Skip to content

Pulse - Mar 7

7 Mar 202620 min
ContentsOverview

Oil is a tax; the Fed still can’t blink. War premium, AI multiple compression, and no clear bullish market structure just yet. Another “survive first” week where you prioritize staying liquid and cutting dead money fast.

Key Takeaways

  • Oil premium = inflation risk; don’t front‑run cuts until expectations break.
  • BTC isn’t a war hedge; wait to add size on Fed pivot confirmation.
  • Cashflow tokens with real buy pressure are the only consistent bid.
  • AI selloff is duration unwind; avoid long‑duration multiples without cashflow proof.
  • Alts are exit liquidity unless you can name the marginal buyer.
  • Time is risk: no move = no trade.

Macro: Passive Liquidity Tax

The shock: the U.S. and Israel launched strikes on Iran on Feb 28, and the conflict is still escalating. Iran’s Supreme Leader Ali Khamenei was killed in the strikes and follow-on strikes/retaliation are ongoing. Multiple oil & gas majors / traders and tanker operators have suspended shipments through the Strait of Hormuz after warnings and disruptions, with maritime traffic visibly impaired. This matters because Hormuz handles ~20% of global oil flows (and meaningful LNG volumes). The market response now centers on “oil risk premium.”

Inflation risk is back at the worst time. U.S. PPI (Jan 2026) printed +0.5% m/m, and core pressures in services were already sticky. Now layer an energy shock on top, reducing the odds of Fed cuts in 2026. Oil shock risk + risk-off flows are immediate; rate expectations become a tug-of-war between “growth scare” (cuts) and “inflation via energy” (no cuts). War = oil up = inflation up, which can handcuff the Fed even if growth weakens. If inflation expectations rip, the Fed can’t “save” risk assets quickly.

This is also week one after markets shifted from a when to an if regime for cash flow durability. The viral Citrini “2028 Global Intelligence Crisis” fan fiction scenario catalyzed a violent software selloff, pushing investors from a predictable “growth annuity” SaaS paradigm into radical uncertainty about AI disruption to future cash flows. P/E and revenue multiples were crushed despite the lack of real macro data on AI.

Meanwhile, AI capex anxiety is rising. Last week, Nvidia printed massive numbers, but the stock still dipped as investors question “how long can the spend cycle last.” On top of that, the U.S. Treasury released an AI Lexicon plus a Financial Services AI Risk Management Framework (FS AI RMF), effectively trying to standardize language + controls for AI use in finance.

To top things off, macro came back to bullying crypto. The Iran shock makes the key point bluntly: bitcoin is not trading as a safe haven in these episodes (it fell on the conflict headline and is down hard over the last ~2 months). Next up: Mar 11 CPI, Mar 17-18 FOMC.

Crypto: Liquid Idiosyncratic

Now that everyone is on the same page—that the “free money” days are gone—the time has come for funds and individuals to take a bold stance. Most are getting to the same conclusion: this is a market for liquid concentrated bets. Market structure is dominated by a few VC bets sitting on >100x gains while most tokens lack marginal buyers. In this environment, the only defensible bets are protocols with durable revenue, real moats, and equity-like token value accrual.

Given the barrier to entry to launching a token is so low, liquidity is not a moat. For projects to succeed, they need something structurally defensible, like a unique product, distribution, or integration edges. This is one of the reasons why most buyback programs seem flawed. Unlike equities, crypto’s absence of marginal buyers means that buybacks mostly make sense only after a project has exhausted higher ROI investments. Otherwise, not only will prices barely react, but the project will end up misallocating capital buying their own issuance at ever-depreciating prices.

AI: SaaS Isn’t an Annuity

Michael Burry came out writing about AI, claiming that LLMs prioritize language mimicry without genuine reasoning, implying diminishing returns from brute-force scaling and a risk that markets are overpaying for a path that may not deliver “true intelligence.” However, the market doesn’t need AGI for this to be a huge business. Coding + agent workflows are already monetizing. Bubble risk is more about capex + valuation stacking faster than defensible cash flows. That’s where the crash lives, and not in whether models can “think.”

