Debt Spiral or NEW Golden Age? Super Bowl Insider Trading, Booming Token Budgets, Ferrari's New EV
These notes were generated by AI and may contain inaccuracies.
Jason Calacanis introduces the All-In Podcast with David Sacks, David Friedberg, Chamath Palihapitiya. Topic one: AI acceleration. New study published Monday, February 9th, in Harvard Business Review suggesting AI tools intensify work but do not reduce it. Two UC Berkeley researchers spent eight months embedded at a 200-person tech company. Employees using AI worked faster, took broader scope of tasks, extended work hours. Workers felt more productive but more stress and burnout.
Sacks: Contrarian prediction that AI increases demand for knowledge workers. Study shows employees worked faster, broader tasks, more hours, more effort due to motivation. AI offloads menial tasks, making work more purposeful—shifting from task-based to purpose-based jobs. Key skill: structuring work for self and AI agents. Early adopters (AI natives) will demonstrate superpowers, e.g., completing days-long assignments in 2 hours. References viral article by Matt Schumer: "Something Big is Happening" on career opportunities for AI early adopters.
Massive enterprise adoption of AI agents driven bottom-up by early adopter employees using consumerized tools, not top-down initiatives (which take months with RFPs). Similar to consumerized SaaS spread.
Jason's tweet (2 million views): Laid-off Amazon/Microsoft workers should learn OpenAI o1 (corrected from OpenClaw), automate previous job, return to boss or join startups. Startups hiring for building/managing agents (beyond prompt engineering: managing, educating, offloading to agents). In Jason's org, 4/20 people focused on this have 10-20x leverage. Rolling out agents to all employees for 10x firm leverage. Examples: Podcast agent clips moments (e.g., group chat request for minute 3-6 clip), analyzes YouTube/Instagram/TikTok stats for viral strategies, eliminates reporting.
Recent advances: Claude 3 Opus, ChatGPT o1, OpenAI o1 revolution.
-
On-prem as new cloud? Post-2008 cloud migration for economies of scale (AWS/GCP). AI may reverse: companies fight for survival, avoid leaking proprietary data via cloud LLMs (prompts/responses/agent traces go to providers). On-prem offers control despite higher costs.
-
Court ruling: no attorney-client privilege in cloud environments—data becomes public domain. Enterprises need AI to survive but public endpoints sacrifice control/security/confidentiality. Solution: private provision networks (higher cost, but AI savings may balance).
Recursive output (agents improving agents) drives strides, surprising computer scientists expecting recursive model retraining. Jason obsessed with OpenAI o1—5-10% of venture firm work shifted weekly. Calls them replicants (personas) with Notion/Slack/Google Docs/email. Tech existed 6 months, but companies withheld API keys due to trust/liability. Now monitoring yields unbelievable results. Using Max Studios with Kimi for easy jobs (free), Claude 3 Opus for orchestration. Created OpenAI o1 Ultron meta-replicant managing four others, checking work, summarizing. Skills: deep research, PipeDrive sales database. Offloaded 30% of Athena assistant's work to replicants. 20% of investment team work by agents—perfect, no forgetting/mistakes, no checklists.
Jason upgrading to enterprise Slack (Chamath's #4 investment), ingesting all messages, giving API keys to all emails for Ultron. Investing $125K each in 10 OpenAI o1 startups for accelerator (email openai.co). Enterprise freakout imminent over data leaks. Sacks: Opportunity for enterprise-grade secure OpenAI o1 (partner's Lobster Tank). On-prem resurgence: e.g., Geico won't use proprietary risk data on open LLMs. Solutions: powerful desktops (Mac Studio 512GB), VAX-like terminals, but cloud too expensive (e.g., 80/90 top-20 Bedrock customer). CoreWeave: training-focused, spot pricing surges. Token costs: $300/day/agent via Claude API ($100K/year). Trend: tokens outpace salaries (superstar devs already; rank/file 10-20%, $100s-$thousands). Need 10x productivity or bust. Future: ramped energy density, lower token costs from Nvidia/Grok/Google/AMD, but confidentiality incentives persist.
Hit critical mass: >$1B on Kalshi, $700M on Polymarket, ~$2B total wagering. Media obsesses over manipulation/insider trading.
Examples:
- Day-old anonymous Polymarket account predicted 17/20 halftime bets (Lady Gaga, Ricky Martin appearances), profited $17K.
- Account <24h old bet correctly on Bad Bunny setlist.
- WSJ: Israeli soldiers (Army Reserve) arrested for using classified info on Polymarket (account Rico Suave 666 won $150K+, dormant 6 months, resumed betting Israel-Iran strikes).
