Intel’s New Deal + Waymo C.E.O. Tekedra Mawakana on Scaling Driverless Cars + Trumps in Tech

Summary

• The U.S. government is turning $8.9 billion in previously conditional CHIPS Act funding into an unconditional payment for a 10% Intel stake. Trump said the bargain took “less than 45 seconds” and claimed he had made $11 billion for the country; Casey Newton saw an “air of a shakedown” because Trump had days earlier demanded CEO Lip-Bu Tan resign as “highly conflicted.”
• The bull case is supply security: TSMC makes most of the highest-powered AI chips for major designers in Taiwan, while a new fab costs roughly $20 billion to $30 billion and requires scarce equipment, years, volume, and clustered know-how. Backing Intel Foundry could position it as the “chosen national chip maker,” reassuring customers against failure and offering a domestic hedge against China-Taiwan disruption.
• Kevin Roose’s bear case is that Intel lacks execution, not capital, after missing mobile and AI. A state-backed incumbent may become less innovative, while Trump could pressure Nvidia, AMD, and others toward Intel through tariffs or deals; Newton expects this 10% template to recur and ultimately sees strategically crucial AI labs as plausible nationalization targets.
• Waymo says its paid ride volume can rise from about 250,000 a week to “around a million rides per week” by the end of next year. Already operating in five metro markets, with Miami later this year, Washington, D.C. next year, and New York testing underway, it credits an early machine-learning stack whose edge-case feedback loop eliminated rain pauses and is moving toward snow and freeways.
• Waymo’s safety pitch rests on 71 million fully autonomous miles showing it is five times safer than human drivers on injury-or-airbag incidents, plus Swiss Re analysis of 25 million miles indicating a tenfold advantage using insurance data as a fault proxy. Against the “too safe” objection, Mawakana points to 40,000 annual U.S. road deaths and Phoenix demand for unaccompanied teen rides: “Is the status quo acceptable? No.”
• Waymo is defending premium pricing today while designing more varied and use-specific tiers tomorrow. The 52-minute San Francisco ETA Roose saw was a bug, not fleet scarcity, but fares remain algorithmic and dynamic; Zeekr scaling continues despite the host’s cited roughly 250% Chinese-EV tariff, diversified with other OEMs and U.S. integration at Magna in Mesa.
• Mawakana says true autonomy means “no ability for anyone other than the vehicle to drive,” separating Waymo from teleoperation and camera-only rivals. Remote staff can send a coded cue when, say, “the road went away,” but onboard software chooses the route; politically, she concedes some jobs will disappear, rejects “move fast, break things,” and says Waymo narrows overly broad law-enforcement demands.
• Trump-family tech ties now span a loosely defined K–12 AI contest, prediction-market access, crypto treasury strategy, and federal web design. Donald Trump Jr. joined Polymarket after 1789 Capital invested tens of millions while also advising rival Kalshi; Trump Media plans $105 million of CRO exposure through publicly traded MCGA, and Airbnb co-founder Joe Gebbia wants government sites “as satisfying to use as the Apple Store.”

Deep dive

1. Washington is underwriting a domestic chip chokepoint

• Trump’s account was blunt: he told Intel CEO Lip-Bu Tan, “Give the United States of America 10% of your company”; Tan agreed, he said, in “less than 45 seconds.” The resulting transaction gives Washington a 10% stake for $8.9 billion, while Trump claimed the position had already made the country $11 billion.

• Newton kept the pressure sequence in view: days earlier Trump had called Tan “highly conflicted” over Chinese-company investments and demanded his resignation. Turning that threat into an equity bargain gave the affair “the air of a shakedown,” unlike 2008 auto stakes intended to prevent imminent collapse; Intel was underperforming, Roose stressed, but not existentially threatened.

• The strategic case begins with manufacturing concentration. Nvidia, AMD, and specialized chips designed by Google and Amazon largely depend on TSMC in Taiwan; if China invaded Taiwan or pressured TSMC to withhold supply, the United States could not meet surging AI-chip demand or maintain flows into phones, laptops, and other devices.

