One of those “aipocalyptic” weeks where an article—the Citrini “AI doom” thesis—drops and moves markets. That speaks volumes about how capital allocators are positioning. This shows how most funds aren’t underwriting a thesis they have strong long-term conviction in. Instead, they are playing catch-up by quickly trading headlines. In this environment, capital is set up to react, not to ride out volatility on well‑defined, long-term underwriting.
Key Takeaways
- Weekly recurring theme: the widening gap between productivity/asset owners and labor.
- Narrative shock > data: Markets are hypersensitive to AI doom memes; almost $1T got repriced off a thought experiment.
- Citrini’s “AI doom” thesis: AI will strip out the human friction which currently justifies large layers of intermediaries, fees, headcount, and software moats in the modern services economy.
- Fund flows are reactive: Most capital is trading headlines, not underwriting durable AI theses.
Macro: When Stories Beat Numbers
The theme of the week: essays causing equities to take a hit amid a narrative shock. Almost $1T in market value erased as “AI takes over the world” becomes consensus and, therefore, the “obvious” short is crowded. It was a viral scenario, not even a prediction, that simply modeled a nasty AI-driven loop (white-collar layoffs → weakened demand → more automation). That alone was enough to spook most investors on Monday’s opening.

Citrini’s doomsday scenario, presented in his latest piece, The 2028 Global Intelligence Crisis, erased ~$800 billion in market cap. At this point, the view that AI is taking over the world is becoming consensus. Unsurprisingly, such a doomsday scenario went viral.

The “SaaSpocalypse” thesis, however, assumes fixed demand. Historically, however, collapses in production cost (compute, internet, cloud) have expanded total consumption and birthed new markets. But AI is a different beast: solve AI first; then solve everything else (healthcare, legal, marketing, coding, etc).

Since the real macro shock is service price compression across the ~80% of GDP that is services, “abundance GDP” becomes the key mental model: the bull case is not soaring nominal incomes, but sustained productivity gains plus falling prices in essential services, transmitting AI gains to households via lower living costs rather than wages alone.

AI: Consensus Long, Underwritten Badly
The 2028 Global Intelligence Crisis, by Citrini Research, went viral this week. Rather than making a prediction, it presents a thought exercise about what happens if the AI bulls are right … and that ends up being bearish for the economy. TLDR: abundant machine intelligence drives a reflexive loop of layoffs, leading to weaker consumption, compressing margins further, causing more AI spend, and again triggering more layoffs.

Three stages: i) 2026-early 2027 is when always-on LLMs and consumer agents go mainstream; human relationships as moats get repriced as a result, ii) 2027 is where disintermediation breaks hardest where the fee is obvious (payments) as well as the potential year when white-collar displacement kicks in, iii) and 2028 is when housing becomes existential.

What Citrini’s scenario is saying: i) agentic coding nukes SaaS pricing power (“build it in-house” becomes credible); ii) “friction goes to zero” as always‑on agents cancel subscriptions, negotiate renewals, price‑match, and route around intermediaries; iii) payments take a hit when machine‑to‑machine commerce becomes 2-3% of card interchange (stablecoins via Solana or Ethereum L2s for near‑instant, cheap settlement is presented as a scenario); iv) credit contagion as ARR-backed software loans default; and v) mortgage stress becomes the systemic accelerant when even prime borrowers become questionable because the income assumptions under underwriting no longer hold.
- Software engineering represents half the AI agent market. The rest is wide open across back-office automation, marketing and copywriting, sales and CRM, finance and accounting, etc.

- METR’s capability assessments show that Claude’s agentic capabilities are already far ahead of how founders/builders are deploying them; the real edge is in vertically specialized, high‑trust workflows where you can safely let agents run much longer and deeper.

