The rotation across AI bottlenecks has been a relatively clear downstream move. Hyperscaler spending to chipmakers to fabs, then the market woke up to HBM shortages inside the GPUs themselves, then storage followed (SNDK). And then the consensus bottleneck rotation seemed to stop there. But going a layer deeper, as AI clusters get larger and denser, connectivity is emerging as the next key constraint, something that may only be fully realized further down the line as the buildout happens when upstream production capacity actually catches up.

Copper’s hard physical limit for scale-up interconnect is two meters at 200Gbit/s per lane, and even NVIDIA, which has been going to extreme lengths to avoid moving to optics for scale-up links, runs out of room by the Rubin generation.

-Dylan Patel’s Semianalysis

I’ve been looking into this for the last several weeks, and what I found is that every single layer of it is supply-constrained, confirmed by the companies themselves in their most recent earnings, and NVIDIA has recently already acted on it publicly.

This is a mult-year purchase commitment + future capacity access rights, partially funding a 5th InP fab. In their words, there is “a massive scarcity in indium phosphide.” They treated it identically to locking TSMC capacity at the start of the AI run.

The most informed buyer in the entire AI supply chain looked two hops upstream from GPUs, identified the material layer that breaks first when transceiver demand scales, and cornered supply publicly. Everything below explains why they are probably right, and how I am positioned.

What about the ME conflict?

I’m slapping this section here last minute because of the new developments that happened after I wrote everything. Despite the headline risk, I don’t believe it flows down to the level of cancelling photonics orders.

The orders at AAOI, LITE, and Coherent are contracted over long-term agreements. Copper physics at 800G are unchanged, and delaying cluster activation means delaying AI revenue, which is the thing driving hyperscaler stock prices and competitive positioning. The payoff is heavily weighted toward building as fast as possible, not toward saving $50M on optics while a multi-billion data center sits idle.

On the Iran risk-off day when Korean semis sold 10-12%, Seikoh Giken (Japanese fiber connector equipment) was up 13%. growth beta from the hyperscaler-contracted infrastructure. Unless hyperscalers are revising 2026 capex down, I don’t think the thesis for photonics changes here.

The Bottleneck

Model capability scales with how many GPUs can communicate simultaneously. Nvidia’s GB200 had such a huge performance leap because it now had 72 GPUs (vs 8 previously) in 1 rack. This is what every dollar of AI capex was ultimately buying.

So the question for scaling all this becomes: what limits how many GPUs you can jam into a single rack?

The answer is copper. NVLink runs on copper. Copper at 200G per lane works for 2m, then signal integrity collapses, which means every GPU has to be within 2m of every other GPU squeezed into 1 or 2 racks. NVIDIA has been making those racks absurdly dense to buy time, but power density, thermal management, and manufacturing complexity all compound simultaneously.

At some point you cannot add more GPUs by making the rack denser. The only way to grow the world size further is to extend the interconnect reach beyond 2m, and make them much faster than current gen 400G data rates.

The Rubin generation extracts one more doubling through an engineering workaround where transmit and receive share the same wire. Beyond that, the copper scaling roadmap is declining. Feynman, the generation after Rubin, is identified as the focal point for co-packaged optics (CPO) injection into the NVIDIA ecosystem, and hyperscalers are already committing to suppliers: Celestial AI is estimating a $1 billion run rate by end-2028, driven primarily by a scale-up solution shipping with Amazon’s Trainium 4.

Every CPO, whether on a scale-out switch today or a scale-up NVLink successor in 2028, requires an InP laser as its light source. Silicon cannot emit light at the required wavelengths.

scale-out = horizontal expansion, ie more racks → more interconnect demand

scale-up = vertical expansion, ie more GPUs within a rack

Indium phosphide and its related components are the bottleneck now. NVIDIA, the most informed buyer in the compute supply chain that locked TSMC wafer capacity, cornered HBM allocation, and tracked every bottleneck in the AI hardware stack for a decade, has identified InP as the layer that breaks first.

You cannot build transceivers without InP laser chips. You cannot build laser chips without epitaxial layers grown inside an MOCVD reactor where one company holds 70-90% market share with 12-18 month lead times. You cannot run those reactors without InP substrates from a duopoly. Each layer experiences supply-constrains and is still relatively cheaply priced against its historicals when really the guidance is saying something different.

Earnings transcripts from hyperscalers in Q4 2025 highlights this pretty clearly:

1. Google increasing networking capex.

Guided $175-185 billion in 2026 capex and disclosed the split: 60% servers, 40% data centers and networking. That is ~$72 billion flowing into data center infrastructure and networking from one company in a year. Sundar Pichai named the constraints during Q&A: power, LAN, supply chain. LAN called out explicitly alongside power.

2. Meta needing more optical interconnect infrastructure.

Guided $115-135 billion and is capacity-constrained through much of 2026 until new owned facilities come online. New owned facilities means new fiber, new optical modules, new switching infrastructure.

Combined capex across these 2 alone exceed $200 billion.

The Chain

Right now, pluggable transceivers handle the optical conversion externally. As data rates push through 800G and into 1.6T, the power draw and latency connecting those modules to the switch ASIC become the constraint. Co-Packaged Optics (CPOs) moves the laser directly onto the switch, making it cheaper and faster. Pluggables don’t disappear immediately, but the transition is on NVIDIA’s roadmap and suppliers are gearing towards that shift.

The transceiver and laser chip layer has already moved significantly, >800% from cycle lows, guided by upward revenue revisions in earnings calls.

AAOI: “Growth limited by production capacity, not market demand”, $1B+ guided for 2026 from $456M the previous year.

Lumentum: “We cannot make enough of these lasers”, OCS backlog tripled to $400M, sold out through 2028, NVIDIA’s $2B investment + a 7-year InP substrate deal through mid-2030s).

The remaining asymmetry sits upstream across 3 layers:

• IQE runs MOCVD reactors as an epi foundry for chip manufacturers who don't want to run their own.

• Those reactors are Aixtron's and they are a near monopoly in this sector. Revenue outlook for them was flat for 2026 because of declining Silicon Carbide business in the EV sector that is masking the photonics inflection. The optical revenue hits financials in 2027 due to 12-18 month equipment lead times where they project optoelectronics revenues to double.

• Feeding those reactors is AXTI, supplying InP substrate wafers in a duopoly with Sumitomo. Substrate demand doubled in 2025.

Every CPO module still requires an InP laser chip. The substrate and reactor layers are still needed regardless of whether the laser ends up pluggable or co-packaged, so the CPO transition extends AIXA and AXTI's demand across both generations.

