug research, regulatory work, manufacturing, and commercial operations. The market should reward AI platforms with domain integration, compliance posture, and implementation teams, not just raw model access. Google is getting ahead on that front.
  • OpenAI / GPT 5.5 — The worker model, not just the chatbot: What changed is less “better answers” and more execution: GPT‑5.5 is materially stronger at end‑to‑end coding, spreadsheets, slides, research, and computer-use workflows, scoring 82.7% on Terminal-Bench 2.0, 58.6% on SWE-Bench Pro, 84.9% on GDPval, 78.7% on OSWorld-Verified, and 98.0% on Tau2-bench Telecom.
  • OpenAI / Codex — Enterprise rollout is not implementation-led: Codex is becoming OpenAI’s enterprise distribution channel. OpenAI launched Codex Labs and partnerships with global systems integrators to scale Codex into engineering organizations, starting with workspace agents.
  • OpenAI / Codex — Workspace agents move ChatGPT into team automation: Business, Enterprise, Edu, and Teachers plans can now create Codex-powered shared agents that run in the cloud, work across tools, use memory/files/code, interact in ChatGPT and Slack, and stay inside org permissions.
  • OpenAI / Cyber — Government access as trust moat: OpenAI briefed U.S. federal agencies, state governments, and Five Eyes members on GPT-5.4-Cyber, a defensive cybersecurity model being rolled out first to vetted users. GPT‑5.5 ships with its strongest safeguards yet, after red-teaming and testing with nearly 200 trusted early-access partners, and is tightening cyber-risk controls while expanding trusted access for verified defensive users. OpenAI is explicitly classifying GPT‑5.5’s biology/chemistry and cybersecurity capability as High under its Preparedness Framework.
  • Anthropic / Mythos — Safety risk: Bloomberg reported how Anthropic’s most powerful model (so capable that it’s not available to the general public given cybersecurity risks) was accessed by unauthorized users through a third-party vendor environment. Mythos is part of Project Glasswing, a controlled preview for defensive cybersecurity, and has raised concern because of its ability to identify software vulnerabilities.
  • Anthropic — Productivity gains now come with job-risk anxiety: Anthropic’s 81k Claude-user survey shows the people closest to AI disruption are also the most worried about it: higher “observed exposure” to Claude-used tasks correlated with higher displacement concern, and early-career workers were more anxious than senior workers.
  • Anthropic — AI surplus is uneven: The highest- and lowest-paid occupations reported the largest productivity gains, with scope expansion and speed being the main drivers. High earners use AI to ship more complex work; lower-wage users often use it to jump into new tasks or side projects.
  • SpaceX / Cursor — Coding tools as strategic assets: SpaceX announced it has an option to buy Cursor for $60B later this year, or pay $10B for the partnership. The point is distribution to elite developers plus xAI/SpaceX compute through Colossus. If Cursor gets deeper compute access, Anthropic Claude Code, OpenAI Codex, Google Gemini Code, and Microsoft/GitHub all face a faster arms race.
  • DeepSeek — China AI gets a capital re-rate: Tencent and Alibaba are in talks to invest in DeepSeek at a valuation above $20B. Capital is clustering around national champions, and DeepSeek remains the symbol of China’s “do more with less compute” challenge to the U.S. AI dominance.

Signals Radar

BTC — Performing better than equities on the week driven by the same macro relief bid that helped equities. Strategy disclosed another 34,164 BTC purchase for $2.54 billion, and U.S. spot ETF flows stayed positive.

ETH — ETH keeps acting as dead money. It caught the same midweek risk bid as the rest of crypto, then gave it back fast as Hormuz/oil headlines hit risk assets again on Thursday. That tells you everything you need to know for trading the asset already.

SOL — SOL has been slightly better than ETH but still muted. The institutional-access story is still there with Grayscale’s GSOL trading on NYSE Arca, but price has not reacted like a market that believes a breakout is here; for now this is a hold-in, not a rip-higher.

HYPE — 21Shares added staking rewards to its Hyperliquid ETP and framed Hyperliquid as the dominant DEX for perpetuals, citing more than 50% of DEX perpetual open interest, roughly $8B in daily volume, and over $4T in cumulative volume. The same day Kalshi and Polymarket declared their intention to expand into perps, just as the CFTC moves toward clarity and exchanges are racing to launch perps for equities and commodities.

KERNEL — Pending an official declaration of next steps, which is likely to be delayed due to legal issues, Kelp published their incident response, stating that their 1-of-1 DVN setup was a recommendation by LayerZero Labs, whose RPCs failed and provoked the capital drain.

