NVIDIA fell 5.5% — largest drop since April — after a modest 3% earnings beat the market read as deceleration. CoreWeave missed. Dell crushed it. Salesforce announced a $50B buyback with Agentforce at $800M ARR. Iran talks ended without a deal but with technical teams heading to Vienna. Trump added another 10% on China starting March 4. February ends with the Nasdaq down 2.5% — worst month since March.
Crypto data provided by CoinGecko
NVIDIA fell 5.5% Thursday — largest single-day drop since April — despite beating Q4 revenue by ~3% ($68.1B vs $66.2B consensus). Morgan Stanley called it "the largest and cleanest beat in semis industry history." The market disagreed. A 3% beat is a narrowing beat by NVIDIA's standards (15-20%+ in prior quarters). The sell-off tells you the market is no longer rewarding beats — it's punishing deceleration. — Full analysis in Big Story #1.
CoreWeave fell 11% after hours — revenue beat by 1.3% ($1.57B) but EPS missed and Q1 guide of $1.9-2.0B came in well below the $2.29B consensus. $66.8B revenue backlog is massive, but the market wanted profitability, not just growth. The GPU-as-a-service model has demand. It doesn't yet have margins. — AI capex thesis test, Big Story #7.
Dell crushed Q4: $33.4B revenue (+39% YoY), EPS $3.89 (beat by 13%), and guided FY27 to $138-142B vs Street's $120B. Record revenue, record FCF at $11.2B. Dell is the quiet winner of the AI infrastructure cycle — server revenue surging while NVIDIA gets the headlines and the sell-off. The 18% beat on FY27 EPS guidance ($12.90 vs $10.56 consensus) is the kind of number that deserves the word "crushed."
Block +21% premarket after announcing it's laying off ~4,000 employees — roughly half its workforce. Cost up to $500M. When Jack Dorsey's company gets its biggest pop in years from firing people, the market is telling you what it values right now: margins over growth, discipline over ambition.
Trump announced an additional 10% tariff on China starting March 4. On top of the existing 15% Section 122 tariff. Effective rate on Chinese imports heading toward 25%. The tariff regime that SCOTUS was supposed to constrain is being rebuilt brick by brick under different legal authority.
10Y yield fell to 4.01% Thursday and heading below 4% Friday — approaching levels not seen since September. Flight to safety as tech sells off and Iran uncertainty persists. When rates fall while the Fed is on hold, the bond market is pricing something the equity market hasn't acknowledged yet.
$8.72B in BTC and ETH options expiring today — February's largest derivatives event. Max pain levels will determine whether the $68K level holds into the weekend. Elevated volatility expected through the close. — Bear market update in Big Story #3.
Citi launching institutional Bitcoin custody, wallet management, and private key services in 2026. Morgan Stanley building similar infrastructure. When the two largest US wealth managers both build Bitcoin custody in the same quarter, the institutional plumbing for the next cycle is being laid — regardless of where the price goes this month.
MARA (Bitcoin miner) signed a deal with Starwood to convert mining facilities to AI data centers. The pivot is accelerating. When miners stop mining and start hosting GPUs, the Bitcoin hashrate narrative changes and the AI infrastructure narrative gets another supply source.
BTC holding ~$68K ahead of options expiry. Weekly close above the 200-period EMA (~$68K) is the bull case for March. Close below and the $60K retest is back on the table. The options expiry could provide the catalyst in either direction.
Jensen Huang: "I think the markets got it wrong" on the AI-software fear trade. In a post-earnings interview, Huang pushed back on the narrative that AI agents will destroy SaaS companies. His argument: AI creates more demand for software, not less. The CEO of the company selling the picks has an obvious incentive to say the gold rush isn't over. But the rebuttal is worth tracking — if he's right, the SaaS repricing is an overshoot.
Intuit Q2: revenue $4.7B (+17% YoY), EPS $4.15 (+25% YoY), and announced a multi-year partnership with Anthropic to integrate Claude into Intuit Enterprise Suite. TurboTax and QuickBooks seeing "rapid adoption" of AI features. Intuit is the test case for the "AI-augmented incumbent" thesis — growing faster because of AI, not despite it.
Autodesk Q4: revenue $1.96B (beat by 2.6%), EPS $2.85 (beat by 8%), FCF +54% to $2.4B. Billings +30% YoY. Construction and manufacturing software demand accelerating. Autodesk is proving that verticalized software with deep workflow integration resists the AI agent threat. The SaaS repricing is indiscriminate — the market will eventually discriminate.
Salesforce Q4: $11.2B revenue (+12%), EPS $3.81 (beat by $0.76), and a $50B buyback — the largest in enterprise software history. Agentforce hit $800M ARR, up 169% YoY. But here's the tension: organic growth ex-Informatica is still high single digits, and the stock is down 36% over 12 months at ~$199. The $50B buyback is Salesforce telling you it believes its stock is cheap. The market is asking whether Agentforce is real revenue or demo magic. The answer matters for every SaaS company betting on AI agents. — SaaS repricing update, Big Story #2.