- LessWrong is openly saying “AGI is here” (specifically calling out Opus 4.6 + GPT‑5.3 as broadly human-capable).
- Altman: his “inside view” is “faster takeoff than I originally thought.”
- Altman again: ChatGPT is “probably” now more energy-efficient than humans at answering questions.
- Anthropic released Claude Code Security (automated vulnerability scanning for codebases).
- Anthropic identified industrial-scale distillation attacks on its models by DeepSeek, Moonshot AI, and MiniMax.
- Anthropic acquired Vercept to accelerate Claude's “computer-use” / agentic capabilities.
- Anthropic is rolling back its original “safety-first” pledge: it won’t unilaterally slow frontier deployment absent regulation.
- The Pentagon threatens to cut off Anthropic’s Claude. The dispute escalated around the military’s reported use of Claude (via Palantir) in an operation to capture Venezuela’s Nicolás Maduro.
- Anthropic DoD contract is up to ~$200M, and Claude is described as the only model currently deployed in classified systems, making replacement hard in the near term.
- Trump ordered federal agencies to cut off Anthropic; Secretary of Defense Pete Hegseth labels Anthropic a “supply-chain risk”; Anthropic says it will challenge the designation in court and argues the legal scope should be limited to DoD contract work, not broad commercial bans; Pentagon simultaneously shifts to OpenAI; OpenAI says DoD accepted similar “red lines” and they wrote them into the deal.

- Perplexity Computer goes live, unifying every current AI capability into one system for research, design, code, and managing any project end-to-end
- Google releases Nano Banana 2.
- Gemini 3.1 Pro solved a FrontierMath Tier 4 problem no model had solved.
- OpenAI closes $110B raise from Amazon, NVIDIA, SoftBank in largest startup fundraise in history @ $840B post-money
- OpenAI reportedly has ~200 people building AI devices; first product described as a $200–$300 camera-equipped smart speaker.
- Square announced a mass layoff of “over 4,000” employees
- NVDA earnings beat consensus expectations estimated around $66.2B and adjusted EPS $1.53; NVDA printed $68.1B and $1.62 non‑GAAP EPS. Price remained fairly muted after hours, as the market had already positioned itself for a “beat and raise”.
“Rents and frictions to zero” is structurally bullish for consumers and new entrants but structurally bearish for legacy intermediaries: SaaS, two-sided marketplaces, credit-card economics, real-estate agents, and any business built on lock-in, opacity, or inertia are implicitly short ubiquitous agents that arbitrage them away. At the same time, if everyone has powerful agents, then those same agents arbitrage AI providers too: you cannot assume stable, high-take-rate “AI tollbooths.”
Agent competition will compress margins unless a provider has genuine, durable capability or distribution moats. That’s why ubiquitous agents destroy not just rent extraction but also many mechanism designs that relied on friction (credit-card chargebacks, loyalty programs, signup bonuses, sloppy underwriting, “gotcha” fees). This forces a rapid redesign of credit, reputation, and fraud systems. It’s the reason why macro “ghost productivity” (high output, low employment, weak demand) is politically unstable in democracies. History and events like Covid suggest that if unemployment spikes, large-scale fiscal and monetary response is the default. Thus, the relevant question is: what gets taxed and who owns AI.

Any scenario where superhuman agents run most of the economy yet humans retain full control for years is inconsistent; the real tail risk is not a slow deflationary spiral but a rapid shift of de facto decision power from states and firms to AI systems and labs, making alignment, governance, and “who is actually in charge” first-order over any specific macro path.

This week, the sharp reaction to a single speculative AI piece reveals the persistent under-modeling of AI second-order effects by institutional capital; the alpha isn’t the specific story, but the meta: narratives can move large pools of capital when the underlying causal map is still foggy.
Signals Radar
BTC — Capital is still hiding, as we only see selective nibbling while U.S. ETFs keep seeing outflows.

BTC is hovering around its 200MA level where the last cycle low formed, with RSI at near extreme historical readings.

ETH — Trading near prior-cycle cost basis around April 2021 levels, network’s fee economics look weak, L2s “no longer make sense,” and the core value proposition is being contested.

BNB — Introduces the zBNB privacy layer for confidential transactions, with ZK technology to decouple transaction history from wallet addresses through a burn-mint mechanism. It’s compatible with standard wallet interfaces.

SOL — Chain fees are starting to notice the impact of decreasing memecoin activity. Memecoins DEX trading fell to ~30% of Solana DEX trading volume, down from 60% in January.