AAVE — Not an Aave smart-contract exploit. Still ugly. The attacker used rsETH as collateral and pulled WETH/ETH liquidity before risk controls caught up; Aave froze rsETH/wrsETH across deployments and later froze WETH on multiple instances as a precaution. The harsh read of Aave’s rsETH incident response is that the high LTV/eMode on bridged LRT collateral was a growth decision masquerading as risk efficiency (rsETH eMode at 93%, meaning only 7% overcollateralization in the event of collateral haircut).

Another line of debate is pushing rate limits, borrow throttles, credit-score-style systems, and isolated risk. The setback came just a week after Aave v4 went live after multiple key service providers like BGD and Chaos Labs left, and the protocol is now being forced to rethink whether “blue-chip LRT/LST” deserves anything close to ETH-like treatment.

Aave was not hacked but it was used as the exit door. Contract safety doesn't prevent balance-sheet damage and now on top of governance problems the protocol is forced to deal with a liquidation and bad debt problem, which presents two scenarios: ~$123.7M total bad debt under a uniform rsETH haircut, or ~$230.1M if losses are isolated to L2 rsETH, with WETH absorbing almost all of it. rsETH positions were opened with high LTV assumptions against collateral whose bridge backing could fail faster than liquidations could clear. WETH reserves on Ethereum, Arbitrum, Base, Linea, and Mantle were all at 100% utilization, with idle balances below $20 on every chain

MORPHO — Relative winner on architecture merit. Isolated markets did what they are supposed to do. As usual during exploits, Morpho Blue remains an immutable permissionless protocol, where curators manage risk on top of a 500-lines-of-code substrate. Asset managers with exposure to rsETH reacted quickly and only ~$1M was borrowed against the asset hours after the exploit (across two isolated markets). It’s in times like this when isolated markets shine versus rehypothecated shared pools, as risk doesn’t spread from one market to the next. Only 2 out of the hundreds of Morpho vaults were impacted.

EUL, FLUID, SPK — Each with their own peculiarities, all these lending markets were also hit by association. Euler got dragged into the contagion bucket, Fluid paused on time and had no material rsETH exposure on L2s, and SparkLend, as a fork of Aave v3, was forced to freeze and pause rsETH-linked markets.

ZRO — LayerZero’s whole v2 pitch was “immutable, neutral infra; apps own their security stack.” If a team wires that into a 1/1 multisig, that’s their decision—not a protocol bug—yet LayerZero actually ran this specific DVN (Decentralized Verifier Network) on their infrastructure. Kelp was affected because that was their single DVN, despite LayerZero’s docs recommending having multiple. Still, 47% of ~2,665 active LayerZero OApp contracts used 1-of-1 DVN security over the prior 90 days.

It was a sophisticated attack (LayerZero post-mortem). The bridge’s transaction verification RPC Nodes were replaced with fake ones while the legit ones were DDoS’d (Distributed Denial of Service — spamming traffic to trigger downtime) to get traffic through the malicious one. Nuance wouldn’t save price in the first wave, as ZRO dumped ~20% alongside AAVE as a result of collateral damage, yet long-term the thesis remains unchanged.

ENA — The Aave/Ethena loop has been a major growth engine for both protocols, with more than 50% of USDe-related assets deposited on Aave and users looping sUSDe/USDe/PT collateral against stablecoin borrows; now, reported full utilization and deeply negative loop economics force the opposite trade: repay, redeem, shrink supply (loopers have been seeing -50% APY). If the ~$1.4B unwind lands, USDe supply takes a ~25% air pocket. That is not fatal, but it kills the “only up” TVL story until new demand replaces forced loop demand.

XPL — Aave Core reportedly had $3.29B USDT deposits with only ~$400k available liquidity, while Plasma’s $1.35B USDT0 reserve had less than one cent of available liquidity, effectively 100% utilization. That is the practical death of the loop.

ARB — Arbitrum’s Security Council froze 30,766 ETH, worth about $71M, tied to the Kelp exploit and moved it into an intermediary wallet accessible only through further Arbitrum governance action. That is net-positive for victims and net-complicated for ARB as the chain showed it can stop a state-linked thief, but also reminded everyone there is an emergency lever.