Zscaler -9% — deferred revenue and billings missed in fiscal Q2. Cybersecurity was supposed to be the AI-resistant SaaS category. If Zscaler is missing, the "security is special" thesis needs revision.
Iran talks ended without a deal — "most intense" round yet, with technical teams assigned to meet in Vienna next week. Araghchi: "We identified the main elements of a possible agreement." Witkoff and Kushner were "disappointed" after the morning session. The gap: US wants enriched uranium stockpile surrendered and an indefinite deal; Iran insists on continued enrichment under IAEA oversight. Technical discussions in Vienna = progress. Kushner's disappointment = timeline pressure. Both can be true. — Full update in Big Story #4.
DHS shutdown Day 13 — paychecks due today. CBP diverting funds from the "One Big Beautiful Bill Act" to pay ~57,600 employees. FEMA still halted on non-disaster response. The workaround funding shows the shutdown is now a political performance, not a fiscal one — the money exists, the deal doesn't. — Details in Big Story #5.
Trump tariffs on China increasing another 10% starting March 4. Effective rate heading toward 25%. The Section 122 framework allows escalation within the 150-day window. Every week brings another brick. Businesses can't plan when the rate changes monthly.
NVIDIA fell 5.5% on a beat. Dell guided 18% above consensus and the stock will gap up. Block is surging 21% because it's firing half its people. CoreWeave missed on margins and dropped 11%.
Read those four sentences again. They're not random. They're a new scoring system.
The old scoreboard was simple: beat estimates, go up. Revenue growth was king. Miss growth and the market punished you. Beat growth and you got rewarded. This worked for the entire post-COVID cycle and through the first phase of the AI trade. NVIDIA was the purest expression — beat after beat, each one bigger, stock up every time.
The new scoreboard has three columns: growth, margins, and the ratio between capex commitment and visible returns. NVIDIA beat on growth (revenue +3% vs consensus) but the beat is narrowing quarter over quarter. The market didn't sell NVIDIA because the numbers were bad. It sold because the trajectory of beats is decelerating — from 20%+ beats to 3%. That's not failure. It's maturation. And the market prices maturation differently than it prices hypergrowth.
Dell is the proof that the new scoreboard rewards differently. $33.4B revenue, $3.89 EPS (beat by 13%), FY27 guide 18% above Street. Record FCF of $11.2B. Dell isn't a hypergrowth story — it's a disciplined infrastructure company catching the AI tailwind with operational leverage. The market is rotating from "who's growing fastest" to "who's converting growth into cash."
Block is the extreme case. Firing 4,000 people — half the company — and the stock jumps 21%. The market isn't celebrating layoffs. It's celebrating a company choosing margins over headcount growth in an environment where AI makes many of those roles redundant. Block is doing voluntarily what AI will force on others involuntarily. The market is pricing in the end state.
CoreWeave is the cautionary tale. Revenue backlog of $66.8B is extraordinary. But the Q1 guide ($1.9-2.0B vs $2.29B consensus) and the EPS miss tell you the market wants to see the GPU-as-a-service model produce margins, not just revenue. Demand without profitability is a story, not a business — and the market is done paying for stories.
Salesforce is the most interesting case because it's trying to play both sides. $50B buyback (margins and shareholder returns) plus Agentforce at $800M ARR (growth story). The stock is at $199, down 36% in 12 months. The market hasn't decided whether Salesforce is an AI winner or an AI casualty — and the $50B buyback is management's bet that the market is wrong. If Agentforce's organic growth reaccelerates above 10%, the buyback looks genius. If it doesn't, it's capital allocation theater.
The framework: We're in a regime transition from growth-at-any-cost to profitable-growth-with-visible-returns. This transition happens every cycle but this one has a unique accelerant: AI both creates the growth opportunity and compresses the timeline for proving returns on it. The companies that navigate this — Dell, Intuit (growing 17% with AI integration), Autodesk (FCF +54%) — will outperform. The companies that can't demonstrate the return path — even if their revenue is growing — will get sold.
What this means for the next 6 months: Every earnings call becomes a margin story, not a revenue story. NVIDIA's next quarter isn't about beating $78B — it's about showing that Blackwell-to-Rubin margins expand. Salesforce isn't about Agentforce ARR — it's about organic growth reaccelerating. CoreWeave isn't about backlog — it's about proving unit economics. The scoreboard changed. Most management teams haven't noticed yet.
"We suffer more often in imagination than in reality."
— Seneca
Friday energy is different. There's a pull to look back at the week and inventory everything that didn't go the way you planned. The meeting that went sideways. The thing you said that came out wrong. The decision you're still second-guessing.
Seneca's point isn't that your problems aren't real. It's that the rehearsal of problems is almost always worse than the problems themselves. The replaying. The projecting. The building of a catastrophe from a setback.
Most of what you're worried about right now won't matter in a month. Not because it isn't real — because it will resolve itself in ways you can't predict from here. The suffering you're doing right now about it, though? That's real and it's happening now.
Before the weekend starts, write down the three things from this week that are taking up the most mental space. For each one, ask: "Has the worst version of this actually happened, or am I rehearsing it?" Let the real ones stay. Let the rehearsals go.
