Connect, Aggregate
and Analyze

Anthropic’s IPO share structure is one of the most unusual we’ve seen in public markets. When the company goes public later this month, investors buying Class A shares will get economic exposure — but almost no real governance power. Here’s the breakdown of the five share classes: Class A – The public shares. One vote per share, full economic rights. But the entire class is permanently capped at 49.9% of total voting power. Class F – One single share held by a Founder LLC (th

Today's jobs report: the labor market is stable, but it's no longer hot. Ahead of this morning's release, the Labor Department's own data tells a consistent story. please take a look at the charts below: ⚖️ Openings per unemployed worker are back near 1.0. At the peak there were two openings for every job seeker. Employers no longer have to compete as hard for workers. 🔄 Turnover is low. Hiring and quitting rates are back at or below 2010s levels, while layoffs remain histor

The market's first read on the midterms: not all AI trades are equal. By 9:50 a.m. ET on Nov. 4, the S&P 500 was up 1.6%. Under the surface, the moves by industry were very uneven. You can see that in the chart below: 📈 Solar and clean energy: First Solar +11.9%, Sunrun +10.5%, Enphase +9.2% 📈 Hospitals and insurers: Centene +8.0%, HCA +7.5%, Universal Health +7.0% 📉 Defense: Lockheed -4.2%, Northrop -3.8%, RTX -3.2% 📉 AI, data centers and power: the weakest block on the

The midterms just repriced three things investors care about: data centers, Iran, and the policy path for the next two years. Democrats are projected to take both the House and the Senate. Three implications: 1. Data centers: regulation risk just got real. Democratic governors won in Texas, Georgia, Michigan, Pennsylvania and Arizona. Texas alone holds about 27% of planned U.S. data center capacity, and these five states together hold roughly 39%. Several of these candidates

What a Democratic sweep could mean, and why markets may care less than you think Polls open today. Forecasts lean toward a Democratic House, and the Senate is the real toss-up. The Decision Desk HQ projection I'm watching shows a Democratic sweep (D 241 in the House, D 52 in the Senate). That's a projection, not a result. If it happens: 🏛 Domestic: Expect subpoenas and oversight hearings, and fights over spending bills. Democrats would have leverage on appropriations. A divi

Take a look at the chart below. Morgan Stanley estimates the US data center capacity shortfall will grow from 15 GW in 2026 to 43 GW in 2028. Demand rises from 97 GW to 180 GW, while supply only reaches 137 GW. So what does it cost to close that gap? A 1 GW AI-optimized campus runs roughly $35-40B all-in (Epoch AI puts it near $38B). About 60% of that is servers and GPUs, not concrete. Land and grid hookups are under 1%. $40B x 43 GW = ~$1.7 trillion That is the order of magn

Midterm years: first the drawdown, then the rebound. 13 times out of 13. Chart 1 👇 Every midterm year since 1974 has delivered a drawdown. 📉 The range runs from -7% (2014) to -35% (1974). 📉 The average is about -18%. 📉 Six of the 13 fell 20% or more. Chart 2 👇 Now look at the 12 months after the election. 📈 All 13 were positive, with an average of about +13.6%. Counting back to 1950, it's 19 for 19. 📈 The range runs from +1% (1986) to +27% (1994). 📈 The worst drawdown

The Nasdaq is down nearly 6% over the month of October, and I think the market is starting to price something it ignored for a long time: the AI buildout is increasingly financed with (debt) commitments, not just cash flow. You can see that in the upper chart below. Hyperscalers, NVIDIA, Broadcom and SpaceX now disclose $4.11tn of off-balance-sheet commitments in their latest filings. That's up from $3.29tn in Morgan Stanley's August estimate. The biggest: • Alphabet: $1,059b

Credit is flashing amber. Equity is being asked to pay the bill. Two charts, one story. 📈 Lower-quality credit (CAA-B OAS) has climbed to ~780bp. That's close to the 2016 peak, and every prior peak was met with policy support: global QE in 2008 and 2020, the ECB's CSPP in 2016. 📈 Meanwhile, net equity issuance just turned sharply positive. Q2 2026 is roughly $150bn, the biggest print in the series since 1997 and about three times the 2000 and 2021 peaks. For years, buybacks

The Fed held. The real decision is still ahead. The FOMC kept the target range at 3.75%–4% by a 10–2 vote, with Cook and Hammack preferring a 25 bp hike. Chair Warsh also announced that one meeting a year will become a strategy session with no rate vote, starting in 2027. Market reaction: a hawkish hold that traded dovish. The 2-year yield dipped as low as 4.678% after Warsh declined to give a clear steer on December. Equities rallied modestly, though narrowly, while the 10-y