TAO — The question hanging over everyone’s watchlist: does TAO deserve a valuation premium that’s heavily skewed to the speculative side, or does Bittensor actually have a moat in the AI sector? Justin Bons (Founder & CIO of Cyber Capital, Europe’s Oldest Crypto Fund) went on the offensive this week with a post on X directly calling TAO a “crypto-ponzi.” He claims TAO has no PMF and points to a $328M annual emissions bill versus roughly $15M in annual revenue. The bootstrapping mechanism works exactly like that, with an emissions-led and halving model capped at 21M and post-halving issuance around 3,600 TAO/day.

It was an opportune moment to add fuel to the fire, just one week after Covenant made the abstract risk tangible as they exited subnets 3, 39, and 81. Community miners had to revive them, and the proposed Conviction/Locked Stake fix (when you stake TAO into a subnet, you receive that subnet’s Alpha token, and that Alpha stake is what determines subnet-specific weight and emissions) is basically an admission that “team commitment” was not sufficiently priced or enforced onchain (a subnet could become systemically important, attract holders/miners/validators, and still depend on a team that could walk away faster than the market could reprice). Bull case is decentralized AI optionality; bear case is that, if emissions are the product, holders are the customer.

RAVE — Between ZachXBT offering a $10k bounty and CEXs like Bitget and Binance doing diligence on insider trading (reports saying nine initial-allocation addresses controlled ~95% of supply), RAVE’s 10,000% up move finally reverted >95%, wiping out $15B in market value, from $26 to $1.

WLFI — After last week’s incident (frozen tokens, disputed voting power, proposed lockups, and insider-controlled mechanics), Justin Sun filed a California federal lawsuit against World Liberty Financial, alleging WLFI froze his 4B tokens, blocked voting rights, and threatened to burn holdings.

CHIP — Token debuted trading ahead of hard proof. USD.AI now stands at ~$345M in deposits, 7.11% current APR, 13.52% expected APR, and a $236M active loan pipeline; DefiLlama shows about $283.9M TVL plus $60.6M in active loans and $10.37M in annualized protocol revenue, while USD.AI says it has already executed $225M of loans with $1.2B+ in approved facilities.

Spot and perps came fully loaded (Binance, Coinbase, Upbit, Robinhood, Hyperliquid, OKX and others), so price discovery was violent, from a $0.0558 ATL on April 22 to a $0.1384 ATH on April 23, or roughly $200M market cap / $1B FDV with 2B of 10B tokens circulating.

Team and investors have 0% unlocked before month 12, then 33% at month 12 and the remaining 67% linearly over the following 24 months. USD.AI’s own docs are explicit that CHIP is a governance/staking token, not equity, debt, or a claim on protocol assets, and while fees can be routed to staking rewards or buybacks, that is governance-dependent, not hard-coded.

Tempo — Honoring their branding as the “blockchain for payments at scale,” Tempo is partnering with enterprises to support stablecoin flows, starting with DoorDash, Stripe, Coastal, and ARQ. They also launched a stablecoin advisory unit with forward-deployed engineers and are working with DoorDash on stablecoin pay options for delivery employees.

USVC — Naval Ravikant, entrepreneur and angel investor best known for founding AngelList, promoted USVC on X, an AngelList-backed, SEC-registered closed-end fund that lets non-accredited U.S. investors buy venture exposure with a $500 minimum check size. They strategically invest through three channels: LP stakes in venture funds, growth rounds, and secondaries. The disclosed portfolio currently lists seven companies, including xAI, Anthropic, OpenAI, Crusoe, Sierra, Vercel, and Legora. So far only 44.34% of the capital has been deployed, with xAI the largest disclosed weight at 20.23%. The public conversation on X split fast between people calling it a long-overdue break of the accredited-investor wall and skeptics framing it as retail access only after the easy upside is gone (an attempt to offload late-stage private AI risk onto retail at peak narrative heat).

Prediction Markets — Both Kalshi and Polymarket are crossing into perps. Polymarket posted that Perpetual Contracts are coming with early-access open; Kalshi is reportedly preparing U.S. crypto perpetual futures and has also teased a “Timeless” April 27 New York reveal. The why is obvious: fixed-expiry binary contracts cap liquidity, force churn, and make size hard; perps remove the expiration cliff, add margin, improve market-making, and turn sports, politics, macro, crypto, and private-company valuations into 24/7 leverage products. The opportunity is Robinhood + CME + Hyperliquid for events.

Conclusion

The market keeps rewarding simplicity. Don’t try to outsmart it. Own what has marginal buyers around, avoid what’s surrounded by tail risks. If something doesn't move on good news, it may be a signal to flatten the position.

Alea Research · 25 Apr 2026 · 20 min

Back to top