Platforms CFTC-regulated. Kalshi CEO on CNBC: Cameraman seeing rehearsals like hotel staff seeing CEO rehearsal—no NDA typically.
Info either public or confidential (e.g., Super Bowl surprises). Platforms create betting without organizers' obligation to prevent.
Sacks: Is side bet with edge "insider trading"? Challenge balancing open trading vs. insider info (like securities). Chart: Few big accounts (sharps with edge/arbitrage) profit; many small lose quickly—risk of burning retail liquidity.
Friedberg: Sharps (edge) vs. squares (grist). Traditional sports: edge cases like throwing games/points, but leagues study betting lines analytically.
Shaving points and sharps are involved, but sports leagues study these analytically to prevent it. Prediction markets involve fundamental markets based on inside information. Regulatory bodies or society can do little about it because these markets are too fluid, dynamic, and ephemeral to regulate like securities. Too many prediction markets allow manipulation via asymmetric information with no arbitration.
This existed in securities markets pre-Reg FD (2000), which prohibited CFOs from sharing info with individual stock managers without disclosing to everyone. Previously, inside info sharing was not illegal: call your buddy, say quarter was blockbuster, buy stock. Warren Buffett's returns pre-Reg FD were double market returns; post-regulation, they matched market with zero alpha.
Markets thrive when there's asymmetry. Billions will be made in asymmetry.
Prediction markets, unless regulated out of existence, will resemble pre-Reg FD stock markets: sharps vs. squares.
Jim: Some prediction markets aid societal function by using inside info to uncover truth faster, incentivizing around whistleblower laws—benefits society. Others are manipulative setups. No way to discern type upfront; society must decide if faster, accurate current events info outweighs risks.
New event: Liquidity retreat/summit in wine country (Yountville), May 31–June 3, hosted by All-In Besties for capital allocators, LPs, GPs.
Chamath: Modeled on elite closed-door events like Iris ($10k/ticket, top traders present long/short ideas—Aman, David Einhorn, Cliff Asness have done it), off-record IB trader conferences for public stocks, and private company CEO pitches. Open it up: convene top public/hedge/private/growth/credit investors, LPs with trillions, fastest-growing tech CEOs. Presentations, best ideas, relationship-building, potential investments. Shut down Yountville/French Laundry for exclusive 2-day playground. Apply at allin.com/events; allocations for emerging managers. Goal: combine and publicize closed rooms.
Freeberg excited for Yountville. Ample networking/meetings. Chamath curating tightly.
CBO report (Feb 11): 2026 deficit $1.9T (6% GDP, vs. 3% target). Social Security trust fund depletes 2032 (1 year earlier). Debt: $31T today → $56T by 2036 (avg $2.5T/year 2026–2036). Debt/GDP: 120% now → 135% 2036 (House Budget Committee). Comparisons: Japan 237%, Singapore 176%, Venezuela 164%, Greece 154%, UK 94%; US was 60% 20 years ago.
The fiscal trajectory is not sustainable.
Freeberg: No path to 3% deficit/GDP. Assumes short rates ~3.1%; at 5%, adds $650B/year interest ($2T total), spiraling debt higher. Trigger worry: Democrats win midterms + 2028 White House → federal bailout of state/local pensions (e.g., CA ~$1T unfunded). CBO excludes this; could be catastrophic with SS insolvency. Links socialist city/state movements to USD weakness, deficits. Hard to cut spending; bailing states worse.
Sacks: Agrees on problem. CBO assumes low growth: 2.2% 2026 (vs. 4% Q3'24, 5% Q4 prelim), then 1.8%. AI capex could drive higher growth to escape spiral. Ideal: federal net outlays ~20% GDP, receipts ~17% (deficit ≤3%). Post-COVID outlays hit 30%, now ~23%. Solution: freeze spending until economy catches up (no cuts needed). Credit: federal employment lowest since 1966 (3M → 2.7M, -10%). Moving to private sector boosts productivity.
Chamath: Debt/GDP doesn't matter much historically—trend up/right since 1700 across top economies (wars spike then reduce it). Moves in unison; could hit 200–300% globally without change. Key: contain for inflation/earnings/wealth protection. Trump/DOGE identified hundreds of billions waste, but Congress inaction despite conservative majority. Hedge with durable assets (gold) as currencies erode. Governments addicted to spending.
Freeberg: Fed tightening (Kevin Warsh?) stops treasury buying → higher yields → exploding interest expense → forces spending restraint.
Jason: Consumer economy strong. Job openings: 12M → 7M (still high). Unemployment 4.4–4.5% (lifetime lows). Participation 62% (peaked 68% Clinton era). Mismatches (geo/skills/pay) sideline 7%. Border closure forces wage hikes. Prediction: Trump raises federal min wage ($7 → +$1–2/year over 3yrs)—populist, popular, boosts participation/affordability (vs. inflation counterarg).