# ▸ THE TAKE
The Market Is Building a New Scoreboard — and Most Companies Don't Know the Rules Yet
NVIDIA fell 5.5% on a beat. Dell guided 18% above consensus and the stock will gap up. Block is surging 21% because it's firing half its people. CoreWeave missed on margins and dropped 11%.
Read those four sentences again. They're not random. They're a new scoring system.
The old scoreboard was simple: beat estimates, go up. Revenue growth was king. Miss growth and the market punished you. Beat growth and you got rewarded. This worked for the entire post-COVID cycle and through the first phase of the AI trade. NVIDIA was the purest expression — beat after beat, each one bigger, stock up every time.
The new scoreboard has three columns: growth, margins, and the ratio between capex commitment and visible returns. NVIDIA beat on growth (revenue +3% vs consensus) but the beat is narrowing quarter over quarter. The market didn't sell NVIDIA because the numbers were bad. It sold because the trajectory of beats is decelerating — from 20%+ beats to 3%. That's not failure. It's maturation. And the market prices maturation differently than it prices hypergrowth.
Dell is the proof that the new scoreboard rewards differently. $33.4B revenue, $3.89 EPS (beat by 13%), FY27 guide 18% above Street. Record FCF of $11.2B. Dell isn't a hypergrowth story — it's a disciplined infrastructure company catching the AI tailwind with operational leverage. The market is rotating from "who's growing fastest" to "who's converting growth into cash."
Block is the extreme case. Firing 4,000 people — half the company — and the stock jumps 21%. The market isn't celebrating layoffs. It's celebrating a company choosing margins over headcount growth in an environment where AI makes many of those roles redundant. Block is doing voluntarily what AI will force on others involuntarily. The market is pricing in the end state.
CoreWeave is the cautionary tale. Revenue backlog of $66.8B is extraordinary. But the Q1 guide ($1.9-2.0B vs $2.29B consensus) and the EPS miss tell you the market wants to see the GPU-as-a-service model produce margins, not just revenue. Demand without profitability is a story, not a business — and the market is done paying for stories.
Salesforce is the most interesting case because it's trying to play both sides. $50B buyback (margins and shareholder returns) plus Agentforce at $800M ARR (growth story). The stock is at $199, down 36% in 12 months. The market hasn't decided whether Salesforce is an AI winner or an AI casualty — and the $50B buyback is management's bet that the market is wrong. If Agentforce's organic growth reaccelerates above 10%, the buyback looks genius. If it doesn't, it's capital allocation theater.
The framework: We're in a regime transition from growth-at-any-cost to profitable-growth-with-visible-returns. This transition happens every cycle but this one has a unique accelerant: AI both creates the growth opportunity and compresses the timeline for proving returns on it. The companies that navigate this — Dell, Intuit (growing 17% with AI integration), Autodesk (FCF +54%) — will outperform. The companies that can't demonstrate the return path — even if their revenue is growing — will get sold.
What this means for the next 6 months: Every earnings call becomes a margin story, not a revenue story. NVIDIA's next quarter isn't about beating $78B — it's about showing that Blackwell-to-Rubin margins expand. Salesforce isn't about Agentforce ARR — it's about organic growth reaccelerating. CoreWeave isn't about backlog — it's about proving unit economics. The scoreboard changed. Most management teams haven't noticed yet.
Every action creates ripple effects beyond its immediate, intended impact. Second-order effects are the consequences of consequences — the indirect results that emerge after the first-order effect plays out. They are difficult to predict because they involve complex chains of causation across interconnected systems. What makes them powerful and often dangerous is the time delay: second-order effects frequently don't become visible until long after the original action, making it hard to connect cause and effect.
Look past the immediate result of any intervention. Resistance to change often comes from people affected by second-order consequences, not the first-order ones. Resistance to self-checkout in retail isn't about the technology — it's about the downstream job losses, reduced human interaction, and increased theft that emerge months later.
# ▸ THE BIG STORIES The macro trends that matter through the daily noise. Updated when news moves the needle. Silent when it doesn't.
Current state: NVIDIA -5.5% on a 3% revenue beat. Largest drop since April. Nasdaq -1.18%. Semiconductor sector broadly weak. Today's update: The sell-off confirmed what the modest beat implied: the market is no longer rewarding NVIDIA for beating — it's evaluating whether the beat trajectory justifies the multiple. Revenue beat narrowed from 20%+ to 3% over four quarters. Morgan Stanley calling it "cleanest beat in semis history" doesn't matter when the market is scoring on a different rubric. Jensen's pushback that "markets got it wrong" on SaaS disruption is notable but self-interested. The question for March: does the $78B Q1 guide and Vera Rubin ramp restore confidence, or has the AI hardware trade peaked on expectations?