• Building a substitute is not simply a financing exercise. A fab can cost $20 billion to $30 billion, requires specialized lithography equipment with years-long waits, and needs enormous chip volume to repay its fixed cost. Taiwan also possesses a dense cluster of experienced workers and accumulated manufacturing knowledge that money cannot instantly reproduce.

2. Intel’s constraint is execution, not cash

• Intel Foundry is the industrial-policy vehicle: it currently manufactures mostly for Intel but aspires to serve outside chip designers and compete with TSMC. The hoped-for government backing would position Intel as the “chosen national chip maker,” potentially reassuring customers that the foundry will not fail after they move critical designs onto it.

• Newton initially questioned whether the deal changed anything because Intel had already been awarded $8.9 billion under the CHIPS Act. Roose’s clarification was material: the old award depended on Intel meeting benchmarks; the new arrangement delivers the money automatically while making Washington an equity holder invested in keeping the company successful.

• Roose’s objection was that “Intel’s problem is not access to cash.” It has repeatedly failed to anticipate market transitions, missing both mobile and AI, and its broad ambitions—building a TSMC-scale foundry while competing with Nvidia and AMD in advanced AI chips—have not convinced him, investors, or the public that execution has improved.

• The government stake could nevertheless manufacture demand. Nvidia, AMD, and other designers may shift orders to Intel to curry favor, while the administration could connect foundry purchases to tariff relief or other concessions. Roose would have preferred funding multiple smaller U.S. manufacturers and creating competition instead of “picking a winner.”

3. The 10% precedent widens Trump’s deal pipeline

• The transaction scrambled conventional politics. AEI economist Michael Strain called it less a thoughtful move toward state capitalism than “an opportunistic display of corporate shakedowns”; Rand Paul described government ownership of production as a step toward socialism, while Bernie Sanders argued taxpayers deserve “a reasonable return” when subsidized chipmakers profit.

• Newton had previously predicted that Washington would eventually nationalize an AI lab because sufficiently powerful AI would carry military and intelligence value. What surprised him was starting with Intel, yet the governing logic now looks reusable: “You want something from me? I gotta get something back.” He does not expect Intel to be the last deal.

• His preferred alternative was institutionally slower but more coherent: convene experts, work with Congress, legislate a funding pool, and pursue a structured plan for reducing Taiwan dependence. The one-off Intel bargain, in his view, has only a “tangential connection” to a comprehensive semiconductor strategy.

4. Waymo’s learning loop is turning into ride volume

• Mawakana listed active service in San Francisco, Los Angeles, Phoenix, Atlanta, and Austin, with Miami planned later this year and Washington, D.C. next year; Waymo has also begun New York City testing. Its expansion involves challenges including rain, snow, freeways, dense pedestrian environments, and new city approval processes.

• Waymo currently supplies hundreds of thousands of weekly rides, with Roose citing roughly 250,000 paid rides. Mawakana said that, “in all likelihood,” the service will reach around one million rides per week by the end of next year—roughly a fourfold increase.

• She attributed that transition to safety discipline plus an early decision to embed machine learning in the original technology stack. Humans continually identified edge cases and fed them back into the system; Waymo once paused service during hard San Francisco rain, but now “we don’t even think about it anymore.” Snow and broader freeway operation are next.

5. Safety data cannot eliminate the behavioral test

• Waymo’s latest report covered 71 million fully autonomous miles and found the cars “five times safer than a human driver” for incidents normally involving injuries or airbag deployment. Mawakana emphasized that this comparison does not assign fault, an important distinction when assessing who caused a collision.

• Swiss Re supplied the fault-oriented proxy: analyzing insurance data across 25 million autonomous miles, it found Waymo “10 times better than humans.” Mawakana’s conclusion remained directional rather than tied to one multiplier: whether five or ten times, she believes the technology is making every community where it drives safer.