The reality is, we are about to enter the SpaceX + OpenAI + Anthropic IPO era. SpaceX is reportedly considering a confidential IPO filing as soon as March 2026, with talk of a valuation >$1.75T; Anthropic has been discussed as potentially prepared to list in 2026; and OpenAI just raised the largest round in history for ~$110B.

Two things have changed versus prior mega‑IPOs: i) index providers are rewriting rules to pull big IPOs in faster, and ii) passive scale is now big enough that even “small” index weights become enormous dollar flows. The latest Nasdaq rules dictate that the top‑40 by total market cap can be added after 15 trading days with ≥5 days’ notice, and QQQ alone is ~$390–$410B AUM. This “AUM × weight = forced dollars” math is why index changes can create real price events. For reference, Tesla’s S&P 500 addition forced index funds to buy about $85B of stock in 2020, with equal selling across other constituents.

  • Anthropic’s Pentagon stance turned into a growth hack: Claude ranks number #1 in the iOS App Store hours after the Pentagon's ban and supply-chain risk label.
  • Claude rolled out Import Memory: a 2‑step copy/paste flow to move your preferences/context from other assistants into Claude’s memory settings with a pre-written export prompt.
  • Anthropic is now a revenue-scale contender: Bloomberg reports its run‑rate revenue recently surpassed $19B, putting it within range of OpenAI’s last‑disclosed ~$20B.
  • GPT 5.4 launched, introduced as a materially more capable, more tool- and computer-use–native frontier model that justifies higher per‑token cost via better reasoning, coding, and token efficiency for serious agentic/professional workflows.
  • Altman says ChatGPT is back to >10% MoM growth, an updated “Chat” model is imminent, and the company is circling $100B funding.
  • We are entering the post-benchmarks era: “Opus 4.6 vs Codex 5.3” framing shifts from leaderboards to real workflow feel + consistent testing.
  • ChatGPT backlash hit hard: U.S. mobile uninstalls +295% on Feb 28 after OpenAI’s Pentagon/“Department of War” deal; U.S. downloads -13% same day; 1‑star reviews +775%.
  • OpenAI walked it back (fast): Altman says the deal is being amended so OpenAI services won’t be used by intelligence agencies (e.g., NSA) unless there’s a separate follow-on contract modification.
  • Reddit claims 80M weekly search users after merging core search with AI‑powered “Reddit Answers,” pulling queries off Google.
  • Google pushes back on “AI killed search,” claiming search revenue growth accelerated every quarter of 2025 despite AI summaries.
  • Google keeps attacking on cost/latency: Gemini 3.1 Flash‑Lite preview is $0.25/M input + $1.50/M output, adds “thinking levels,” and claims 2.5× faster time‑to‑first‑token vs 2.5 Flash plus 1432 Elo on Arena.
  • Qualcomm chips kick off the 2026 AI wearables season, launching its new Snapdragon Wear Elite platform designed to power AI wearables, including pins, watches, earbuds, pendants, and glasses.
  • In a rebuttal to Citrini’s 2028 Global Intelligence Crisis report, Citadel pointed out that software engineering job postings are up 11% YoY.
  • Cursor isn’t slowing down. They reportedly crossed $2B annualized revenue run-rate, doubling in ~3 months, with ~60% of revenue from enterprise.
  • Nous Research ships “Hermes Agent,” its open-source (MIT) persistent agent you install on a machine, then access across Telegram/Discord/Slack/WhatsApp + CLI, with cross-session memory and a “skills” system that records solutions so it improves over time.
  • Beyond agents, there is a regime-change in software production: production code volume outpaces human attention; winners build trust/accountability/control primitives early.
  • Hardware gets eyes: AirPods Pro 4 rumor adds cameras to “see around you” for gestures + enhanced spatial audio, with ambient sensing as the next UI surface.
  • Consumer AI M&A stays hot: MyFitnessPal acquired Cal AI; Cal AI claims 15M+ downloads and $30M+ annual revenue in <2 years; app remains independent but plugs into MyFitnessPal’s 20M‑food database.
  • SaaSpocalypse: Coding agents flip “build vs buy” toward build, break per‑seat pricing, and put venture‑backed SaaS IPOs on ice until a new pricing model stabilizes.
  • Apple (local LLM bet): M5 Pro/Max marketing is explicit about LLM throughput (“higher token generation for LLMs”), with neural accelerators in each GPU core and “>4×” peak GPU AI compute vs prior gen.
  • Donald Knuth says Claude Opus 4.6 cracked his long‑running Hamiltonian‑cycle conjecture for all odd sizes.
  • Ex‑OpenAI research chief Bob McGrew is raising $70M for Arda to train AI on factory-floor footage (manufacturing automation).
  • Unity’s CEO teased an AI beta that can generate full casual games from plain English prompts.
  • NotebookLM added “cinematic” video overviews.
  • Google dropped an open-source Workspace CLI with 40+ agent skills across Drive/Gmail/Calendar.
  • xAI committed 1.2GW for AI data centers, expanding what’s described as the world’s largest Megapack deployment.
  • Broadcom printed record Q1 revenue ($19.3B); AI revenue doubled ($8.4B); projecting AI chips >$100B by 2027.
  • Barclays estimate: humanoids + AVs + industrial automation could be a $1.4T market by 2035.

