The numbers don’t add up.
A short industry briefing from last week stated Instinct, the personal AI assistant startup, raised $2.5 billion and is now seeking an additional $10 billion, pushing total funding to $3.5 billion. Basic arithmetic kills that claim: 2.5 + 10 equals 12.5, not 3.5. The headline says “an extra $10 billion,” so the real total is $12.5 billion. This isn’t a typo. It’s a symptom. The entire narrative around Instinct suffers from the same broken math.
The macro shifts. The chart follows.
Let’s audit the architecture. Instinct is not a foundation model company. It uses open-source models heavily — likely Meta’s Llama family. That means Meta is both the model supplier and a direct competitor. Platform dependency risk is not a footnote; it is the core of the balance sheet. In my 2020 NLockdown audit of Compound Finance, I learned that code is law only if mathematically sound. Instinct’s codebase has no mathematical moat. It is a consumer layer built on rented ground.
Context: The global liquidity map for AI startups is shifting. Central banks are tightening or holding, and venture dollars are fleeing to hard assets. Instinct’s $2.5 billion raise came at a $100 billion valuation — a 40x multiple on zero revenue. That is not a bet on technology. It is a bet on narrative. The additional $10 billion is framed as needed for compute capacity: buying chips, building data centers. The founder, Noah Shinn, wants to operate his own infrastructure. But $10 billion is pocket change in the AI infrastructure world. Microsoft is spending $80 billion on data centers this year alone. Instinct’s capital is a rounding error.
Core: Instinct’s technology stack is a bundle of engineering decisions, not research breakthroughs. Evidence shows: - Heavy reliance on open-source models. No evidence of fine-tuning, distillation, or post-training. - Service has been at full capacity multiple times, with response delays. That is not demand overflow. It is an inference capacity bottleneck. - The founder’s stated plan to buy chips and run data centers confirms they are currently renting cloud servers. Renting means variable cost, no margin control.
In my ZK-rollup latency study on StarkNet, I demonstrated that cryptographic efficiency directly correlates with global trade velocity. Instinct is not using cryptography to differentiate. It is using off-the-shelf language models. The latency problem here is not cryptographic—it is engineering debt.
The machine economy is coming. AI agents will transact autonomously. But Instinct’s technology cannot scale to that. Their current architecture cannot even handle a surge of human users.
Contrarian angle: The decoupling thesis is that Instinct is an independent AI company. The truth is that it is a derivative asset. Meta’s free assistant, Muse, can shop, book flights, and handle daily tasks. Community testers say Muse is faster. Meta has self-developed models, its own compute, and a social distribution layer. Instinct has none of that. The trust narrative—that Instinct is a startup with superior privacy—is a liability. Trust is a liability, not an asset. Meta can offer the same features for free because it monetizes through advertising and ecosystem lock-in. Instinct cannot.
In my Swiss regulatory negotiation with FINMA, I pushed for recognition of zero-knowledge proofs for compliance. Instinct has no such privacy infrastructure. Their personal assistant handles sensitive data—calendars, payments, travel—but there is no evidence of encryption, access control, or red-team testing. The security risk is catastrophic. Prompt injection attacks could enable unauthorized transactions. Data leakage from open-source model fine-tuning is unaddressed.
Takeaway: Instinct’s valuation is a bet that compute bottlenecks will be solved by capital. But compute is not the bottleneck. Trust is. And user trust is already migrating to the free, faster, integrated option from Meta. The macro environment is squeezing high-burn, low-differentiation plays. Instinct’s $100 billion valuation will be repriced in 18 months or it will be acquired by one of the big five. The machine wants execution, not PowerPoint promises. Instinct is not ready for the machine economy.
Ledgers don’t.