Current state: Software ETFs down 10%+ in February. Indiscriminate sell-off starting to discriminate. Today's update: Salesforce Q4 ($11.2B, EPS beat by $0.76, $50B buyback, Agentforce $800M ARR) is the biggest data point this week. Intuit (+17% revenue, Anthropic partnership) and Autodesk (FCF +54%, billings +30%) both beat. Zscaler missed. Jensen says AI creates more software demand, not less. The SaaS sell-off is entering Phase 2: the market sold everything in January-February. Now it starts separating the companies AI helps from the companies AI kills. Salesforce's $50B buyback is management betting the sell-off overshoots. Intuit and Autodesk are early evidence of "AI-augmented incumbents" working. Zscaler's miss punctures the "security is special" thesis.
Current state: BTC ~$68,100. $8.72B options expiring today. Weekly close at the 200-period EMA is the technical inflection. Today's update: BTC holding $68K but directionless ahead of the largest February options expiry. Citi and Morgan Stanley both building institutional custody — structural plumbing for the next cycle. MARA pivoting mining facilities to AI data centers. The institutional infrastructure is being built while the price goes nowhere. If BTC closes above ~$68K, March bullish case activates. Below = $60K retest. Options expiry provides the volatility catalyst.
Current state: Third Geneva round ended without agreement. Technical teams assigned to meet in Vienna at IAEA HQ next week. Today's update: "Most intense" round yet. Iran identified "main elements of a possible agreement." US envoys Witkoff and Kushner were "disappointed" after the morning session — unclear if afternoon improved their assessment. The key gap: US wants Iran's 10,000kg enriched uranium stockpile surrendered and an indefinite deal. Iran wants continued enrichment under IAEA monitoring. Technical talks in Vienna are progress — they mean both sides see enough overlap to put details on paper. But "disappointed" envoys reporting back to Trump raises military strike probability if Vienna doesn't produce framework agreement. Oil at $70 still prices diplomacy. Two carrier groups still in theater.
Current state: Paychecks due today for 260,000+ DHS employees. CBP diverting funds to cover 57,600 workers. Today's update: The funding workaround (CBP redirecting money from existing legislation) proves the shutdown is political theater, not fiscal necessity. The money is there — the deal isn't. Underlying issue remains ICE reform after the Minneapolis shooting. No negotiation framework exists. The 43-day shutdown last fall plus this 13-day gap means DHS employees have spent nearly two months of FY26 in shutdown conditions. Institutional degradation is the compounding risk.
Current state: Gold $5,185. Silver $90.41. 10Y below 4%. Today's update: Gold pulling back slightly from $5,231 — healthy consolidation after +43% over 12 months. Silver above $90 and holding. The new development: 10Y yield dropping below 4% for the first time since September. When rates fall, the dollar weakens, and gold is already elevated, the floor keeps rising. The bond market's flight-to-safety bid is reinforcing the gold bid. February ends with gold up ~6%, silver up ~7%, and both metals structurally overbought but with no catalyst for reversal.
Current state: NVIDIA supply commitments $95.2B. TSMC capex $52-56B. CoreWeave: demand without margins. Today's update: CoreWeave's miss is the most important data point for the AI infrastructure thesis this week — more important than NVIDIA's beat. CoreWeave proved demand exists ($66.8B backlog) but can't yet prove unit economics work at scale. Dell proved the opposite — disciplined infrastructure with record FCF. The AI capex cycle has demand. The question is now who captures margins. Dell's answer: hardware with operational leverage. CoreWeave's answer: not yet. NVIDIA's answer: depends on Rubin transition.
Current state: Section 122 tariffs at 15%. Additional 10% on China starting March 4. Effective rate heading to 25%. Today's update: Trump announced the China tariff increase despite the SCOTUS constraint on IEEPA. The Section 122 framework allows escalation within the 150-day window. Each week brings higher rates. The tariff regime SCOTUS was supposed to constrain is being rebuilt through different legal authority — slower, but directionally the same. Businesses can't plan costs when rates change monthly. The March 4 effective date means supply chains are repricing this weekend.
Remaining Big Stories — no change today: The Fed's Impossible Position, Humanoid Robotics, Crypto Regulatory Clarity, India Energy Realignment, US-China Tech Decoupling, Nuclear Renaissance, Strategy BTC Treasury Risk, Silver Supply Deficit, AI Model Architecture Shift, Japan Monetary Policy, European Defense Spending, US Fiscal Trajectory, Global Dollar System, Executive Authority Under Legal Siege, Mexico Cartel Destabilization, Political Violence Trend.
# ▸ TOMORROW'S HEADLINES What the market will be talking about in 12-18 months.
New evidence today: NVIDIA sold on a beat. Dell rewarded for margins and FCF. Block surging on mass layoffs. CoreWeave punished for growth without margins. In 12 months, "profitable AI deployment" will be the phrase on every earnings call. The companies that figure out the unit economics of AI first will define the next phase of the trade. The hypergrowth-at-any-cost era is ending.
New evidence today: Citi and Morgan Stanley both building Bitcoin custody and trading services simultaneously. When the two largest US wealth managers commit to crypto infrastructure in the same quarter, it's not speculation — it's plumbing for a financial system that includes digital assets by default. 12-18 months from now, institutional crypto access will be unremarkable. The price impact lags the infrastructure by 6-12 months.
Full reference list (23 items) unchanged — see bottom of brief.