AAVE (Alea Memo: Lending Monopoly, Crisis Multiple) — The current governance narrative is pushing toward a more aggressive framework (“Aave will win”) that’s meant to improve competitiveness and token-holder alignment, but the risk is non-trivial. BGD Labs (core builder) is leaving, citing governance tensions and disagreements around focus on Aave v4 vs v3. This is an execution hazard for such a relevant “upgrade cycle.”

Aave Labs proposes ratifying Aave v4, directing 100% of Aave‑branded product revenue to the DAO, and creating a funding framework to execute that roadmap. In response, ACI’s Marc Zeller published an onchain “audit” arguing Labs has already received ~$86M in capitalization ($16.2M ICO + $32.5M VC + $31.93M DAO payments + ~$5.5M in unapproved swap fees) while the founding team retained 23% of the original LEND supply (migrated 100:1 to AAVE), yet still provides no wallet disclosure or cost‑per‑outcome accountability. He also flags Horizon as a “headline TVL vs thin real RWA usage” case with roughly ~$216K of revenue against millions in incentives/carry costs. This is an overhang on AAVE, even if the end‑state (revenue alignment + brand protection) is directionally bullish.

UNI — Governance already passed UNIfication, which included a 100M UNI burn and turned on protocol fees on mainnet (and a burn mechanism that converts collected fees into UNI burns via a fixed‑UNI auction mechanic). That’s the structural bid UNI didn’t have for years, but that’s relatively considered “news of the past.” What’s most recent, however, is the Feb 18 proposal (Snapshot Feb 18–23) to expand protocol fees to eight more chains and to enable fees across all v3 pools via a new open adapter.

FLUID — With Aave’s latest mess in mind, Fluid launched a governance proposal to create the Fluid Foundation, a Cayman Islands non-profit that would own all IP like smart contracts, domains, and trademarks, transferring control from founders to FLUID token holders via voting.

This—becoming a DAO-owned protocol—is starting to look like the new norm: launch a non-profit Foundation with no owners and that owns all IP, allowing token holders to retain ultimate authority.

APT (Alea Memo: Use It Or Lose It) — Aptos is trying to brute‑force a re‑rate by cutting staking emissions ~50% (5.19% → 2.6%), raising gas 10x (still basically free in USD; a stablecoin transfer ~$0.00014), and burning all fees so usage translates into supply reduction, while tightening the long‑term supply via a proposed 2.1B hard cap (vs ~1.196B issued), a 210M APT foundation lock “never sold,” milestone‑gated grants, and a post‑Oct 2026 unlock roll‑off that reduces annualized unlock pressure ~60% (plus lower grant distributions). The trade is path‑dependent: none of this matters if activity doesn’t show up—today fee burn is negligible (~$500/day), and even 10x fees won’t move the needle without a real throughput app; Decibel Trade is the fulcrum because perps/spot/margin are where fees and burns scale, and a Decibel “points rush” could pull mercenary volume fast enough to create a reflexive, short‑lived deflation window. Expressed as a pair trade, a SUI short strips out market and sentiment beta, improving the signal-to-noise.

ZRO (Alea Memo: One Token, Many Flows) — Bullish long-dated; near-term mostly noise. Their new chain, Zero, is aimed at “institutional finance,” but it’s targeted for fall 2026. The announcement includes collaboration names like Citadel Securities, DTCC, and ICE, plus a strategic investment in ZRO. The LayerZero Foundation has been explicit about ZRO buybacks / fee-switch mechanics, but the market will only care as we get closer to an imminent “on.”
POL — Leaning toward a bullish setup if BTC behaves and we start seeing flows. Right now, there isn’t a clean weekly catalyst, but there is a credible path for a deflationary case, with a sweet combination of staking yield denominated on an asset that is burned with increasing activity. Polymarket and stablecoin volumes solidify this case. Reclaiming $0.13–0.15 is the first “trend might be back” signal. Continued sub‑$0.10, however, may signal “dead money.” Time-stop it if it doesn’t reclaim lost levels in 2–4 weeks.

