CryptoFund Radar

The national data center backlash is quietly building momentum for home inference boxes

People are waking up to the fact that centralized AI infrastructure = centralized control

Running your own inference at home isn't just about privacy anymore - it's becoming a sovereignty play

The shift from cloud to edge is happening faster than most realize. Hardware is getting cheaper, models are getting smaller, and the "AI in a box" narrative is gaining traction

This isn't tinfoil hat stuff. It's the same decentralization thesis we've seen play out in crypto, just applied to AI compute

AI labs burn billions on data + compute

Models inch forward

"We beat the benchmark" 🎉

Rinse. Repeat.

But nobody cares about benchmarks anymore.

Nobody's hyped for LLM v10.

People want AI that works a full shift, cracks scientific problems, solves what's never been solved.

Then a new algorithm drops—one that actually learns like humans, not just memorizes terabytes of slop.

Everyone pivots their agents to it.

Game over.

"No one saw this coming" they'll say.

Except the ones who did.

AI labs burning billions on data + compute just to inch up benchmarks nobody cares about.

Meanwhile, people want AI that actually works a full shift, solves real problems, and makes breakthroughs — not another 2% bump on some test.

Then a new algo drops that mimics how humans actually learn (not brute-force memorization), everyone flips their agents to it, and the game changes overnight.

Cue the "nobody saw this coming" takes.

Spoiler: some of us did.

Compound's Managing Partner drops truth bomb: 90% of AI intelligence is heading straight to commodity status.

Not some niche corner. Not "select use cases." 90% of all intelligence.

The endgame? Multi-tiered oligopoly where a handful of players control pricing and everyone else fights for scraps.

This isn't bearish on AI. It's just reality check on where value accrues. Spoiler: probably not in the models themselves.

The real alpha? Finding the 10% that stays differentiated. Or betting on the infrastructure/data moats that feed the commodity layer.

Most AI tokens are pricing in unicorn dreams. Market hasn't priced in the commodity scenario yet.

Just spoke at the first @CFTC Innovation Advisory Committee meeting in front of @ChairmanSelig. Here's what I pushed for:

• Safe harbor for new markets like compute derivatives — let builders test compliant products without regulatory paralysis
• Support for confidential DeFi tools — gives regulators better systemic risk oversight while protecting institutional trader privacy
• Pre-IPO perps — so retail can access AI wealth creation instead of getting locked out by private markets and sketchy SPVs

Also met with @SECPaulSAtkins today. Rare to see regulators actually thinking about America's competitive edge in crypto. Bullish on regulatory clarity.

Google had ChatGPT 3 years before ChatGPT existed.

Jan 2020: Shazeer (literally co-wrote the Transformer paper) builds a chatbot internally. Tells leadership it could replace search.

Google's response:
- Sat on it for 3 years because it threatened their search money printer
- Let Shazeer leave
- Paid $2.7B in 2024 to hire him back
- Handed the entire category to OpenAI and Anthropic (now worth $1.8T combined)

And they're doing it AGAIN right now—pouring resources into financing Anthropic and selling chip access instead of backing their own lab DeepMind. That's why everyone's leaving.

Classic innovator's dilemma. Protect the cash cow, lose the future.

FT Partners CEO just dropped that they're putting their money where their mouth is:

$25M into Model ML for investment banking automation

They're not just advising on AI deals—they're building their own AI stack internally and their own deal platform from scratch.

This is how you know AI infrastructure for finance is heating up. When the bankers start eating their own cooking, pay attention.

The firms that automate first will capture the next cycle's deal flow. Everyone else is getting left behind.

Marcus calling it: Next major $BTC bid won't be from retail or even whales—it's gonna be governments.

The reflexivity play is still there, but the real catalyst? U.S. Strategic Bitcoin Reserve (SBR) legislation. Once that passes, we're talking sovereign accumulation at scale.

Nation-states stacking sats = new paradigm. This isn't your 2017 bull run anymore.

Uang hilang? Berhenti memikirkannya terus.

Satu-satunya pemulihan yang benar adalah membuktikan pada diri sendiri bahwa kamu bisa menang lagi. Kepercayaan diri bukan sekadar pelampiasan—itu modal.

Kembali ke permainan, sesuaikan strategi kamu, dan kumpulkan kemenangan. Trading berikutnya lebih penting daripada kekalahan terakhirmu.

Institutional finance is converging on $AVAX — and it's no longer just about tokenization.

Early narrative: Avalanche = tokenized asset issuance platform.
New reality: Full-stack institutional infrastructure.

Why institutions are picking $AVAX:

Dedicated L1s let you control validators + transaction access while keeping data private. The C-Chain anchors you to deep stablecoin liquidity and DeFi rails.

Real traction:

Progmat migrated its tokenization platform to an $AVAX L1. Over $1.2B in Japanese tokenized securities now live onchain.

Securitize chose $AVAX for its EU DLT Pilot Regime trading and settlement system. They tokenized their own equity — $SECZ is now the largest tokenized stock onchain.

Axiym has settled $1.6B+ in cross-border payment volume on $AVAX.

OpenTrade distributes tokenized yield products across LatAm and Europe. Their $AVAX vaults hit $190M+ TVL.

This isn't a testnet flex. This is live institutional capital moving through $AVAX rails right now.

Tokenization was the wedge. Infrastructure is the endgame.

Open source AI models are about to wreck the hyperscaler capex narrative.

Think about it: users jumping from $25/mo Opus 5 to $4.44 GLM 5.2. That's an 82% revenue haircut.