SemiAnalysis

Co-Packaged Optics (CPO) Book – Scaling with Light for the Next Wave of Interconnect

Co-Packaged Optics (CPO) has long promised to transform datacenter connectivity, but it has taken a long time for the technology to come to market, with tangible deployment-ready products only arriving in 2025. In the meantime, pluggable transceivers have kept pace with networking requirements and remain the default path thanks to their relative cost-effectiveness, familiarity in deployment, and standards-based interoperability…

Read more

5 months ago · 251 likes · 4 comments · Dylan Patel, Daniel Nishball, Myron Xie, Ivan Chiam, Clara Ee, and Patrick Zhou

A part that gets missed entirely though, before any fiber connection carries signal, the connector end-face must be polished to sub-nanometer flatness or the link is unusable. And as data rates increase, fiber density per connector increases, meaning more polishing cycles are needed. Seikoh Giken makes these machines and consumable films.

Asymmetric Opportunities

Seikoh Giken (TSE:6834)

Read more

8 months ago · 12 likes · 5 comments · Asymmetric Opportunities

Japan

Lumentum locked a 7-year InP substrate supply agreement with a non-China supplier. In a market that is effectively a duopoly between Sumitomo and AXT, that agreement is likely Sumitomo. Lumentum didn't announce it that way, but the structure of the deal points directly at them:

multi-decade capacity secured, non-China sourced, locked through the mid-2030s.

Fujikura did 14x over 2 years after locking US fiber contracts worth up to $20B. Sumitomo sits at 11.8x PE against Fujikura’s 29x. Furukawa is at 20x.

Citrini flagged the upstream rotation this week for photonic components and substrates. The Sumitomo angle is different from their Furukawa valuation gap call since this to me is more of a high probability bet that they are the counterparty, hence proving the demand for upstream photonics companies.

Positioning

The transceiver layer has already repriced (LITE +800%, COHR +250%). The remaining value sits upstream where the market hasn't followed the cascade.

Primary upstream bet:

Aixtron (AIXA) Revenue down (because of EV segment), but optoelectronics segment growing rapidly, and they are essentially the monopoly in MOCVD reactors used for pluggable transceivers and subsequent CPOs. What’s interesting is Citrini flagged it out this week as well, which is confluence on the thesis.

AXT Inc (AXTI) as a supply chain bottleneck bet because of Lumentum’s 7-year deal with (most likely) Sumitomo.

Side bet:

Applied Optoelectronics (AAOI) America positioning (<10% China-sourced) with vertically integrated laser production facilities. $1B guidance, 2x from 2025

Japan layer:

Seikoh Giken has a dominant position in connector polishing, with polishing volume increasing as fibre density increases.

I believe the photonics chain is still relatively early in the physical buildout that all of that capex actually depends on to generate returns. Oracle’s Stargate delay is a prime example of optical interconnect demand from data center facilities not yet being realized.

As the one thing that converts the hundreds of billions in sunk investment into functioning compute clusters is the fiber, the lasers, and the connectors that link them together.

DISCLAIMER: The information, opinions, and projections contained in this report are provided exclusively for discussion and educational purposes. They do not constitute investment, legal, accounting, tax, or other professional advice and should not be relied upon to make, or refrain from making, any investment decision. You should perform your own independent analysis (including consultation with appropriately qualified advisers) before acting on any information presented herein.

All quantitative data were gathered from publicly available sources. Wherever possible those figures were cross-checked against offical copmany or government reports. Nonetheless, historical metrics can be restated and dashboards may contain errors. No representation or warranty, express or implied, is given as to the accuracy, completeness, or correctness of the data as of either the pull date or the date you are reading this report.

References

https://tspasemiconductor.substack.com

Then I went offline, spent time with family, with friends of family, with functioning adults who work in real industries spanning Accountants, Engineers, Managers, Consultants in their late twenties to early fifties.

Surprisingly, none of them actively use AI. They don’t use it daily or for real work regularly, a few had tried ChatGPT when it went viral, one mentioned the Ghibli image thing. That was it.

These are people with graduate degrees and real careers and zero reason to be dismissive of new tools. People who should be adopting AI but for some reason, that exposure started and stopped at viral moments of curiosity. They never used it for a single meaningful brainstorming prompt, nor redesigned a workflow around AI. They describe AI as “interesting” the same way they would describe a new restaurant that they’d try eventually.

I thought I was behind the curve. I realized I was just online too much.

Offline, the median knowledge worker is still operating in 2019.

That’s the paradox of this moment: AI is experiencing a capital-markets bubble but is behaviorally under-adopted. Both measure different layers of the same phenomenon.

The Spread

The gap between what’s been invested and what’s been adopted is the most important number in tech right now. Nobody is tracking it carefully.

Here’s the framework. Call it the Reality Gap:

Reality Gap = Upstream Enthusiasm - Downstream Penetration

Upstream refers to the capital allocation level that the markets have priced in. These include Hyperscaler capex, VC funding flows, GPU orders, data center buildouts, AI-exposed multiple expansion. This is the source of the headlines of SNDK 10x-ing, NVDA annual revenue going from $60B to over $200B in 2 years, Memory prices skyrocketing… (ie. where the bubble perception is).

Downstream refers to the the behavioral layer of what humans actually do at work. Daily AI tool usage for real deliverables, enterprise seat activation rates, workflow redesign, junior hiring trends. This is where "still early" lives.

When upstream enthusiasm dramatically outruns downstream penetration, you get both camps being right simultaneously. The bubble critics are correct about valuations but there is also a lot of headroom when it comes to adoption.

What Markets Priced Upstream

These numbers are genuinely hard to digest. Hyperscaler capex across big tech like Amazon, Microsoft, Google, Meta, and Oracle hit $256B in 2024, up 63% YoY. Projected to reach $443B in 2025 (+73% YoY) and $602B in 2026. Roughly 75% of that is going to the AI buildout: GPUs, servers, data centers built for training and inference.

To fund this, hyperscalers now spend almost 80% of their operating cash flow on capex (up from 40% in 2024), with firms like Amazon going as far as 94%. They raised over $100B in debt this year alone which is more than triple the average of the past 9 years. Meta and Oracle issued $75B in bonds and loans in just two months (Sep - Oct 2025) to fund AI buildouts. Hyperscalers plan to add $2T in AI-related assets to their balance sheets by 2030. At 20% annual depreciation on AI assets, that’s $400B in annual depreciation expense (more than their combined profits in 2025).

This essentially that captures what this actually means:

“The reason RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist.” - @jatinkrmalik

This is defensive capex where if Hyperscaler 1 is doing it, Hyperscaler 2 can’t be seen not doing it. Spend gets pulled forward in anticipation of demand that may be mis-timed by years. It’s an arms race where everyone has to play and nobody knows whether the war is coming.

VC funding is equally concentrated with AI startups capturing 52.7% of all global venture capital in 2025 which is the first time AI has taken majority share. In Q3 2025, AI absorbed 62.7% of US venture dollars. OpenAI and Anthropic alone accounted for 14% of all global venture investments, and over 2,000 AI-related deals in Q2 2025 alone.