OP — Unlike Polygon, the default stance is bearish until the Superchain economics re-clarify. The bull case hinges on 50% of Superchain revenue allocated for OP buybacks. The problem is the timing, since Base is moving away to build their own stack. This forces the market to question whether the “Superchain tax” model holds if the biggest revenue contributor optimizes away from it. At ~$0.12, OP is trading like the market is discounting that uncertainty. There is an offsetting bright side though, given that EtherFi’s Cash is migrating to OP Mainnet with ~70k active cards and ~300k accounts.

ETHFI — Partners with FalconX, M11 Credit, Cap, and Symbiotic to use 7,500 wstETH as collateral for institutional stablecoin loans, still earning ~4.5% in real yield via restaking-backed credit markets. This is a rare “consumer DeFi” product with a strong institutional angle that makes it unique versus other DeFi projects. Cash is doing real activity (thousands of swaps + spend tx/day) and is migrating the Cash/card stack to OP Mainnet over the coming months.

RPL — The Saturn One upgrade brings a bullish tokenomics angle (fee switch), but it needs adoption. Saturn introduces MEGAPOOL validators with 4 ETH bonds, adds a withdrawal buffer for rETH stability, and (most importantly for tokenholders) activates the RPL fee switch. RPL stakers earn a share of protocol commission paid in ETH, with inflation rewards set to phase out later in the year.

JTO — Given the denomination in SOL, TVL is hugging ~$1B, almost -50% MoM and down 70% from a ~$3.9B peak in January 2025. It’s fair to say that speculative activity compression hit Jito right in the face. Even if Solana is “fine,” the specific profit center Jito rode during meme mania got weaker. Besides, staking economics got worse at the margin (this matters for LST TVL). Solana has no in-protocol way to programmatically share priority fees with delegated stakers today; therefore, as priority fees become a bigger slice of rewards, it mostly hurts stakers. Lower/less-attractive yield means less sticky TVL.

NEAR — Launches NEAR.com Super App for AI and confidential transactions on this week’s NearCon held in San Francisco, where it introduced a renewed vision. It’s positioning itself as a single-account, multichain DeFi hub: swap, earn yield, and “go confidential” across 35+ networks from one Near account. The product surface: i) swap: market makers compete to fill orders across 35+ networks; ii) earn: deposit idle assets for yield with no lockups or commitments; iii) confidential: deposits, transfers, and swaps stay confidential, only visible to you, and whoever you choose; and iv) private deals (P2P): create terms + share a link with execution enforced by smart contracts, no intermediaries required. Under the hood, the “Super App” is a front-end to NEAR’s chain-abstraction stack, powered by the NEAR Protocol (base L1), NEAR Intents (engine behind every swap), and NEAR AI (privacy-preserving AI layer).

LUNA — Terraform’s wind‑down administrator sued Jane Street in Manhattan federal court, alleging insider-info-driven trading around the May 2022 UST depeg. Specifically, they claim that Jane Street got advance visibility into Terraform’s liquidity moves via a backchannel tied to a former Terraform intern (“Bryce’s Secret”), then positioned to unwind exposure “mere hours” before the system broke. Jane Street denies it and calls it a blame-shift / money grab.

CRCL — Crypto stocks show that they can still outperform tokens, especially when supported by positive earning results: $770M total revenue & reserve income (+77% YoY); $75.3B USDC in circulation (+72% YoY); $11.9T USDC onchain transaction volume (+247% YoY); $166M+ Arc total transactions (since Oct 28, 2025 testnet launch).

Axiom — Crypto sleuth ZachXBT exposed employee insider trading; Axiom says it’s “shocked and disappointed” to learn internal customer support tools were misused to query user wallets, revoked access to those tools, and says it will continue investigating and hold responsible parties accountable / pursue legal liability.
Conclusion
Hard numbers; fragile markets. A single narrative shock moved stocks, showing how the map is still foggy. Positioning was already tight, and conviction is missing all over the place. For most, this was a call to action to stop buying proxies, to trade smaller, and to take profits faster. Turns out AI is both bullish and bearish. AI is bullish for compute owners and consumers; bearish for intermediaries whose moat is friction. This is an environment where headlines can gap prices faster than fundamentals can correct them. Preserve ammo, avoid illiquid side-quests, and only add when price action confirms.


