Trump's First 100 Days, Tariffs Impact Trade, AI Agents, Amazon Backs Down
Speakers
Jason Calacanis
— Angel Investor / Founder, LAUNCH
David Sacks
— General Partner, Craft Ventures
Ryan Petersen
— CEO, Flexport
Chamath Palihapitiya
— CEO, Social Capital
Aaron Levie
— CEO of Box
Summary
The All-In hosts grade Trump's first 100 days with guests Ryan Petersen of Flexport and Aaron Levie of Box, then spend the middle of the episode on the trade war: China tariffs near 145-154% have cut China-to-US ocean freight bookings by roughly 60%, importers are deferring duties through bonded warehouses, and small apparel brands face layoffs if nothing is resolved in weeks. David Sacks argues the market is already back above its Liberation Day level and that rare earths, magnets, batteries and pharma APIs were unacceptable national-security dependencies that had to be confronted, while Aaron Levie argues the same outcome could have been reached with incentives instead of shock. The final third covers AI agents: Levie sees software moving from seat pricing to labor budgets, Chamath warns that probabilistic software is not yet safe in regulated industries, and Sacks lays out compounding 3-4x annual gains in algorithms, chips and data centers.
- Panel scores the first 100 days - Chamath a B+, Jason a B - crediting the border, DOGE and deregulation, and faulting communication and economic uncertainty.
- Ryan Petersen: China-to-US ocean freight bookings are down about 60% after tariffs escalated to roughly 145-154%, with warehouse and trucking strain expected by early June.
- Importers defer duties using bonded warehouses in the US, Mexico and Canada while betting China-specific rates come back down.
- Small importers and apparel brands are the most exposed; re-sourcing to Vietnam is not feasible quickly because China's advantage is manufacturing ecosystem, not cheap labor.
- Chamath argues roughly 60% of Amazon sellers are Chinese-registered entities exploiting weak valuation and classification enforcement, and expects policy action.
- Rare earths and magnets, batteries, pharma APIs and AI are framed as strategic China dependencies with national-security consequences.
- Chamath highlights Bessent's proposed full expensing of factory construction costs as an uncovered catalyst that changes ROI and ROE math for most of the S&P 500.
- On AI: Levie sees agents expanding software TAM into labor spend, Chamath sees a trough of disillusionment in regulated industries, and Sacks sees algorithms, chips and data centers each improving 3-4x a year.
Ideas
Uber is the anti-tariff stock
In a tape dominated by trade-war risk, Jason frames Uber as the anti-tariff stock: its business has no imported-goods exposure, so tariffs do not hit it and it keeps doing great on green days while tariff-sensitive names get repriced. He owns it personally and says 88 is his number, and the stock is getting close to it.
Deregulation sets up a Trump boom
Sacks argues the administration is reprivatizing an economy that was being propped up by unsustainable federal spending that also flattered the employment numbers. The pieces are DOGE cuts to the federal workforce, a large volume of deregulation executive orders, unleashing energy by ending the EV mandate and offshore wind projects while encouraging oil and gas exploration, repealing the Biden AI executive order and ending the war on crypto. He says these take time to play out but set up a Trump boom, and he later notes the market is already back above its April 2 Liberation Day level, so the tariff drawdown was a media-fueled panic rather than evidence the policy failed.
China-US ocean freight bookings collapsed 60%
Petersen says bookings of ocean freight from China to the US have fallen about 60% once the China tariff stack went from a planned 54% to roughly 145-154%, a decline far beyond what anyone planned for, and the tariffs are live on departure date so goods arriving now are actually paying. If the freeze persists, the downstream network of warehouses and trucking starts seizing up around early June. He is explicit that this is not past the point of no return: negotiations are active, the administration does not want a tanked supply chain as its legacy, and importers are deferring duties in bonded warehouses while betting the China-specific rates come down.
Tariffs could wipe out apparel importers
In Petersen's bleak case, tariffs stay near 145% on China with no safe-haven rate elsewhere, trade falls off a cliff and small US importers get wiped out. The companies at risk are not just Amazon resellers but known fashion and apparel brands, and he cites a founder saying layoffs begin in his group chats if this is not resolved within two to four weeks. The reason they cannot escape is that China is no longer a cheap-labor destination: firms are there for manufacturing capability and ecosystem depth, anyone who could have moved already did under the 25% tariffs of Trump's first term, and a t-shirt supply chain cannot be physically restarted in Vietnam on this timeline.
Full factory expensing would lift S&P
Chamath flags Bessent's statement that the tax bill will let companies fully deduct property, plant and equipment plus the incidental costs of building a factory. He says that single provision would create an economic bonanza if enacted because it completely changes the ROI and ROE calculation for roughly 90% of the S&P 500, and he immediately took it to his wife's pharma business to rebuild their factory business case. His edge is that the measure is narrow, precise and actionable yet has had essentially zero mainstream coverage, so most S&P 500 CEOs and investors do not know it exists.