What Humans Actually Do Downstream

Now for the other side.

In BCG’s AI at Work survey of over 10K workers across 11 countries found that only 51% of frontline employees (BCG terms them as white-collar non-managerial roles) frequently use AI in their daily work, which is a decrease from 52% in 2024.

Interestingly, higher AI adoption correlates with higher fear of job loss, which actually rhymes with the thesis whereby the people closest to the technology are most aware of what it can replace while the offline majority isn’t worried because they haven’t crossed that literacy gap yet.

The most useful data point to consider here however is what’s happening wiht Microsoft's own Copilot AI. Microsoft 365 has >450M paying subscribers, but Copilot’s monthly active paying users sit at a reported 15M (~3.3%) 2 years after launch. What I intend to point out here isn’t a Microsoft-specific criticism, but rather the description of what “adoption” actually means across enterprise AI beacuse seats sold =/= seats used. MAU numbers can easily be inflated with a low threshold for what is defined as an “active user” since i can accidentally open Gemini on my Samsung phone when I actually intended to search Google.

Why Adoption Is Slow

(And no, it’s not what you think)

The standard explanation for slow adoption is that people are lazy or resistant to change, which isn’t quite right. Structurally there seems to be 3 things that drastically reduce the level of true “people-replacing” capabilities of AI usage:

1. AI literacy isn’t binary, there is a knowledge curve to it. The gap between “I tried chatgpt bro” vs “I use AI daily to cut research time from 12 hours to 2 hours” is enormous, and frankly not visible from the outside. Crossing it requires redesigning workflows, prompt engineering and context-setting knowledge , and iterative refinement skills that take (a lot of) time to develop. I believe most people, including many young professionals who should theoretically be most motivated, haven’t crossed this chasm. The analogy I keep coming back to is:

   AI is the new Excel. Excel has been standard since the 90s, yet a meaningful chunk of the workforce still can’t build a functional pivot table. AI has a steeper learning curve than Excel, a less standardized (open-ended chat) interface, and a significantly larger variance in output quality based on how you use it.

2. Institutional friction and stigma, which is severely underappreciated. Even capable individual users often don’t use AI openly beacuse of a “cheating” stigma in many workplaces. There is an underlying fear of looking replaceable and being seen as cutting corners, all while organizations haven’t redesigned processes around AI. Then there are regulated industries (finance, law, medicine) with active restrictions so AI gets capped at the individual level but never compound across the organization.

   BCG found that only about 25% of frontline employees say they receive strong leadership support for AI adoption. When leaders demonstrate strong support, the share of employees who feel positive about AI rises from 15% to 55%. The bottleneck to stigma is actually mostly just permission and cultural cover to use it at work.

3. Lastly, the friction to change workflows. From an economic pov, change happens when the pain of staying exceeds the cost of switching. Most knowledge workers haven’t hit that threshold yet as their current tools work well enough. The forcing function hasn’t personally affected them yet.

The Mirror

What makes this more interesting than “normies don’t get it” is that the Twitter-native crowd which is supposed to be a more motivated, chrnically online, more tool-obsessed cohort that should be furthest ahead, is also failing to compound.

As Ryan Watkins put it:

“Dude I have 10 agents running while I sleep. No one is prepared for AGI in 2 years man.”

“So what are you building?”

“Bro all my smartest friends are vibe coding until 3am every night. It’s all about agency. Intelligence is a commodity man.”

“So what are you building?”

“Do you even study exponentials? Have you seen the latest METR chart? You’re going to be stuck in the permanent underclass bro.”

“So what are you building?”

“I promise you I’m 10x more productive bro! You just don’t understand! Please bro just…. I know you use this stuff everyday too, but you must not be prompting it right! Please broo…”

The “10x more productive” claim without any shipped work. In relation to my New Year observation, its a complete mirror image where one end of the distribution under-adopts entirely, the other over-signals without building anything. Both failure modes confirm the same gap that actual, compounding, output-generating AI use is rare across the whole distribution, not just the offline majority.

If even the most motivated, risk-native, tool-obsessed cohort hasn’t figured out how to reliably compound with AI, what does that tell us about the timeline for everyone else? To me it suggests the adoption gap is going to be wider than consensus assumes.

The Gap Doesn't Close Smoothly

Tne adoption gap among individuals exist but is largely irrelevant to the overall outcome because the displacement mechanism bypasses the individual level entirely. Perhaps this is portrayed more clearly in Citrini’s piece which I’m sure many have read.

The core question it asks, “what if AI bullishness keeps being right... and what if that’s actually bearish?” This mechanism is worth considering.

In this scenario, the adoption gap doesn’t close because individuals finally get it. It is forcefully closed because CFOs start cutting headcount to fund AI tooling with institutional adoption growing at a pace where workers (especially juniors) can’t match.

The Flashing Signal

The leading indicator for this dynamic is entry-level hiring.

What I intiially expected to see (and what I’m watching now) isn’t massive unemployment announcements. Its likely something quieter with fewer junior, entry-level roles hired, slower replacements when someone leaves, flatter/leaner teams shipping the same output, budgets shifting from headcount to AI tools.

A Stanford working paper from 2025 documented a 13% relative employment drop for 22-25 year-olds in AI-exposed occupations (customer service, accounting, software development) since 2022. This isn’t as severe among more experienced workers in the same fields.

The one category worth stress-testing this against is software. The BLS projects +15% employment growth for developers through 2033. There’s a Jevons paradox argument here, and loosely quoting CampbellRamble:

If AI makes code 10x cheaper to produce, does the world want 10x less code or 100x more?

Every company that couldn’t afford a dev team now can. Demand for what software does is arguably infinite. So software might restructure rather than shrink where senior engineers are fine, but junior CRUD grinders not.

But Jevons doesn’t save the other categories in that Stanford paper as there’s no latent infinite demand for more customer service tickets or accounting reconciliations. These are the jobs that don’t expand when they get cheaper. They just disappear.

Where This Lands

The whole point of writing this is to address the bubble vs early debate by pointing out that both camps are measuring different things. The bubble is evident in valuations, capex intensity, and VC concentration. But we’re also early in low daily usage rates and in the sheer number of knowledge workers who have never crossed the AI literacy chasm.

The AI buildout likely continues regardless of job displacement because capex is defensive and the Hyperscalers pouring billions into this cannot afford to be wrong. The first leg ran through chip designers and GPU manufacturers, the second through fabs, HBM, and memory. The next logical stops are energy infrastructure powering the clusters and the interconnects pulling GPU servers together at scale.