AI will commoditize “algorithm-thinking” (fast pattern recognition, legible metrics work); durable human edges (storythinking + embodied judgment + responsibility + presence) don’t score well on dashboards. Obsolescence fear is rational. The AI timelines being discussed by the people building frontier systems (exceed Nobel-level ability across many fields and operate at massive scale) are aggressive (AGI by 2027). Every time models get close to “human level” on a benchmark, the benchmark evolves to expose what wasn’t measured (generalization under novelty, changing rules, long-horizon interaction, etc). The takeaway isn’t “benchmarks are fake.” It’s: benchmarks create a treadmill, and the important question is what remains after the treadmill resets.

Indeed, a lot of what modern knowledge work values is getting commoditized. AI eats your edge in: information access, synthesis, pattern recognition, and analysis; humans still get paid for: problem selection (asking the right question), narrative modeling under uncertainty, accountability (owning outcomes), trust, and distribution (people still choose who they trust in high-stakes settings).

Signals Radar

BTC — Preferred long for convexity (options; or perps on confirmation). War headline risk is live, add size after the Fed materially pivots (cut, emergency liquidity language, or unmistakably dovish guidance; by March 17-18 FOMC), and set close stops near ~$67-65k. Risks: oil-led inflation blocks cuts; conflict de-escalates fast (vol crush); correlation spike → everything sells first. The edge isn’t “war = buy BTC.” It’s “war increases odds of the Fed put; buy when the Fed put shows up.”

HYPE — Where price discovery happens when TradFi sleeps, with Bloomberg now citing Hyperliquid’s oil perps. This weekend’s catalyst was Middle East escalation risk: traders pushed into oil / gold / silver perps on Hyperliquid to hedge and trade the headline while waiting for Monday’s reopen. Indeed, Hyperliquid is the canonical “good buyback” case: extremely high revenue, ultra-lean team (~11 people), continued product expansion (HIP3, HIP4) and no crowding out of growth investments; buybacks are funded by durable cash flow, not treasury desperation.

AAVE — After BGD, ACI (Marc Zeller) exits Aave, not seeking renewal and winding down over four months while continuing governance activity and executing outstanding commitments during the handoff. Zeller frames the exit as a governance failure: ACI spent years building “accountability” norms (drove 61% of governance actions, built revenue strategies tied to 48% of protocol income, deployed $101M in incentives), but when those standards were applied to the largest budget request in DAO history (route 100% of Aave-branded product revenue to the DAO treasury, ratify V4 as the future foundation, solve brand protection, and set a DAO-funded budget for continued development), the process “stopped working.”

MORPHO — Amidst a governance crisis in its most direct competitor, the “crypto mullet” thesis—front-end remains TradFi (institutions, RWAs), back-end is DeFi infra—stands out as one of the few opportunities with a credible chance to be repriced as critical infra businesses. Liquid tokens in this category are among the few that could still deliver venture-scale returns.

ETHFI (Alea Benchmark: Q1 2026) — Cash is now the main character: protocol revenue runs ~$60M annualized (Cash ~55%; Stake ~39%) and has held ~$1.0–$1.2M/week since Q4 ’25, while spend is ~$663M annualized (~$1.8M/day) with ~8k daily active cards (+29% QTD) and ~54k cards issued—still mostly referral-led ahead of paid acquisition tests in Q1‑26 (target $50–$100 CAC). The token hook: ~$19M annualized buybacks (cumulative ~$16M, ~3.5% of supply); they’re a forced buyer and sETHFI stays engineered around ~10% yield.