Amazon's Chinese-seller base faces enforcement crackdown
Chamath argues Amazon's marketplace has been built on foreign sellers that cannibalize American purveyors on price and margin. A foreign company can import into the US without forming any registered US entity, and he says roughly 60% of Amazon sellers are Chinese-registered companies with no US presence, so when they cheat there is effectively no enforcement: they can misstate valuation or reclassify goods to pay a lower tariff, or ship unsafe product. He calls this the biggest micro-level issue the administration is missing while it negotiates at 40,000 feet, and expects both the administration and a bill moving through Congress to act on it.
Rare earths, batteries, pharma APIs strategic
Chamath says the tariff fight has finally put the four things that actually matter on the national agenda: batteries, AI, pharmaceutical APIs and rare earths. His framing is national security rather than trade economics. If China invades Taiwan, or China and India fight and the US has to pick a side, Beijing can simply withhold pharma APIs, rare earths and batteries and set American life back 50 years, so the US currently has no strategic optionality to act on what is morally right. He adds that China's national-champion model, which blurs public and private partnership, law and capital, is the thing the US still has no answer to, and none of this was on the table until April 9.
China controls rare-earth processing and magnets
Sacks gives a specific structural reason the trade confrontation was necessary: over 25 years China strategically annihilated US rare-earth processing capability and the ability to cast rare-earth magnets, because the market went to the lowest bidder and that bidder was subsidized by the Chinese government in a way the WTO permitted. The result is a critical supply-chain dependency for basically every electric motor in every product, including cars, which he calls an unacceptable national-security exposure that free trade created and that the political conversation had to be shifted to address.
AI agents expand software into labor budgets
Levie argues AI agents change the software business model from selling seats to selling the work itself. Historically a vendor selling to a ten-lawyer department could sell at most ten seats at roughly 120 dollars a year each; if agents do paralegal-grade or professional-services work, the vendor can sell a multiple of those seats because software starts going after labor spend. He calls that a massive TAM increase that also resets how moats are built. He pushes back on the replacement framing: about 90% of future AI usage will be work companies never did before because it was unaffordable - reviewing every contract, automating an invoice process, running marketing campaigns in every language - with only about 10% displacing current work.
Enterprise AI stuck in disillusionment trough
Chamath says enterprise AI buyers are hitting the trough of disillusionment after two years of board-driven, CEO-mandated purchases and billions of dollars of spend. The unsolved problem is technical: replacing deterministic software with probabilistic models introduces error, and in regulated industries - life sciences, healthcare, financial services - an error means a fine or a shutdown, so QA, unit testing and integration testing go from throwaway work to the only thing that matters. He guarantees class-action lawsuits when those errors land and says budgets will not move from experimentation to mainline production until hallucination risk is provably contained; agents are real, but production-grade enterprise reliability is still far away from the toy apps and vibe coding people see online.
Exponential AI progress favors chipmakers
Sacks argues AI's impact is nowhere near peaking because three dimensions are each compounding at three to four times a year. Algorithms moved from chat LLMs to reasoning models that decompose a question, try approaches and validate them, with agents as the next leap. Chips went H100 to H200 to GB200 to GB300, with Nvidia back on a roughly annual generation cadence and NVL72-style rack networking lifting performance at the data-center level. Deployed compute is scaling from Colossus at 100,000 GPUs to 300,000 and toward a million, with Stargate on the same path, and power moving from 100 megawatt sites to the first gigawatt data centers. At 10x every two years each is about 100x in four years, and multiplied together that is roughly a million-x increase in available AI capability, partly taken as price reductions and partly as a higher performance ceiling.
Token-hungry agents justify data-center capex
Sacks gives the demand-side case for AI capex: each new generation of application is far more token-intensive than the last. Basic LLM answers need relatively few tokens, a reasoning model can spend a thousand times more tokens on a single answer, and agents will be more token-intensive still - a deep-research query effectively fires off about 200 sub-queries at once. Because serving these applications requires massively more compute, he concludes the data-center capex buildout actually makes sense rather than being overbuilding.
Up Next
This All-In Podcast video, published May 03, 2025, features Jason Calacanis, David Sacks, Ryan Petersen, Chamath Palihapitiya, Aaron Levie discussing UBER, SPY, SHIPPING, US apparel brands, AMZN, REMX, BATTERIES, Pharmaceutical APIs, IGV, Agentic enterprise software, NVDA, SMH, AI Data Centers. 12 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Jason Calacanis, David Sacks, Ryan Petersen, Chamath Palihapitiya, Aaron Levie · Tickers: UBER, SPY, SHIPPING, US apparel brands, AMZN, REMX, BATTERIES, Pharmaceutical APIs, IGV, Agentic enterprise software, NVDA, SMH, AI Data Centers