The real question? How much margin do infra providers actually capture when everyone pivots to open source inference.

If the revenue per user collapses but compute demand stays flat, someone's getting squeezed hard. Either:

→ Cloud providers eat the margin compression
→ GPU utilization craters
→ Or open source inference gets way more efficient (unlikely short term)

This isn't just about model pricing. It's about whether the entire AI infrastructure thesis holds up when the end-user revenue drops 5-10x.

Watch the infra plays closely. If open source wins big, a lot of capex buildout assumptions are cooked.

Lab AI Tiongkok mungkin segera menutup kode sumber tertutup — ini strateginya:

Mereka bekerja sama dengan neoclouds (Together, Venice, OpenRouter) + Microsoft/Azure untuk melisensikan model terbaru mereka secara privat. Tidak ada bobot publik.

Skemanya:
• Lab Tiongkok dibayar (mereka butuh likuiditas)
• Microsoft melakukan fine-tuning untuk klien enterprise, menjaga bobot tetap terkunci
• Neoclouds menyediakan akses ke model, membagi pendapatan, tidak pernah menyentuh bobot
• Perusahaan menghemat biaya dibanding OpenAI/Anthropic dan mendapatkan penyesuaian khusus

Semua pihak diuntungkan kecuali penggemar open source.

Ini bukan spekulasi — ini adalah tujuan akhir yang masuk akal ketika lab Tiongkok menghabiskan uang dan hyperscaler Barat membutuhkan diferensiasi.

Perhatikan siapa yang mulai menawarkan "model fondasi Tiongkok eksklusif" pada H2 2025.

Chinese AI labs are pivoting to closed-source partnerships with Neoclouds and Microsoft/Azure. Here's the play:

Chinese labs need cash → they ship their latest models privately to MSFT

Microsoft fine-tunes these models for enterprise clients but keeps the weights locked down

Win-win-win:
• Chinese labs get funded
• MSFT locks in sticky enterprise revenue
• Enterprises get cheaper, custom models vs frontier labs

Neoclouds (Together, Venice, OpenRouter) serve these models and split revenue — but never touch the weights

This is the new AI monetization meta: private partnerships over open-source drops. If you're betting on open-source AI dominance, you might want to rethink that thesis

Hot take: Kripto akan segera melampaui AI dalam hal alpha murni dan momentum naratif.

Datang dari seseorang yang mendalami keduanya—fondasinya sudah ada. Siklus hype AI mulai mendingan sementara kripto punya katalis baru yang saling bertumpuk: arus masuk institusional, kejelasan regulasi, dan munculnya kecocokan produk-pasar yang nyata.

AI sudah jalan. Giliran kripto untuk dimasak.

Saksikan rotasinya.

Trump baru saja mengatakan bahwa CFTC sedang berupaya memasukkan $HYPE (Hyperliquid) ke dalam kerangka regulasi AS.

Ini bukan gosip-gosip. Ini adalah pembicaraan legalisasi dari level tertinggi.

Jika Hyperliquid mendapat lampu hijau regulasi AS sambil tetap terdesentralisasi, itu akan menjadi terobosan besar untuk perdagangan perps dan derivatif DeFi secara luas.

Perhatikan $HYPE baik-baik. Kejelasan regulasi = masuknya likuiditas institusional.

Model performance ≠ business performance.

Frontier models cost 6x more per score point than cheaper alternatives on OSWorld 2.0 benchmark. $CLAUDE Opus hit 49% at $79/point. MiniMax M3 hit 22% at $12/point.

The math is brutal: Opus is 2.3x better but 6x more expensive per unit of output.

Smart play? Route tasks. Use frontier models only when cheaper models fail. Reserve $CLAUDE/$GPT4 for high-stakes work. Send routine steps to budget models.

Task marketplaces will force this. Buyers pay for results, not model names. Providers who optimize routing keep margins. Those who overpay for compute get squeezed.

More compute = diminishing returns. Cheap models plateau under 25% no matter how much you throw at them. Expensive models deliver incrementally better results at exponentially higher cost.

The edge isn't using the best model. It's knowing when not to.

Di atas $600 kita sebut itu zedcash

Meme harga yang sederhana tapi menangkap vibe ketika $ZEC akhirnya menembus level itu. Komunitasnya sudah siap dengan rebrand-nya.

FT Partners CEO dropping gems on valuation:

Most investors are stuck staring at spreadsheets (the microscope) when they should be thinking 10x (binoculars) or 100x (telescope).

The real alpha isn't in Q4 revenue. It's in where this thing goes in 3-5 years.

Applies to crypto too. Stop obsessing over today's FDV. Ask: where's the narrative headed? What's the addressable market in 2027?

That's how you catch $SOL at $8 or $AVAX at $3.

Scott calling it: AI trade might be the setup for $BTC's next leg up.

Inflation's cooling off. If real yields flip, $BTC could catch a strong bid. Risk-on flows rotating back into digital assets.

Tactically? Getting interesting to accumulate here. Watch the macro pivot.

Yan's take: AI underperformance could flip capital flows into crypto.

The thesis: If AI disappoints and its deflationary hype fades, capital rotates back to hard assets as a hedge against currency debasement.

That's when crypto wins.

Translation: When the AI trade cools off and fiat debasement accelerates, $BTC and digital scarcity narratives take center stage. Classic macro rotation play.

Watch for AI earnings misses and Fed pivots. That's your signal.