What's the optimal gold allocation in a stock/bond portfolio? The data from 1970-2024 says ~14%. Incrementum AG ran a gold allocation sweep on a stock/bond portfolio (S&P 500 TR and US 10Y TR), using data from Robert Shiller and LSEG. Gold was added in 2% steps from 0% to 30%, then in 10% steps up to 60%. You can see in the charts below what the numbers show: 📉 Volatility is U-shaped. It starts at 8.28% with no gold, bottoms at 7.81% with 14% gold, then climbs. By 30% gold y

Anthropic just opened its books. Meta just went to the equity market. Together they tell you where the AI trade is headed. Anthropic's S-1 landed Monday. The headline numbers: 📈 Revenue grew about 1,300% to $16.3B in the first half of 2026, including $11.56B in Q2 alone. 📉 The net loss was $80.7B, but most of it came from fair-value changes on convertible bonds, which is largely non-cash. ⚙️ Q2 revenue was $2.08 for every $1 spent on compute. Operating income swung from a $

Moody's Analytics just updated its "What Could Go Wrong" matrix, which maps risks by likelihood and economic severity over the next 12–18 months. A few things stood out: 🔴 Four risks are on Negative Watch: an AI stock selloff, Fed missteps, Iran war escalation, and bond market turmoil. Notice that two of them (Fed missteps, bond market turmoil) tie directly to this week's meeting. 🔵 Newly added: a major AI cyberattack and a debt limit debacle. ⚠️ Highest severity (far right

A decade of quiet erosion: what $1 in 2016 buys today The chart below shows the purchasing power across ~80 government currencies since 2016 (IMF data). It's worth a few minutes of your time, to use it as a framework for your portfolio return over thr last 10 years, measured in fiat currencies. 📉 The "strong" currencies didn't hold their value either: • Swiss franc: -8% • Euro: -22% • US dollar: -26% • British pound: -27% And the tail is brutal: Nigerian naira -79%, Turkish

The 10-year Treasury just fell below 5%. Here's what's behind it. On Thursday the 10-year yield dropped about 19 basis points to roughly 4.95%, its biggest one-day move since March 2023. Just weeks ago, long-term yields were at their highest levels in over 20 years. It wasn't one catalyst: 📉 AI jitters. Nasdaq fell 2.6% and semiconductors dropped nearly 7% on concerns about AI spending and access to capital. Money rotated into Treasuries. 🛢️ Oil and geopolitics. Crude climb

Meta now has $900B in liabilities and future commitments. Almost none of it is traditional bond debt. Meta's latest 10-Q, plus the $68B July lease subsequent event, adds up to roughly $900B in liabilities and future commitments, including contingent cloud purchases. Three quarters ago it was $384B, then $587B in Q1. You can see that in the two charts below. What's in it: • $349B in noncancelable purchase commitments • $279B in leases not yet commenced, plus another $68B signe

SPCX just delivered a strong Q3 2026 — and the real debate is already shifting from the beat to the forecast. Headline numbers: Revenue: $14.6B (+161% YoY) Net income: $4.4B (vs. a $1.1B loss a year ago) Adjusted EBITDA: $8.8B (+306%) Super Intelligence (AI) revenue: $7.6B — now the largest segment Connectivity (Starlink): $5.2B (+69%) Compute capacity: 1.7 GW at quarter-end The print was clearly better than expected. AI scaled faster than most modeled, Starlink remains a hig

In AI data center credit, the tenant matters more than the building. Take a look at the chart below, showing the G-spreads on 24 high yield data center bonds, grouped by who actually pays the rent. The pattern is hard to miss. At the bottom of the pack, the safer tenants: when Amazon or Meta are the tenants, the spread ranges from 166 to 345bp. For CoreWeave's own unsecured bonds: roughly 980 to 1,008bp Same asset class: powered shells, GPUs, concrete, cooling. The spread gap

When the long bond breaks the weakest borrowers, the long bond eventually comes down. Please take a look at the chart below. It plots the US 30-year yield (blue) against CCC & lower option-adjusted spreads on an inverted axis (gray), so a falling gray line means credit stress is rising. For most of the past 20 years, the two lines moved together. In 2008 and again in 2020, CCC spreads blew out and the 30Y collapsed right alongside them. Stress built at the bottom of the credi

Europe's migration story is more complicated than the headlines. This map of immigration growth since 2000 shows how uneven the continent is. Spain is up roughly 700%, Turkey 446%, and Iceland and Finland about 300%. Ukraine, Serbia and the Baltics have seen sharp declines. Meanwhile, France is in the headlines. Nationwide student protests have disrupted over 700 secondary schools and led to roughly 5,000 arrests. Politicians abroad have been quick to blame migration. But Reu