ETHFI is -35% QTD and FDV multiples are at cycle lows (TVL/FDV ~14; P/S ~7) with investor unlocks ending March ’26 and new surfaces (Hyperliquid Trade, tokenized stocks, EUR/SEPA rails) queued to add fee lines beyond ETH beta. Breaks if Cash growth mean-reverts once CAC grows, or if incentives/emissions swamp net buybacks.

SYRUP — The Q1 tokenholder update was a “distribution + buyback engine” pitch: overall AUM is down to ~$3.8B (‑23% vs Oct ’25 peak $4.97B), but the only thing that matters for durability—syrupUSDC/USDT flows— is still sticky (+$17M net in Feb; +$325M YTD). The drag is utilization: outstanding loans ~ $850M, roughly ‑50% YTD, which is why trailing APYs are at lows even if the “loop it on Aave” spread still exists.

Next catalysts are non-crypto rails: team flagged an imminent exchange-earn program and a goal to sign 3 fintech partners by end‑Q1, targeting $500M–$1B of fintech-sourced capital by year-end; plus an onchain securitization vehicle “in the coming months” and in-house trading strategies to smooth yield. Token linkage stays clean: 25% of protocol revenue routes monthly to the Syrup Strategic Fund for buybacks/treasury (Q4 SSF bought back ~2M SYRUP, with Dec revenue ~$2.5M).

ENA — Still silently scaling distribution: February shipped suiUSDe on Sui, pushed Ethena Whitelabel stablecoins to $130M+ across MegaETH/Jupiter/Sui, and took USDe cross‑chain supply to $1.1B+ (from ~$120M in May), with USDe‑native perps now >$150M.

PUMP — Introducing “Pump launches,” letting users trade a broader set of tokens inside the app (incl. WBTC, USDC, ETH/WETH via Wormhole, plus other Solana tokens and even assets from other launchpads). They’re trying to become a one-stop Solana trading surface, similar to Jupiter’s vertical stack integration, reducing user churn to DEXs, and capturing more session time + flow after tokens “graduate.”

DRIFT — Roling out “Drift Guild” as a permissionless contributor program to earn DRIFT rewards and “work your way into Drift’s inner circle.” The framing is classic gamified BD: contributors stack XP via work (content, onboarding, ecosystem support) and climb tiers (incl. an “Ambassador” pathway per the initial comms), with monthly DRIFT rewards. The upside is sticky community labor + cheaper growth loops; the risk is mercenary farming and incentives that don’t convert to durable volume/fees. Watch whether this becomes an acquisition funnel for real makers/affiliates vs. a rewards treadmill.

LIT — Fee Credits (MM/HFT-focused utility) were introduced, letting premium accounts pay in LIT for credits that count toward higher discount/latency tiers without staking the full amount upfront. This is a direct “token utility” lever aimed at market makers: bullish for LIT if it meaningfully improves depth/volume; bearish if it’s just fee rebates with no durable venue pull.

CHIP — USD.ai ICO ends with $19.4M raised. From here it’s pure market structure: unlocks/vesting + listing venues + whether “AI infra / AI interest rate” actually becomes recurring demand instead of one-off fundraising heat.

POL — Polygon’s “deflation” story relies on durable payment flow. USDC-heavy stablecoin rails are pushing adoption to new weekly highs after an explicit pivot to an end-to-end payments stack (“Open Money Stack”: wallets + orchestration + on/off-ramps + chain), and “agentic payments” (x402 / machines paying machines) as the high–tx count wedge where tiny fees can actually compound.

Token math: net supply = emissions + incentives − (fee burn + buybacks), and governance is already pointing at the structural leak (the ~2% inflation schedule) with a buyback/burn proposal. If emissions get forced down while stablecoin settlement volume sticks, POL’s scarcity stops being aspirational and becomes a weekly datapoint. A clean relative-value angle is long POL / short OP while “stack rent” narratives crack (Base stepping away is a direct stress test). Invalidations: USDC activity peaks then mean-reverts, routing migrates to stablecoin-native chains (Tempo/Plasma) or Stripe/Bridge-style vertical stacks, Polymarket-driven “usage optics” fades, or tokenomics changes stall (if burn/buybacks stay de minimis vs emissions, the market will keep pricing POL like a low-fee pipe with no value capture).