Minimum wage increases labor costs, making it illegal to hire workers whose labor is worth less, leading to higher unemployment in low-skill segments—a core economic finding. Some benefit from higher wages, but others lose jobs, incentivizing automation. If minimum wage were a panacea, why not set it at $100/hour? Raising it beyond labor value causes unemployment.
Counter: Places like Australia, Seattle, San Francisco, New York, Los Angeles have higher minimum wages, higher prices (10-20% more in restaurants), but happier populations and low unemployment. Businesses can afford small increases ($1-2/hour). Theoretical arguments exist, but reality shows positive outcomes. Raising it could counter socialism concerns by providing a backstop against low-wage labor, making people stoked for higher-paying hourly jobs, similar to universal healthcare.
US at the beginning of an economic boom: Q3 GDP >4%, Q4 >5%. January jobs report: 172,000 new private sector jobs (beat 70,000 expectation), government shed 42,000 jobs, unemployment down to 4.3%. Construction added 33,000 jobs (data centers, AI boom). In Trump's second term: 615,000 new private sector jobs, >300,000 government jobs cut, increasing productivity and re-privatizing economy per Secretary Besson.
AI boom: Capex from four leading hyperscalers expected at $600 billion this year (~2% GDP tailwind), excluding ROI from software, applications, productivity. Avoid blackpilling on CBO's low growth projections (they'll print 6%) or distractions like Epstein. This period could be like late 90s golden age—phenomenal growth, internet boom, high labor participation—despite political scandals like Clinton-Lewinsky.
Unemployment ticked up from 4.1% (Trump inherited) to 4.5% (10% increase), but now down due to closed southern border and strong hiring. All job creation enjoyed by native-born Americans; job losses on non-native-born. October tick-up from DOGE voluntary buyouts (deadline Oct 1).
Top areas for illegal aliens: construction (#1), leisure/hospitality (#2)—2.5 million workers. Enforce by fining businesses hiring illegals: surveil construction sites, photo workers, demand paystubs/taxes from owners. Off-books jobs ($20-40/hour, no taxes) incentivize illegal crossings.
ICE technique (done for decades): surveil mornings (workers check in via trucks), video all day, confront owners: "Show paystubs/taxes." No records = tax evasion, multi-million fines. 2017 case: Justice Dept/ICE recovered $95M ($80M criminal forfeiture, $15M civil) from group hiring illegals—largest immigration case ever.
Target largest construction sites, then hotel/restaurant chains. Not raids, but surveillance (legal, peaceful). Business owners cause problem by providing incentives; without jobs, immigrants wouldn't come. 75% come for jobs (LA Times survey), not benefits/crime (small %). Gang bangers/rapists/murderers separate issue—ICE handles via warrants.
Ferrari's first all-electric vehicle: >1,000 horsepower, four electric motors, 0-60 under 2.5s, 330-mile range. Heaviest Ferrari ever at 5,100 lbs (vs. F40's 3,000 lbs). Launch May 2026; reveal May (illustration from Car and Driver).
Interior viral: Designed by ex-Apple chief Jony Ive's team + Mark Newome (Ford 021C concept). Key like iPhone—insert, yellow Ferrari color drains into shifter. Mac-inspired screen + tactile buttons (sports car enthusiasts love muscle memory, fighter pilot feel). Start: twist/press launch button, car turns Ferrari orange/red like jet startup.
Opinions: Interior balances Tesla glass cockpit + analog Ferrari (tactile buttons with pleasing sounds, heavy-duty). Exterior (projected): Hated—looks like Corvette, Frantzam, Model 3; bad grill, sides, back (hatchback-ish). Ferrari should be swoopier, curvier, fewer pieces.
Autonomy (FSD, Waymo) will make manual driving rare, like thoroughbred racing—risky, niche. Car culture (US freedom, interstate highways, productivity) changing. Ferraris beautiful art, market in China/India, but US insurance for self-driving too expensive. High-end like Ferrari for experience/luxury; rest use FSD/Waymo.
New CEO Benedetti (talented) will deliver beauty, but time racing autonomy.
Lexus LM (not in US): Minivan with first-class airline seats, blacked-out front for privacy, captain's chairs, full screen + divider, rising/falling monitor (driver talk/CNBC). Easy in/out height. Alfaard (Toyota version), LS—#1 in China, Singapore, Middle East for chauffeured use.
Personal: Model Y with FSD or Waymo; need 7-seater (X depreciating, 3-row Y tight). Dream: Abu Dhabi-seen Lexus minivan.
All-In Podcast episode 261. Let winners ride.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