# ▸ THE WATCHLIST Regime-changing 2-10x opportunities. Small bets, big asymmetry. Most will be wrong.
This section is purely illustrative — not investment advice. These are structural theses applied to specific assets to test our frameworks against real markets. Do not invest in anything because it appears here. Do your own work. Size accordingly.
ETN — Software company disguised as an industrial ~$380. Data center orders +200%. Electrical backlog $15.3B. The insight: Every data center needs recurring power management software that isn't broken out. Re-rates from "industrial 29x" to "hybrid 35x+." If the AI-energy convergence plays out, 3-5x over 3-5 years. Upside: Software reclassification at 35x = $560 (47%). AI energy narrative = $700+ (80%+). Downside: Buildout slows, range-bound (~10%). Validates: Software-attached revenue reported. Data center >25% of revenue. Rejects: Margins flat. No software breakout through 2027. Feb 27 note: Dell's record FCF from AI infrastructure validates the "disciplined infrastructure winner" archetype. ETN's power management sits in the same lane — essential physical infrastructure with software margin upside.
SOL — Agent payment rails ~$86, down 75% from ATH. The insight: AI agents need micropayments at machine speed. Banks can't clear $0.001. Solana can. Leading all chains in payment volume during a bear. Staking ETFs creating institutional floor. Upside: Agent commerce + cycle recovery = $200-300 (2.5-3.5x). Machine commerce real = $500+ (6x). Downside: Agents use Stripe. SOL to $40 (~50%). Validates: Agent tx volume on Solana. x402 >$1B cumulative. ETF inflow trend persists. Rejects: Agent commerce centralizes. Reliability issues. Feb 27 note: Citi and Morgan Stanley building crypto infrastructure strengthens the institutional floor thesis. SOL -2.6% on the day — tracking broader crypto weakness, not SOL-specific.
Harmonic Drive (6324.T) — Humanoid picks-and-shovels ~¥3,800. The insight: Every humanoid platform needs the same precision actuators. Near-monopoly on strain wave gears, minimal Western coverage. Wins regardless of which platform wins. Upside: 500K units/yr by 2028, 20-40 actuators each. Revenue triples = 3x+. Downside: Timeline slips. Chinese competitors. 30-50% down. Validates: Tesla ships externally. Humanoid OEM backlog growth. Rejects: Production <10K in 2026.
Gulf Sovereign AI — Thematic, watching for vehicle The insight: Gulf running oil playbook with compute. Cheapest energy, 3.5B people need AI outside US/China. 95% of AI allocation is US tech. UAE at 14x vs S&P 21x. Upside: G42 IPO creates investable category. 30%+ broad. Downside: Captured by hyperscalers. 10-15%. Validates: G42 IPO. Sovereign AI deals bypass AWS/Azure. Rejects: Hyperscalers go direct. No differentiated tech.
The Friendship Paradox Explains Why You Always Feel Behind — In 1991, sociologist Scott Feld proved something counterintuitive: on average, your friends have more friends than you do. This isn't about popularity — it's a mathematical certainty arising from network structure. People with many connections appear in more people's friend lists, biasing the sample upward. The paradox extends far beyond social networks. Your coworkers are, on average, more productive than you (because the most productive people work on more teams, so you're more likely to encounter them). The papers you read are, on average, more cited than your papers (because highly cited papers are more visible). The startups in the news are, on average, more successful than most startups. The systematic bias isn't that you're underperforming — it's that your sample of comparison points is skewed toward outliers. Every network creates this illusion. The practical implication: if you feel like everyone around you is doing better, your sampling method is broken, not your performance. The fix isn't to work harder. It's to understand that your comparison set is mathematically guaranteed to make you feel inadequate.
Current state: NVIDIA -5.5% on a 3% revenue beat. Largest drop since April. Nasdaq -1.18%. Semiconductor sector broadly weak.
Today's update: The sell-off confirmed what the modest beat implied: the market is no longer rewarding NVIDIA for beating — it's evaluating whether the beat trajectory justifies the multiple. Revenue beat narrowed from 20%+ to 3% over four quarters. Morgan Stanley calling it "cleanest beat in semis history" doesn't matter when the market is scoring on a different rubric. Jensen's pushback that "markets got it wrong" on SaaS disruption is notable but self-interested. The question for March: does the $78B Q1 guide and Vera Rubin ramp restore confidence, or has the AI hardware trade peaked on expectations?
Current state: Software ETFs down 10%+ in February. Indiscriminate sell-off starting to discriminate.
Today's update: Salesforce Q4 ($11.2B, EPS beat by $0.76, $50B buyback, Agentforce $800M ARR) is the biggest data point this week. Intuit (+17% revenue, Anthropic partnership) and Autodesk (FCF +54%, billings +30%) both beat. Zscaler missed. Jensen says AI creates more software demand, not less. The SaaS sell-off is entering Phase 2: the market sold everything in January-February. Now it starts separating the companies AI helps from the companies AI kills. Salesforce's $50B buyback is management betting the sell-off overshoots. Intuit and Autodesk are early evidence of "AI-augmented incumbents" working. Zscaler's miss punctures the "security is special" thesis.