TAO — Bittensor is trying to be a credibly‑neutral market for “intelligence” (AI outputs / services), where subnets are specialized AI businesses, and TAO is the base asset that funds and governs the economy. TAO is the reserve asset and behaves like an index on the entire subnet economy, while subnets are the venture layer (many “AI startups”). That’s the bull pitch: TAO is “Bitcoin‑scarcity-ish” with an embedded AI venture economy. It’s one of the few “AI crypto” designs where token demand is mechanically tied to new “apps.” The primary way to get subnet tokens exposure is by staking TAO, not by paying USDC on DEXs.

GMX — The DAO approved a “restore price discovery” plan aimed at neutralizing the CEX supply overhang that’s muted buyback impact. Effective Mar 4, 2026, staking rewards stop being paid out and instead accumulate in the Treasury (they note 27% of protocol fees flows there) until GMX > $90, at which point rewards get distributed pro‑rata by “staking power” (amount * duration) with a hard “80% of peak stake” rule (dip below it and you permanently forfeit accumulated rewards). The Treasury will pull ~600k GMX from external LP venues (Uniswap/Trader Joe) and redeploy into GMX-native liquidity (GMX‑USDC + GMTrade.xyz) targeting ~2% of market cap in onchain depth, plus a one‑week, 1,000,000 GMX buy‑wall at $5, then a weekly deposit‑buy‑withdraw cadence to keep Treasury assets off exchanges between executions.

META — Futardio permissionless launchpad live; bullish for Solana’s “onchain capital formation” stack, but permissionless launchpads are magnets for junk flow.

NEAR — The Feb 23–24 NEARCON SF launch of near.com was NEAR’s first attempt to exit the “mid-cap L1 beta” bucket and reprice as a wallet UX + cross-chain routing business + confidential execution layer (TEE/private-shards roadmap) + AI “agents” interface. The bet is orderflow: win the front-end, or embed as the back-end inside other wallets (Ledger + SwapKit is the template) and negotiate economics with liquidity. What to watch: routing share and take-rate (Intents volume + channel fees).

RPL — Saturn I / Saturn One (live Feb 18, 2026) turns RPL into an ETH‑denominated fee‑switch yield curve. Vote‑eligible (staked) RPL in megapools earns ETH sourced from a slice of protocol commission on borrowed ETH, and the payout is: (megapool validator count × commission stream × voter share) / (total vote‑eligible RPL × RPL price); returns rise with megapool adoption and fall with crowding and RPL repricing. What to watch: i) megapool validator growth, ii) total vote‑eligible RPL (capped via min(1.5× RPL value of bonded ETH, staked RPL)), and iii) whether governance keeps the initial 5% node‑operator commission share / 9% voter share economics intact.

DIME — Paradex’s native token went live this week on Thursday; 25% of total supply allocated to the community airdrop pro-rata XP-based; protocol revenue buybacks, gas coverage, and staking benefits.

HPL — Hyperlend airdrop took place on Friday; liquidity paired with Native Market’s USDH; eligibility criteria based on supply/borrow.

OPN — Binance launchpool and spot listing with initial circulating supply listed as 198.5M OPN out of 1B. total

INK — Kraken becomes the first exchange to gain direct Fed payment access via a Fed master account for its bank entity. This reduces reliance on sponsor banks and improves fiat settlement loops, which is meaningful for on/off-ramp resilience and a regulatory signal that “regulated crypto banking” is being re-opened.

Conclusion

You can be right on macro and still get chopped to death on timing. The Fed wants to ease but the market doesn’t give it cover yet. Risk-taking is getting punished. Keep ammo, stay in the few names with repeatable buy pressure, and don’t average losers. If the Fed pivots, add BTC convexity; if oil keeps inflation expectations pinned, cut beta and sit tight. Some weeks aren’t about being right, but about not getting chopped to death. Respect the oil premium, respect the Fed constraint, treat everything illiquid as optionality, not allocation.

Alea Research · 7 Mar 2026 · 20 min

Back to top