The more interesting short leg pointed out in Citrini’s piece is friction-based consumer businesses, but that trade is obvious enough to be crowded and wrong in its timing. The cleaner expression of this thesis is replicable mid-market SaaS where competent internal teams can replicate million-dollar annual contracts in weeks. Others like Asana (-61% past year), Monday.coms (-73% past year), ie workflow tools whose moats were mostly switching costs gets hit from the labor displacement thesis (fewer knowledge workers, fewer seats).

The people who compound AI literacy now will have a durable edge when the downstream catches up to the upstream. If you're entering the workforce in the next few years (which is uncomfortably close to Citrini's scenario window), I believe the best path to take isn’t necessarily to “find a job AI can’t do”, but rahter to become the person who makes a small team output what a large team used to.

The setup isn’t particularly optimistic, but what needs to be done on the individual-level is clear, and so is the trade: long the infrastructure that has to get built regardless, short the software that gets commoditized by the same buildout.

DISCLAIMER: The information, opinions, and projections contained in this report are provided exclusively for discussion and educational purposes. They do not constitute investment, legal, accounting, tax, or other professional advice and should not be relied upon to make, or refrain from making, any investment decision. You should perform your own independent analysis (including consultation with appropriately qualified advisers) before acting on any information presented herein.

All quantitative data were gathered from publicly available sources. Wherever possible those figures were cross-checked against offical copmany or government reports. Nonetheless, historical metrics can be restated and dashboards may contain errors. No representation or warranty, express or implied, is given as to the accuracy, completeness, or correctness of the data as of either the pull date or the date you are reading this report.

References

https://www.microsoft.com/en-us/investor/events/fy-2025/earnings-fy-2025-q4

https://www.statista.com/chart/35046/capital-expenditure-of-meta-alphabet-amazon-and-microsoft

https://news.crunchbase.com/ai/big-funding-trends-charts-eoy-2025/

https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain

https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

https://web-assets.bcg.com/fd/0d/bcc5dfae4cbaa08c718b95b16cf5/ai-at-work-2025-slideshow-june-2025-edit-02.pdf

https://plusai.com/blog/how-consulting-firms-use-ai

https://aifundingtracker.com/q2-2025-ai-funding-report/

Preface

DeFi has undergone a pivotal shift in how users interact with DEXes and token launch platform. In the early days, each protocol (exchange, lender, etc.) had its own website to onboard users. Today, many users bypass those official UIs entirely. Instead, they flock to third-party aggregator platforms that combine multiple services into one. These serve as one-stop terminals for trading, lending, yield farming and more, abstracting away the need to visit individual protocol websites.

The result is that protocol-owned frontends are losing ground. The bridge between a DeFi protocol and its user is weakening as aggregators become the preferred gateway.

This trend raises a pivotal question: who really “owns” the DeFi user… the protocol or the interface? As independent frontends capture the user’s attention, they effectively control which underlying protocols get used and how. This power shift has serious implications. Protocols risk being commoditized, reduced to back-end utilities chosen only for better exchange rates or liquidity, while frontends capture the user relationship (and even fees) at the front.

This report examines the shift from protocol-specific frontends to independent platforms that have become the user-facing product layer. Are protocols losing control of the end user?

Key Takeaways

• Users prefer aggregation over isolation: Instead of visiting separate sites, users increasingly opt for aggregator interfaces (DEX aggregators, trading terminals/bots) that provide a single point of access to many protocols​. Convenience and breadth of features are winning out over single-protocol experiences.

• Aggregators “own” the user relationship: The platform where users actually interact is the primary touchpoint, effectively owning the customer. This leaves individual protocols with less direct influence over user engagement​ and puts third-party frontends in a position to guide user flow across the ecosystem.

• Value capture is shifting to front-ends: Control of the interface also means control of monetization. Aggregators often take a cut or fee for facilitating user transactions (ie. value that bypasses the protocol). This shift is more pronounced with Solana trading bots and terminals like Photon, BullX, Axiom, etc.

• Protocols are becoming back-end utilities: Protocols can no longer count on users coming to them by default. Every protocol is just one of many options under the hood of aggregators and they must compete on yield, price, and liquidity, since users will gravitate toward whoever offers the best execution​.

Background

This shift in frontend dynamics is drastically more pronounced on Solana. Historically, protocols like Raydium, Orca, and Serum operated with their own frontends. Users went to the protocol’s official site to swap, trade, or provide liquidity. This worked fine for years when the ecosystem was simple.

But once Solana’s memecoin season hit in 2023–2024, user behavior flipped. Pumpfun also had their own frontends, but the user-facing interface was no where near as good as the bots and terminals that came after. New players like Photon, BullX, Trojan, and later Axiom built faster, cleaner and more intuitive CEX-like frontends. Some are bots while some are full terminals that let users snipe launches, copy whales, and flip tokens at speed. Users didn’t just prefer these tools, they abandoned official frontends almost entirely when trading memecoins.

I’ll use Solana as the main case study here, but the dynamic is generalizable. DeFi frontends are fragmenting, and whoever controls the screen the user taps or clicks on will control the flow.

Market Landscape Overview

Memecoins now drive the bulk of DeFi activity on Solana, eclipsing more traditional DeFi trading. During the peak of the craze, Raydium derived about 80% of its volumes from memecoin trades according to Blockworks (with 40% from pumpfun alone). On top of that, 90% of Raydium’s swap fee revenue came from memecoin liquidity.

As the previously go-to AMM for Solana tokens, Raydium has seen a significant drop in volume following the launch of Pumpswap which new memecoins will now migrate to instead of Raydium. The key difference is what the volume comprises: today Raydium’s volume skews more toward non-Pumpfun assets (e.g. “older” tokens like BONK and legit SPL assets and core pairs like SOL/USDC), while nearly all new memecoin trading occurs on PumpSwap.

Third-party Frontends

Solana’s memecoin season changed user behavior entirely. Traders didn’t want to manually copy contract addresses, open five tabs, and pray for a good swap on Raydium. They wanted sniper bots, fast UI, live new token feeds, copy-trading whales, one-click selling when rugs occurred. Protocol-native frontends couldn’t keep up, and so terminals and bots were born.

It’s worth noting that a large percentage of PumpSwap’s trading volume is driven by these third-party terminals, rather than purely via Pump.fun’s own front-end. Examples include:

Photon - Launched in late 2023 by Tiny Astro Inc., Photon positioned itself as a high-speed DEX aggregator tailored for memecoin “snipers.” It quickly expanded beyond Solana to support other chains (Ethereum, Base, Tron, etc.), but Solana remained its core.

GMGN – Founded in 2024, GMGN focused on being a data-rich trading and analytics tool for meme coin traders​. It aggregates token launch info, liquidity changes, and social trends across multiple launchpads.

Trojan - Initially launched as a pure Telegram bot interface, Trojan targeted speed and simplicity for Solana token trading. Users interact by messaging the bot to snipe newly launched tokens or set quick orders.