Current state: BTC ~$68,100. $8.72B options expiring today. Weekly close at the 200-period EMA is the technical inflection.
Today's update: BTC holding $68K but directionless ahead of the largest February options expiry. Citi and Morgan Stanley both building institutional custody — structural plumbing for the next cycle. MARA pivoting mining facilities to AI data centers. The institutional infrastructure is being built while the price goes nowhere. If BTC closes above ~$68K, March bullish case activates. Below = $60K retest. Options expiry provides the volatility catalyst.
Current state: Third Geneva round ended without agreement. Technical teams assigned to meet in Vienna at IAEA HQ next week.
Today's update: "Most intense" round yet. Iran identified "main elements of a possible agreement." US envoys Witkoff and Kushner were "disappointed" after the morning session — unclear if afternoon improved their assessment. The key gap: US wants Iran's 10,000kg enriched uranium stockpile surrendered and an indefinite deal. Iran wants continued enrichment under IAEA monitoring. Technical talks in Vienna are progress — they mean both sides see enough overlap to put details on paper. But "disappointed" envoys reporting back to Trump raises military strike probability if Vienna doesn't produce framework agreement. Oil at $70 still prices diplomacy. Two carrier groups still in theater.
Current state: Paychecks due today for 260,000+ DHS employees. CBP diverting funds to cover 57,600 workers.
Today's update: The funding workaround (CBP redirecting money from existing legislation) proves the shutdown is political theater, not fiscal necessity. The money is there — the deal isn't. Underlying issue remains ICE reform after the Minneapolis shooting. No negotiation framework exists. The 43-day shutdown last fall plus this 13-day gap means DHS employees have spent nearly two months of FY26 in shutdown conditions. Institutional degradation is the compounding risk.
Current state: Gold $5,185. Silver $90.41. 10Y below 4%.
Today's update: Gold pulling back slightly from $5,231 — healthy consolidation after +43% over 12 months. Silver above $90 and holding. The new development: 10Y yield dropping below 4% for the first time since September. When rates fall, the dollar weakens, and gold is already elevated, the floor keeps rising. The bond market's flight-to-safety bid is reinforcing the gold bid. February ends with gold up ~6%, silver up ~7%, and both metals structurally overbought but with no catalyst for reversal.
Current state: NVIDIA supply commitments $95.2B. TSMC capex $52-56B. CoreWeave: demand without margins.
Today's update: CoreWeave's miss is the most important data point for the AI infrastructure thesis this week — more important than NVIDIA's beat. CoreWeave proved demand exists ($66.8B backlog) but can't yet prove unit economics work at scale. Dell proved the opposite — disciplined infrastructure with record FCF. The AI capex cycle has demand. The question is now who captures margins. Dell's answer: hardware with operational leverage. CoreWeave's answer: not yet. NVIDIA's answer: depends on Rubin transition.
Current state: Section 122 tariffs at 15%. Additional 10% on China starting March 4. Effective rate heading to 25%.
Today's update: Trump announced the China tariff increase despite the SCOTUS constraint on IEEPA. The Section 122 framework allows escalation within the 150-day window. Each week brings higher rates. The tariff regime SCOTUS was supposed to constrain is being rebuilt through different legal authority — slower, but directionally the same. Businesses can't plan costs when rates change monthly. The March 4 effective date means supply chains are repricing this weekend.
Remaining Big Stories — no change today: The Fed's Impossible Position, Humanoid Robotics, Crypto Regulatory Clarity, India Energy Realignment, US-China Tech Decoupling, Nuclear Renaissance, Strategy BTC Treasury Risk, Silver Supply Deficit, AI Model Architecture Shift, Japan Monetary Policy, European Defense Spending, US Fiscal Trajectory, Global Dollar System, Executive Authority Under Legal Siege, Mexico Cartel Destabilization, Political Violence Trend.
New evidence today: NVIDIA sold on a beat. Dell rewarded for margins and FCF. Block surging on mass layoffs. CoreWeave punished for growth without margins. In 12 months, "profitable AI deployment" will be the phrase on every earnings call. The companies that figure out the unit economics of AI first will define the next phase of the trade. The hypergrowth-at-any-cost era is ending.
New evidence today: Citi and Morgan Stanley both building Bitcoin custody and trading services simultaneously. When the two largest US wealth managers commit to crypto infrastructure in the same quarter, it's not speculation — it's plumbing for a financial system that includes digital assets by default. 12-18 months from now, institutional crypto access will be unremarkable. The price impact lags the infrastructure by 6-12 months.
This section is purely illustrative — not investment advice. These are structural theses applied to specific assets to test our frameworks against real markets. Do not invest in anything because it appears here. Do your own work. Size accordingly.
ETN — Software company disguised as an industrial
~$380. Data center orders +200%. Electrical backlog $15.3B.
The insight: Every data center needs recurring power management software that isn't broken out. Re-rates from "industrial 29x" to "hybrid 35x+." If the AI-energy convergence plays out, 3-5x over 3-5 years.
Upside: Software reclassification at 35x = $560 (47%). AI energy narrative = $700+ (80%+).
