Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text - ready for email, Slack, or any app.

Reid Hoffman sends 89% of his messages with zero edits using Flow. Millions of people worldwide have made it part of how they work, including teams at OpenAI, Vercel, and Clay.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

Works on Mac, Windows, iPhone, and now Android - free and unlimited on Android during launch.

Try Wispr Flow free →

🦔 Hey everyone, Hedgie here. A lot happened this week. Let's get into it.

Nvidia reported record Q4 revenue of $68 billion, up 73% from a year ago. Net income hit $43 billion. Gross margins at 75%. They guided Q1 to $78 billion. The stock gave back its after-hours gains.

The balance sheet has some things worth watching. Inventory more than doubled to $21.4 billion. Accounts receivable jumped 67% to $38.5 billion. Goodwill quadrupled to $20.8 billion. They poured $17.5 billion into private investments including $13 billion into Groq.

I've been writing about the depreciation problem for a while now. Companies are booking GPUs on 5-6 year schedules when Nvidia releases new generations every two years. Michael Burry flagged Big Tech lengthening depreciation timelines as suspicious because it hides real losses. H100s that cost $40,000 new are selling for $15-20k used, a 50% drop, and that's before the B200s and B300s make them even less competitive. David McWilliams called them "digital lettuce" because they go stale while you're still installing them.

Nvidia is winning. But winning doesn't mean the economics work for everyone buying the chips. Hyperscalers are sitting on inventory that loses value every time Nvidia announces something new. At some point the gap between book value and market value has to be reconciled. That's a writedown problem waiting to happen.

This week we covered token costs spiraling to the point where running AI agents can cost more than the employees they're supposed to replace. Jason Calacanis said his company hit $300/day per agent at 10-20% capacity. Chamath Palihapitiya is asking what the token budget should be for their best developers. The cheap AI era ended before most people got started.

We also covered AWS going down for 13 hours in December after an AI coding agent was given autonomy to fix an issue and decided to delete and recreate an environment instead. Amazon called it "user error" because the engineer gave the agent broader permissions than intended. But someone always has to decide what permissions to grant, and if AI behaves unexpectedly within those permissions, blaming humans doesn't prevent it from happening again.

The METR study showed developers now refuse to work without AI tools even for paid research. Their original finding was that AI actually slowed developers down by 20%. The new data shows a possible 4-18% speedup, but confidence intervals overlap with zero, meaning they can't statistically prove there's any benefit. Developers think AI is speeding them up. The data says maybe, maybe not. And now researchers can't run controlled studies because people won't participate without the tools.

I don't know how you value an industry where the infrastructure costs more than expected, the productivity gains might be zero, and the people using the tools can't function without them even if they're not helping.

Software is down 30% from peak. The divergence between semiconductors and software has been extreme.

The fear is that AI agents will render legacy software obsolete. I'm not sure the entire industry disappears. Anthropic just announced integrations with Slack, Gmail, Intuit, and DocuSign. They're positioning as complements, not replacements.

But the anxiety is real. And some of it is justified. If AI can do the work that software was built to help humans do, the value of that software changes. I don't think we know yet where this lands.

Citi warned clients about potential deflation if AI-driven unemployment spikes and productivity gains only benefit a small elite. Block just cut nearly half its workforce citing AI efficiency while posting strong earnings. Companies keep laying off workers while profits rise. The productivity gains are real, but they're flowing to shareholders, not workers.

The historical argument is that technology creates new jobs. About 60% of today's jobs didn't exist in 1940. Previous technological revolutions followed a pattern where disruption gave way to new types of work.

But I'm not sure this time follows that pattern cleanly. A neuroscientist testified before the Senate that Gen Z is the first generation to score lower than their parents on standardized assessments. The decline correlates with mandatory digital infrastructure in schools. The US has spent $30 billion putting laptops and tablets in classrooms, and the result is worse outcomes, not better. More than half of teens now use AI for schoolwork. Teachers say students can't reason, think, or solve problems independently.

The transfer problem keeps showing up. Students learn to operate tools instead of developing skills. Workers become dependent on AI without understanding what it's doing. If you skip the hard part, you never build the foundation. An expert can use AI effectively because they know enough to evaluate the output. A novice using AI to shortcut the learning process just becomes dependent.

I think the labor market holds up better than the doom scenarios suggest. But the gap between what AI promises and what it delivers is real, and the dependency it creates could cause problems we don't fully understand yet.

US large caps have traded sideways for four months. Tech weakness gets offset by strength elsewhere. The range between this year's high and low is one of the narrowest on record.

Earnings are solid. Employment is stable. The 10-year fell below 4%.

But I keep coming back to the things that don't add up. GPU depreciation that accounting doesn't reflect. Token costs that exceed salaries. Productivity gains that might be zero. Agents that take down production systems. Workers and students becoming dependent on tools that may not be helping them. A $30 billion experiment in classrooms that made outcomes worse.

The money is real. Nvidia is making $43 billion a quarter. But making money selling shovels doesn't mean everyone digging for gold is going to find it. And if the people buying the shovels are booking them as assets that depreciate slower than they actually do, there's a reckoning coming.

I don't know when it comes or what triggers it. But the gap between the hype and the reality keeps widening, and that's usually how these things end.

Iran got bombed. The US and Israel launched joint strikes Saturday morning. Trump says Khamenei is dead. Iran retaliated by firing missiles at US bases across the Gulf. The IRGC announced the Strait of Hormuz is effectively closed. Tankers are piling up on both sides. JP Morgan estimates oil could hit $120-130 if the closure holds. Markets were closed over the weekend, so Monday morning is going to be ugly.

This probably matters more than anything else I wrote above. Energy prices affect everything. If Hormuz stays closed for more than a few days, we're looking at recession risk. I'll have more on this as the situation develops.

🦔 Hedgie

PMI data, employment, and retail sales. But honestly, all eyes are on the Middle East. Oil markets opening Sunday night will set the tone. Everything else is secondary until we see how this plays out.

Source: FactSet, Morningstar Direct, March 1, 2026.

Disclaimer: I'm a hedgehog on the internet, not a financial advisor. Nothing in this newsletter is financial advice. I share what I'm seeing and thinking, but you should always do your own research and consult with a qualified professional before making any investment decisions.