BullX - Launched around early 2024, they branded themselves as an “advanced DEX trading system” spanning multiple chains. It offered trading features like live charts, cross-chain limit orders, and a Telegram-integrated interface​ for users to login.

Axiom - A later entrant that has recently become a popular choice for onchain Solana traders. Axiom differentiated itself by building a feature-rich terminal for serious traders offering tools like Wallet Tracer and Twitter Tracker to follow big wallet movements and social sentiment​.

Volume & Market Share Analysis

The top five platforms (Axiom, BullX, Photon, GMGN, and Trojan) collectively attract the majority of user activity in this sector. Each has tens of thousands of daily active users (DAUs), daily volumes in the tens of millions of dollars, and hundreds of thousands of transactions per day. Table 1 summarizes recent daily usage metrics for these leading terminals, and Table 2 provides their lifetime fee revenue and AUM:

Table 1 – Daily Usage of Top Solana Trading Terminals​

Venue	      DAU      	        Daily Volume	  Transactions/Day
Axiom	      ~17k – 30k	$100M – $150M	  700k – 1.5M
BullX	      ~20k – 30k	$20M – $50M	  200k – 400k
Photon	      ~18k – 27k	$30M – $50M	  250k – 350k
GMGN	      ~10k – 18k	$8M – $20M	  180k – 290k
Trojan	      ~14k – 25k	$7M – $30M	  ~200k

Sources: Dune Analytics (via Adam_Tehc) as compiled by Nico​. Figures represent typical recent daily ranges.

Table 2 – Lifetime Fees and AUM

Venue	Total Fees Collected	AUM (SOL)	AUM (USD)
Photon	~$384 million	       ~539,000 SOL	~$82.6 million USD
BullX	~$187 million	       ~215,000 SOL	~$30.0 million USD
GMGN	~$67 million	       N/A	        N/A
Axiom	~$44 million	       N/A       	N/A
Trojan	~$177 million	       N/A	        N/A

Sources: Dune Analytics (via Adam_Tehc). Figures not disclosed for GMGN, Axiom and Trojan.

Sources: Dune Analytics (via Adam_Tehc)

The five platforms collectively have over 100,000+ daily active traders and over $200M in daily volume​. This constitutes almost 50% of pumpfun daily trading volumes.

Native Protocol Front-Ends Are Increasingly Marginalized

One clear outcome of these trends is that the native web front-ends of liquidity protocols (like Raydium’s own site) have been marginalized.

Prior to PumpSwap, nearly half of Raydium’s volume and fees came from Pump.fun’s funnel​ rather than organic traders on the Raydium app. Even beyond Pump.fun, a large portion of Raydium’s remaining volume is driven by aggregator flows or arbitrage bots, not retail users manually swapping on the Raydium frontend.

Users rarely go directly to a specific AMM’s interface to trade anymore.

For memecoins, the Pump.fun interface initially brought users in, but now those users stick to trading terminals and bots. The result is that these DEXes are now heavily reliant on external order flow sources.

This indicates that the “front-end” of Solana DeFi has largely moved to third parties. Though the liquidity still resides in protocols like Raydium or PumpSwap, the traffic comes from elsewhere. As a result, native DEX tokens and UIs capture significantly less mindshare and product/UX-layer fee revenue from end-users.

Legacy DEX Value Chain (Pre-2023)

DEXes used to control the entire user experience and user traffic. Users went straight to the protocol website frontend to swap tokens, and those trades were executed directly against that protocol’s liquidity pools. A routing/aggregation layer was minimal initially as users usually traded on one DEX at a time. But as the ecosystem matured, aggregators like Jupiter emerged to source best prices across multiple DEXs, but early on many users still went directly to each DEX’s UI.

Source: Aw Kai Shin

In this legacy model, swap fees were largely captured by LPs and the DEX protocol’s treasury. For example, Raydium charged a 0.25% fee on swaps, with 0.22% of the trade value paid out to liquidity providers and 0.03% going to Raydium’s own treasury.

In practice this meant virtually all fee revenue from trades stayed within the protocol (LPs earned the bulk, and the DEX kept a small cut). The frontend itself did not add any extra fees beyond the pool’s swap fee since it was simply the interface to the underlying protocol. Even when aggregators like Jupiter were introduced as a routing layer, they initially charged little or no additional fee. Jupiter’s role was mainly to optimize price and reduce slippage by splitting routes across DEX pools, while the standard pool fees still went to the utilized protocols.

In summary, the value chain was vertically integrated: the protocol owned the UI, supplied the liquidity, and captured the fee value (with LPs and the protocol sharing the swap fees).

Trade Routing Infrastructure in the New Model

Starting around 2023, a new breed of third-party trading interfaces created a layer above the DEXs and changed the value chain. These were terminals and bot frontends that improved trading experiences and provided users with a much better UI to trade on. Now, instead of going to a Raydium or Orca site, users began routing swaps through these independent frontends that aggregate liquidity in one place.

These third-party terminals effectively hijacked the user flow. A trader could use Axiom’s UI or a simple Telegram command via Trojan to execute a swap. Under the hood, those services route the order to the best Solana liquidity source, but the user interacts solely with the terminal’s interface.

This practically works the same as the legacy model, just with an added frontend that abstracts away UX complexities compared to being directly on Jupiter. Terminals may even integrate aggregators or various direct liquidity sources and query between both to find which route is best. The execution path would then look like:

User → Terminal UI → Router (some)→ DEX Pools (liquidity) → Settlement on Solana.

One side effect of adding an extra layer is a potential increase in total fees (as the terminal’s fee is added) and slightly more complex transactions (the aggregator might do multi-hop swaps), but Solana’s efficiency has kept this overhead small and even highlighted a demand in terms of DeFi UI/UX that users would pay a small premium for.

Fee Extraction: How Frontends Capture the Revenue Stream

As mentioned previously, third-party frontends have restructured fee capture in DEXes. Now, liquidity has become a commodity that is easily tapped by external aggregators, and now the user-facing layer has become the prime source of revenue.

Historically the protocol UI guaranteed the protocol a share of user activity, now the frontends capture significant value despite being “just UIs”. In fact, many of these platforms now rival the base protocols in fees generated.

For example, over the course of 7 days, Pumpfun (launchpad + Pumpswap) pulled in $18.27M in fees.

Frontend trading terminals on the other hand, pulled in over $24M in fees combined. To put it in context, Axiom’s frontend alone attracted more trading fees than Raydium’s own liquidity pools (~$9–10M weekly) over that period.

All this trading volume and demand is now outside the control of the protocol. Terminals can monetize order flow and take a cut on every swap they broker even though the liquidity and market making was done by another protocol. For protocols, they still earn their small share of pool fees, but they no longer control the trader’s experience or have a direct line to monetize those users beyond the raw liquidity fee.