Downside: Buildout slows, range-bound (~10%).
Validates: Software-attached revenue reported. Data center >25% of revenue.
Rejects: Margins flat. No software breakout through 2027.
Feb 27 note: Dell's record FCF from AI infrastructure validates the "disciplined infrastructure winner" archetype. ETN's power management sits in the same lane — essential physical infrastructure with software margin upside.
SOL — Agent payment rails
~$86, down 75% from ATH.
The insight: AI agents need micropayments at machine speed. Banks can't clear $0.001. Solana can. Leading all chains in payment volume during a bear. Staking ETFs creating institutional floor.
Upside: Agent commerce + cycle recovery = $200-300 (2.5-3.5x). Machine commerce real = $500+ (6x).
Downside: Agents use Stripe. SOL to $40 (~50%).
Validates: Agent tx volume on Solana. x402 >$1B cumulative. ETF inflow trend persists.
Rejects: Agent commerce centralizes. Reliability issues.
Feb 27 note: Citi and Morgan Stanley building crypto infrastructure strengthens the institutional floor thesis. SOL -2.6% on the day — tracking broader crypto weakness, not SOL-specific.
Harmonic Drive (6324.T) — Humanoid picks-and-shovels
~¥3,800.
The insight: Every humanoid platform needs the same precision actuators. Near-monopoly on strain wave gears, minimal Western coverage. Wins regardless of which platform wins.
Upside: 500K units/yr by 2028, 20-40 actuators each. Revenue triples = 3x+.
Downside: Timeline slips. Chinese competitors. 30-50% down.
Validates: Tesla ships externally. Humanoid OEM backlog growth.
Rejects: Production <10K in 2026.
Gulf Sovereign AI — Thematic, watching for vehicle
The insight: Gulf running oil playbook with compute. Cheapest energy, 3.5B people need AI outside US/China. 95% of AI allocation is US tech. UAE at 14x vs S&P 21x.
Upside: G42 IPO creates investable category. 30%+ broad.
Downside: Captured by hyperscalers. 10-15%.
Validates: G42 IPO. Sovereign AI deals bypass AWS/Azure.
Rejects: Hyperscalers go direct. No differentiated tech.
# ▸ DISCOVERY
The Friendship Paradox Explains Why You Always Feel Behind — In 1991, sociologist Scott Feld proved something counterintuitive: on average, your friends have more friends than you do. This isn't about popularity — it's a mathematical certainty arising from network structure. People with many connections appear in more people's friend lists, biasing the sample upward. The paradox extends far beyond social networks. Your coworkers are, on average, more productive than you (because the most productive people work on more teams, so you're more likely to encounter them). The papers you read are, on average, more cited than your papers (because highly cited papers are more visible). The startups in the news are, on average, more successful than most startups. The systematic bias isn't that you're underperforming — it's that your sample of comparison points is skewed toward outliers. Every network creates this illusion. The practical implication: if you feel like everyone around you is doing better, your sampling method is broken, not your performance. The fix isn't to work harder. It's to understand that your comparison set is mathematically guaranteed to make you feel inadequate.
Proposed changes based on today's brief:
NVIDIA: from "beat" narrative to "deceleration" narrative. The 5.5% sell-off on a 3% beat confirms the market is scoring on beat trajectory, not absolute beat. Update Big Story #1 framing from "AI Capex Audit" to acknowledge the maturation signal. The AI hardware trade isn't dead — it's transitioning from hypergrowth to show-me.
New framework: "The New Scoreboard." Growth → Profitable Growth with Visible Returns. NVIDIA (sold on narrowing beat), Dell (rewarded for margins/FCF), Block (surging on layoffs), CoreWeave (punished for growth without margins). This framework applies to every earnings call for the next two quarters.
SaaS repricing entering differentiation phase. Salesforce ($50B buyback, Agentforce $800M ARR) is the biggest test case. Intuit and Autodesk prove AI-augmented incumbents can grow. Zscaler's miss shows even "protected" categories aren't safe. Jensen's "markets got it wrong" pushback is self-interested but worth tracking. Update Thesis #1 to reflect Phase 2: indiscriminate sell-off → selective repricing.
Iran: technical talks in Vienna = meaningful progress. Upgrade from "proposal on table" to "technical framework discussions." But Kushner's "disappointment" narrows the window. If Vienna doesn't produce a framework, military strike probability rises materially.
Tariff escalation: China +10% March 4. Section 122 being used for country-specific escalation. The legal constraint from SCOTUS is being circumvented within its own framework. Update Thesis #7.
Tomorrow's Headlines additions: (1) Growth-to-Profitability Regime Transition. (2) Institutional Crypto Infrastructure Buildout (Citi + Morgan Stanley).
10Y below 4%. The bond market is front-running something. Add to Fed watch — if rates continue falling while the Fed holds, the market is pricing recession risk or a dovish pivot that hasn't been communicated.
Source check: NVIDIA earnings call transcript is now available — confirm SemiAnalysis and Acquired episode exist for Deep Read. CoreWeave and Dell transcripts for deeper analysis. Monitor Vienna technical talks timeline next week. Track March 4 China tariff implementation.