DEX treasury income is now largely limited to their built-in fee cut, whereas the new frontends enjoy a new income stream that many users are surprisingly willing to pay more for. Raydium’s AMM fees are 0.25% while Axiom charges up to 1% for the lowest tier. All this just goes to show how important it is to own the end user and why the “UI layer” that directly connects protocols to users is so important. You can visualize this by going 1 layer higher to Phantom wallet itself, their built-in swap generated >$40M a week during peak retail meme season, more than many standalone DEXes.

Liquidity is still king on the backend, but order flow is the new gold on the frontend. Protocols that once prided themselves on their shiny apps now see guerrilla UIs siphon volume away.

Fee Destinations and Revenue Distribution

In nearly all cases, the terminal’s team accrues the platform fees as revenue. Some projects have or plan community distribution via referrals or tokens. Referral programs are common: Trojan redistributes 25–35% of its 0.9% fee to the referring user as an incentive, and BullX offers an even larger rebate with 35% of fee revenue from direct referrals, plus smaller percentages from sub-referrals. This is effectively a marketing spend.

At present, none of these terminals have an active profit-sharing token, so “fee to token holders” is not yet a channel (though some are planning tokens/airdrops to reward users in the future via points)

Terminal Monetization Models: Aligned vs. Extractive

Most terminals can be argued as extractive since they introduce an added fee on top of the underlying infrastructure that is not shared within the ecosystem or improves on-chain liquidity for the broader good.

Terminals extract value via high fees without directly reinvesting in the backend liquidity layers. None of the major terminals share their fee revenue with LPs or DEXes, nor do they reduce the underlying fees, meaning users pay full DEX fees plus the platform’s cut.

There are a few nuances. By not artificially routing trades to a favored platform, they at least ensure liquidity is used efficiently across Solana so they are not worsening trade execution.

Source: Axiom

Additionally, a few platforms offer tooling that indirectly aids the ecosystem such as Axiom’s integration of Solana’s HyperLiquid perps and MarginFi lending is an example of collaborating with other protocols to expand functionality. While Axiom still takes fees on trades, the integration of external protocols (perpetual swaps, margin trading) can increase overall on-chain activity and diversify liquidity usage, rather than just siphoning swap fees.

(Shameless Axiom referral shill here.)

Although frontends are purely an added cost layer, the value proposition of these terminals lies in their superior user experience rather than improving underlying liquidity. Though often criticized as high-fee, extractive frontends, I would argue that they contribute to volume and price discovery indirectly due to lower friction to purchasing assets, and sometimes smoother and better execution. In practice, the explosion of on-chain trading/memecoining via bots/frontends did benefit LPs on various DEXs because of the sheer volume, so they indirectly “give back” by generating fees for LPs. But this is a side effect of user activity.

Also, users have the freedom to not use terminals and simply use the underlying protocol to swap if they wanted to. For example, Pump.fun’s own UI makes a point to note that its web and mobile apps charge no additional frontend fee beyond Pump.fun’s built-in fees.

Protocol Strategic Responses

The implication is clear. To stay relevant, DEX protocols must either integrate into these aggregators, offer competitive frontend experiences, or find ways to align with the terminals. Jupiter is one of them.

Jupiter launched Ape Pro, closely rivalling terminals like Photon and BullX in offering real-time charts, one-click swaps, social logins, and MEV-protected trading via Jito validators. Critically, Jupiter undercut competitors on cost by charging a 0.5% swap fee on the frontend (half the market rate). In effect, Jupiter built its own “extractive terminal” to compete head-on, but with more favorable fees and the Jupiter brand. However, overtime, it still fell off to newcomers offering better UIs and fresher unfarmed points.

In January 2025, Jupiter acquired Moonshot, a popular Solana memecoin app whose primary audience is retail. The $TRUMP frenzy garnered over $400M in volume for Moonshot and the app hit over 1M DAUs. By taking a majority stake in Moonshot, Jupiter instantly gained a huge userbase and integrated features like fiat on-ramps as Moonshot allowed buying memecoins with retail-friendly payment rails like Apple Pay.

Risks, Dependencies & Long-Term Implications

Fragmentation of User Interfaces

The launch of various trading terminals has fragmented the user experience with each terminal offering its own unique UX with different dashboards, layouts, execution engines etc. This fragmentation also leads to inconsistent information where one frontend might not display a token’s warnings or pool stats that the protocol’s own UI would show, which is a security concern. For example, you could still end up buying rugs on photon especially newer degens without much experience or knowledge.

Impacts on Protocol Survival, Partnerships, and Token Utility

In the long term, the dominance of third-party terminals forces protocols to adapt. Those that failed to respond are already seeing declines. For example, when Pumpfun introduced PumpSwap, Raydium’s volumes from memecoins nuked, endangering a major source of its revenue. Protocol survival seems to hinge more on how vertically integrated it is and how much of the user journey it actually owns, rather than just how deep liquidity is.

This links to the next point on token utility. Most protocol tokens derive value from platform usage (trading fee revenue, staking revenue etc.). With usage heavily shifting to third-party UIs, the connection between the token and the end-user can weaken. If Photon launches a token that gives holders fee discounts or a share in its revenue, traders might favor that over holding the AMM tokens that based on the data, don’t generate nearly as much fees as frontends would.

Conclusion

Highlighted by the success of frontend terminals, it has become more important than ever for protocols to not just focus on the tech, but also on UI/UX. Vertically integrating the token lifecycle from launch to LP (like Raydium’s launch of Launchlab), matters far less than owning the end user.

Liquidity protocols are becoming increasingly commoditized as frontends take control of the end user, effectively deciding where trade volume goes, and which pool gets used, all while taking a fat cut without offering a single $ of liquidity. It may be value extractive but my argument that it highlights a bigger pain-point to demand that level of value extraction is why I remain bullish on terminals.

For now, most Solana terminals remain tokenless after months of points farming and surging adoption, barring $UNIBOT as many recent terminals far eclipse the offerings and usage that Unibot had 2+ years ago. The first to launch a well-structured token could seize first-mover premiums and position itself as a dominant true utility project.

DISCLAIMER: The content in this article is provided solely for general informational purposes based on current facts and sources and should not be considered professional advice. Please perform your own research and consult a qualified advisor before making any decisions. The author is not liable for any outcomes based on the information presented.

That being said, none of this should be construed as financial advice. Always do your own due diligence and research before investing in this casino clownworld we call crypto.

Preface

As we navigate this cycle, one thing is clear: this isn’t 2021. Back then, reflexivity was driven by a few dominant narratives (DeFi, NFTs) and a flood of liquidity. Fast forward to today, the market is fragmented. Every day, a new “next big coin” drops, a new narrative emerges, and liquidity gets stretched thinner. Reflexivity is still alive, but it’s spread thin across dozens of tokens and narratives at once. The result? A market that’s a mile wide but an inch deep where lots of things going up a little, but few staying up.

In this piece, we’ll break down how reflexivity is playing out in this new environment, why liquidity is the silent killer of this cycle, and how I’m positioning myself in this part of the cycle.

What Stage Are We In This Cycle?

I’d like to believe that we’re on the cusp of a bottom/already bottomed (obviously, for the sake of muh bags). Almost every sector is downbad YTD, with AI and Memes taking the biggest hit (-80-90%)

I think by now, you probably would’ve noticed the dispersion and thin liquidity in your daily narrative-chasing and “finding the next big thing” arc. Since the start of this bull run (which I know most people like to classify it as Nov 22 or Jan 23 post-FTX days, I would put Jan 24 as the start of this new paradigm), apart from BTC, ETH, and DeFi, we’ve seen a tremendous uptick in narratives.

• Animal - The OG meta, even though dog/cat deserve their own categories because millions of sub-, sub-sub-, sub-sub-sub- categories were spun up.

• Real-World Assets (RWA) - a TradFi favorite, can be convincingly pulled off as a “fundamentals” trade rather than pure mimetics. [ONDO, PRCL, CPOOL…]

• AI (agents) - H1 of 2024 you had the original AI thesis revolving RNDR, NEAR, FET, AGIX, etc. Then came truth terminal and now the “AI narrative” has revolved almost entirely around agents and agentic frameworks. [$VIRTUAL, $ARC, $AIXBT, $AI16Z, $pippin, $AVA…]

• DeFAI - Right… forgot to mention this small subset within AI that has become pretty big on its own. Agents now being able to perform DeFi tasks, with their own sub-classifications [$GRIFFAIN, $ANON, $GRIFT, $BUZZ…]

• President - Pretty self-explanatory. [$TRUMP, $MELANIA, $BARRON, $KAI…]

• Web2 Founders - If you’re chronically on CT, you would’ve seen this. Web2 founders on their redemption arc coming into crypto. [$VINE, $JELLY]

These are purely the ones that are still I think in the spotlight now. But you probably didn’t remember that just a few months ago we had things like the wifhats, celebrity coins, zoo animals, cute animals, euthanised animals, quants, baby, boomer, zoomer, tiktok… the list goes on.

On the larger picture, let’s look at TOTAL3, BTC.D and stablecoin supply.

Total 3 tracks the total market cap of crypto excluding BTC and ETH, essentially serving as the aggregate value of altcoins, stablecoins, and meme coins. This is currently at near November 2021 highs.

BTC.D is chilling at 58%, down from 61% in November 2024.

It is pretty clear though that November 2024 till January 2025, there had already been an alt season in the form of onchain, specifically with AI and memecoins. BTC.D fell, TOTAL3 surged, and stablecoin supply climbed alongside it (now sitting at just under $215B)

Reflexivity in Past Cycles

Soros defines reflexivity as a theory that positive feedback loops between expectations and economic fundamentals can cause price trends that substantially and persistently deviate from equilibrium prices. This is what people call “price driving narratives rather than narratives driving the price”.

Crypto creates the perfect environment for reflexivity: 1) There’s no clear valuation framework, ie. pure speculation; 2) Low liquidity; 3) Attentionomics. Your favorite KOLs on CT, Tiktok and Telegram group chats collectively shilling.

In 2017 it was ICOs; in 2020 it was DeFi yield farming; in 2021 it was meme coins and NFTs. From Jan to May 2021, $DOGE pulled off almost a 200x.

Dogecoin is the perfect asset to highlight this reflexivity and the change in the markets as a whole back then vs now. It had no fundamental valuation framework, and was the first mover for the term we now call “memecoins”.

High-profile endorsements, particularly from figures like Elon Musk, ignited a self-reinforcing feedback loop.

Stablecoin liquidity, while comparable to today’s levels, was funneled into fewer outlets, creating a "crowded theater" effect where capital and speculation concentrated heavily on Dogecoin. Additionally, the market’s novelty and retail-driven mania, fueled by pandemic-era stimulus checks and lockdown boredom, lowered skepticism, enabling memes to dominate.

The best part of all this was it was driven almost entirely by retail spot demand rather than leveraged derivatives. At Dogecoin’s peak price, there was only ~$60M in OI, compared to now at half the ATH price, yet OI is >$1.5B.

Reflexivity Now

2024 has bucked the trend, with Bitcoin holding strong while most altcoins struggle to steal the spotlight​.

It's like the market developed ADHD where attention jumps from one shiny object to the next before any single trend can gain serious traction.

While stablecoin liquidity today is comparable to 2021, reflexivity is diluted, and harder to sustain across dozens of narratives (AI, DePIN, RWAs, 100+ memecoins). The key differences are:

• Fragmentation: Capital is spread thin across hundreds of low-cap tokens. This dilutes the reflexive feedback loops.

• Leverage saturation: Traders now use perps more heavily, making OI a critical metric.

• Risk awareness: Post-2022 trauma (LUNA, FTX) has made investors wary of “dumb money” pumps.

The result is that most new coins/narratives end up like Bitconnect charts.

Altcoin season in the traditional sense has been elusive.

The usual rotation out of BTC into alts hasn’t materialized in force.

Intuitio noted that unlike past cycles, this time around Ethereum and other alts have lagged significantly… (and no, eth never broke aths)

What we have now is a market structure defined by breadth: lots of tokens seeing mini-pumps, but shallow depth of conviction behind any single one.

To illustrate how fractured things are, consider that by late 2024, Bitcoin dominance had risen to levels not seen since early 2021​. In Jan 2025, BTC dominance hit 65%. All this happened while total crypto market cap was growing, which meant that other coins were all lagging.

Plenty of tokens existed and had some activity, but few had sustained long enough to outpace corn. The Altcoin Season Index stayed in “Bitcoin Season” territory for most of 2024 too.

Attentionomics

In the current crypto cycle, attention has become the most coveted asset.

Fundamentals and traditional tokenomics have taken a back seat to memes, viral moments, and reflexive hype. This phenomenon is loosely termed “Attentionomics” where the value of many tokens now hinges more on capturing eyeballs than on fundamentals.

In a market fragmented into thousands of coins, human attention is the only truly scarce resource. Projects that command it, will see it in their price action.

As Redphone puts it:

“Attentionomics is a coin’s capacity to capture attention is more important than other metric. The greater its capacity to capture attention, the greater its potential upside. And there are very real (& identifiable) factors that determine just how large a coin can grow.”

The Attention Flywheel

Attentionomics comes down to a self-reinforcing attention flywheel in today’s social-media-driven crypto markets. The cycle typically follows a similar flow:

1. Viral Catalyst: A meme or event that creates a new narrative and curiosity that inspires someone to mint a token. Take Ghiblification for example.