NVDA -5.5% on 3% revenue beat ($68.1B vs $66.2B). Largest drop since April. Beat trajectory narrowing (20%+ → 3%). Jensen: "markets got it wrong." Q1 guide $78B. Vera Rubin H2 2026.
Updated Feb 27.
Salesforce Q4 $11.2B, $50B buyback, Agentforce $800M ARR (+169%). Intuit +17% with Anthropic partnership. Autodesk FCF +54%. Zscaler missed. Market separating AI winners from AI losers.
Updated Feb 27.
BTC ~$68,100. $8.72B options expiring. Weekly close at 200 EMA (~$68K) is the inflection. Citi + Morgan Stanley building custody. MARA pivoting to AI data centers.
Updated Feb 27.
Third Geneva round ended. "Most intense" yet. Technical teams to IAEA HQ in Vienna next week. US envoys "disappointed." Gap: enrichment stockpile + deal duration. Two carrier groups in theater. Oil $70.
Updated Feb 27.
Day 13. Paychecks due today. CBP diverting funds to pay 57,600 workers. No negotiation framework on ICE reforms. Second shutdown of 2026.
Updated Feb 27.
Gold $5,185 (pulling back from $5,231). Silver $90.41. 10Y below 4%. Bond market flight to safety reinforcing gold floor. Feb: gold +6%, silver +7%.
Updated Feb 27.
CoreWeave: revenue beat, EPS miss, Q1 guide below consensus. $66.8B backlog but no margins. Dell: record FCF, guidance +18% above Street. Demand confirmed. Margins are the new question.
Updated Feb 27.
Section 122 at 15%. China +10% starting March 4 (effective ~25%). SCOTUS constraint being circumvented. Monthly rate changes prevent business planning.
Updated Feb 27.
Rates 3.50-3.75%. 10Y at 4.01% and falling toward 4%. Markets pricing cuts, zero hikes. Warsh takes chair May. Bond market front-running dovish pivot. January PCE March 13.
Updated Feb 27.
Tesla Gen 3 mass production. Figure at BMW. 1X at $20K consumer. Watching GTC March 16.
Last updated Feb 20.
GENIUS Act regs due July 18. SEC "Project Crypto" with CFTC. Stablecoin capital rules eased. Citi + Morgan Stanley building custody.
Updated Feb 27.
Potential pivot from Russian oil. Trade talks paused post-SCOTUS.
Last updated Feb 23.
El Mencho killed. No successor identified. Succession battle expected. Port of Manzanillo disrupted. Supply chain impact ongoing.
Updated Feb 27.
Export controls tightening. Two AI ecosystems forming. China tariffs heading to 25% effective.
Updated Feb 27.
AI power demand driving nuclear restart. MARA converting mining to AI data centers adds supply.
Updated Feb 27.
717K BTC at $76K avg. BTC ~$68.1K. Unrealized loss persists. Options expiry day volatility risk.
Updated Feb 27.
6th consecutive year. Silver at $90.41. Industrial + monetary demand. JP Morgan sees $81/oz average for 2026.
Updated Feb 27.
Training diminishing returns. Inference-time compute emerging. Vera Rubin 10x efficiency. NVIDIA cadence: Blackwell → Rubin → Feynman.
Last updated Feb 26.
BOJ exiting zero rates. Yen strengthening. Largest foreign holder of US Treasuries.
Last updated Feb 22.
Stoxx 600 eighth straight monthly advance — longest in a decade. NATO budgets rising.
Updated Feb 27.
$36T+ debt. Interest exceeding defense. Fiscal dominance intensifying.
Last updated Feb 22.
DXY ~97.74. 10Y below 4%. Yuan strong. Central bank gold buying. Dollar weakening on multiple fronts.
Updated Feb 27.
Mar-a-Lago breach, Capitol breach, Minneapolis shooting. Pattern accelerating.
Last updated Feb 23.
"Who has the power" > "who has the chips."
x402, Coinbase Wallets, Lightspark. Machine-speed settlement.
$1.2T globally. Pure labor arbitrage.
IBM + Microsoft converging.
Gulf first. India and Brazil next.
HBM4 supply-constrained. SK Hynix/Samsung bottleneck.
Inference energy economics transformative.
GENIUS Act Jan 2027. Banks can issue.
Built-on-AI > bolt-on-AI. Intuit and Autodesk proving the augmented case.
If parity, value shifts to application layer.
Privacy, latency, cost advantages.
Pay-per-task robot labor. First contracts 2026-2027.
Biology as manufacturing platform.
Voluntary + compliance converging.
Proof of humanity becomes real need.
GLP-1, CRISPR, anti-aging Phase 3.
Desalination improving. Core infra.
Starship 10x cost reduction.
On-chain risk transfer. $6T market.
AI displacing white-collar faster than blue-collar.
AI changing engineering work. Spotify evidence. *Added Feb 25.*
Infrastructure builds agents, agents displace SaaS, displacement drives more infra. *Added Feb 26.*
Gold, silver, copper vs falling DXY. *Added Feb 26.*