   

2. Speculative Surge: Early speculators pile into the token, sending the price vertical. In crypto, price itself is content. Charts showing 10× gains in hours start making the rounds on socials.

   

3. Reflexive Hype: The price pump becomes proof of the meme’s “strength,” attracting even more attention. Viral posts bring in a larger second wave of buyers who don’t want to miss the next moonshot. Liquidity floods in, further lifting the price. Betas also start to appear.

   This feedback loop: Attention → Price → More attention, can unfold rapidly, sometimes within a single day of a meme’s inception.

4. Mainstream Amplification: If the frenzy grows large enough, it spreads beyond niche crypto circles. Media coverage, exchange listings, or endorsements amplify this reach, effectively creating value from virality.

   

This reflexive loop means that attention itself becomes a form of momentum. As the great Cobie put it:

People talk a lot about scarcity in crypto. Whether it’s enforcing digital scarcity through NFTs or the idea that “there’s 55m millionaires but only 21m bitcoins”. In reality, the only scarce resource in crypto is attention. Risk-seeking capital is certainly not scarce.

Projects/Coins that win the attention lottery can easily see their market caps go parabolic, a phenomenon less seen in TardFi.

Shitposts With a Price Feed

Consider that many of the hottest tokens in 2024–2025 were essentially launched as shitposts with a price feed.

$ROUTINE was spawned purely for the lulz (and profits) around a trending topic. Such honesty doesn’t deter investors. On the contrary, it’s part of the ironic humor.

Recognizing the short-termism of attention-fueled projects, some of the most successful meme projects are now experimenting with adding real utility or infrastructure.

Does this work tho?

Pepe has floated the idea of a dedicated Pepe Chain and related products to capitalize on its massive community. The notion is that a Pepe-themed L2 or DEX could give $PEPE holders something to actually do with their tokens beyond swapping in and out. It’s a strategy of leveraging “brand” recognition to bootstrap a user base for real platforms.

Many so-called utilities from meme projects feel duct taped as a post-pump justification. A meme-branded DEX or merch store might exist, but more often than not doesn’t meaningfully drive value back to the token.

Often these “utilities” are thin wrappers around the same community speculative impulse.

Attention is still the engine. The product is an accessory.

Musical Chairs of Capital

What happens when attention doesn’t stay for long? Traders enter a constant game of rotation.

Moving capital from one sector to another (or going down the risk curve to buy betas) has become the meta strategy.

Since no single narrative delivers 10x returns (or for most who miss the main wave), the next best thing is to catch a series of smaller waves.

This was exactly how the Euthanisia Coaster meme came about.

We saw this in action after people made bank on $ROUTINE, and profits soon rotated to related tokens (like $SARATOGA, another meme from the same viral video)​.

This kind of hot money rotation is why you’ll get weird weeks where, say, dog-themed memecoins all pump together, then the next week it’s AI tokens, then perhaps old DeFi coins get a random bid (because someone says “hey Yearn hasn’t pumped yet, maybe that’s next”).

It’s a speedran reflexive game:

See green,

buy,

make it greener,

then dump before it turns red.

Rinse and Repeat.

Leverage - Reflexivity on Steroids

Another huge shift from 2021 to now is the role of leverage.

The 2021 Dogecoin mania was largely driven by spot buying. Millions of retail buyers on Robinhood and Coinbase simply buying DOGE with cash from their covid stimmies.

Today, a significant chunk of market action comes from derivatives in the form of perps and options.

A lot of this is just crypto traders apeing with margin on Binance, Bybit, etc. When you have that much OI, price moves can get violent.

We saw Bitcoin rip from $75k to $90k in just 2 days in November 2024, with plenty of short squeeze scenarios along the way. It is reflexivity driven by leverage (shorts getting liquidated = forced buying = price goes up = more shorts liquidated, etc.). But this cuts both ways.

High leverage = high reflexivity, but often not in a healthy, sustainable way.

We start seeing situations where price swings way out of proportion faster and more frequently, driven by leverage and then reverting back to the mean since the pamp was not built on a steady inflow of new money. An insight here is that open interest can pump price, but it doesn’t equal new capital. It’s a PVP game if anything.

To put this into context, from November to December 2024, total OI increased by about $70B. Yet stablecoin supply only grew by $30B.

The fact that 2024’s OI eclipsed 2021’s by such a huge margin tells me that this cycle’s reflexivity was more mechanical and less organic. When coins pumped in 2021, people bought and held with undying conviction. This time, when things pump, a lot of it is traders yelling, “I’m long, don’t rug me bro!” with their cursor hovering over the sell button.

Conclusion

We are seeing a breadth-focused cycle with many narratives and many tokens popping off in its own isolated mini-cycle.

It’s possible we’re in an interlude and that a more profound, unified cycle is yet to come. Perhaps the groundwork laid by institutions (ETF approvals, RWA integration, etc.) will eventually ignite a broader bull run that lifts many boats and stablecoin dry powder finally floods into alts and BTC.D drops giving us an old-school altseason.

On the other hand, maybe fragmentation is the new normal and are signs of a maturing market. Crypto has grown so large and diverse that expecting everyone to FOMO into the same trade might be unrealistic. No longer does the tide lift all boats like a 2017-style “all coins go up together” moment. Being selective, nimble and skeptical is now more important than ever to thrive in this market.

One thing that is for sure for both instances is that Reflexivity will remain, just in differing forms and magnitudes. The challenge (and opportunity) is in identifying which feedback loops are merely transient hype and which could spiral into something bigger.

Just when you think a narrative is dead, it resurfaces.

Who had “Trump meme coins” on their wishlist? Yet there we were​.

Just when you think an asset is too big to fail, it falls more (ETH meows from $1800s).

As we move forward, I’m keeping the lessons of this cycle in mind. Stay agile, but also know when to sit tight and be skeptical of every narrative.

I admit that “All breadth, no depth” sounds like a complaint, but it’s also a description of a market that’s maturing in unpredictable ways. Perhaps depth will return in the next phase, or perhaps we’ll fragment further into a hundred more echo chambers. There will always be opportunities for the prepared (and pitfalls for the careless).

Reflexivity isn’t gone. It’s just more complex.

Stay safe, stay sharp, and don’t forget to secure your freedom when your memes turn to condos. I’ll leave you with a legendary quote by mgnr:

In discretionary the right position is almost always flat

do nothing. 5x / yr there's free money on the floor

pick it up. then do nothing

outperformance

Cheers!

This is simply my 2cents on current cycle dynamics, I’m sure there are 101 different takes out there, which are all valid btw!

Thanks for reading!

If you haven’t, please subscribe - it’s completely free :)

Subscribe now

DISCLAIMER: The content in this article is provided solely for general informational purposes based on current facts and sources and should not be considered professional advice. Please perform your own research and consult a qualified advisor before making any decisions. The author is not liable for any outcomes based on the information presented.