• Roose preserved Malcolm Gladwell’s counterargument that autonomous cars may be too safe: pedestrians will exploit their predictable stopping behavior, creating congestion and widespread jaywalking. Newton illustrated it—he feels nervous entering a crosswalk before a human-driven vehicle, but when he sees a Waymo, “I’m leaping into the crosswalk.”

• Mawakana returned to the baseline of 40,000 annual U.S. road deaths: some people may gamify cautious vehicles, but that cost must be weighed against preventable fatalities. Phoenix’s new unaccompanied-teen product is her concrete adoption case—working parents want a safer option, while the Waymo Driver itself follows the same seat-belt and driving rules for teenagers.

6. Product tiers and supply diversification define Waymo’s economics

• Waymo progressed from Chrysler Pacifica minivans to Jaguar vehicles, with Zeekr vehicles and IONIQs following. Mawakana expects multiple tiers: the “date night experience” will differ from the “soccer team experience,” while some European cities will require smaller form factors. The product portfolio, not one universal robotaxi, is her scaling model.

• Roose identified the Zeekr relationship as a tariff risk, citing a roughly 250% rate on Chinese EV imports. Mawakana said Waymo already has vehicles in the United States, continues testing them, and has not changed its scaling plan, while its broader OEM roster provides diversification as administration policy changes.

• She also drew a line around the domestic value layer: the Waymo Driver is installed at a Magna-run facility in Mesa, Arizona. Although the vehicles rely on a global supply chain, “where this technology was born, where it’s applied, where it’s understood” remains in the United States.

• On unit economics, Mawakana offered positioning rather than margins. Roose’s 52-minute San Francisco wait was caused by a bug—“period, full stop”—not insufficient supply, but pricing is dynamic across a fixed fleet. Today’s clean, consistent JLR I-PACE experience competes above UberX; lower-priced tiers are a future portfolio decision.

7. Labor and regulation remain city-by-city constraints

• New York’s Transport Workers Union warned that Waymo would turn pedestrians into “cannon fodder,” obstruct emergency responders, and ultimately replace rideshare, taxi, and transit workers. The charge combines a near-term safety challenge with the longer-term political risk that organized labor views autonomous fleets as direct job substitution.

• Mawakana pointed to the approaching five-year mark since removing drivers from Waymo vehicles in Arizona: driving jobs still exist, while autonomous fleets create integration, maintenance, fleet-technician, depot-leader, and depot-worker roles. Her concession was explicit—“there are jobs that will go away”—but she rejected the claim that autonomous vehicles will eliminate them all.

• On market entry, she chose neither permissionless Uber-style blitzscaling nor maximum regulation. The bar for launch should require extensive prior safety data, she argued, but not necessarily a complicated framework. Safety cannot become “a race to the bottom”; Waymo’s formula is “super deliberate and deeply engaged,” with “move fast, break things, absolutely not.”

8. Surveillance trust depends on disciplined disclosure

• Five Waymos were destroyed during June’s Los Angeles ICE protests, but Mawakana did not interpret that as Waymo-specific targeting because parking structures, privately owned cars, and other property were also damaged. The larger issue was whether a camera-covered autonomous fleet could become part of a government surveillance system.

• Asked whether footage could go “straight to ICE,” Mawakana did not offer a categorical no. She said Waymo publicly explains its practices, requires law enforcement to follow legal process, and narrows requests when their scope is “overly broad.” Her governing test was community trust—both in the driving system and in how captured information is used.

9. Waymo draws autonomy’s line at onboard control

• Tesla’s Austin robotaxi service creates a direct market comparison: Tesla relies on cameras, while Waymo insists on a “full sensor suite.” Mawakana welcomed multiple companies pursuing safer roads but argued Waymo’s sensor architecture has been essential to removing the driver across a large operating domain.

• Her competitive claim was that no other company has removed the human driver and reached Waymo’s scale across surface streets and freeways without either a human in the car to take over or an “extremely limited ODD.” Employees are already taking rides in San Francisco, Phoenix, and Los Angeles, with wider freeway availability still forthcoming.

• Mawakana corrected Roose’s assumption that remote operators occasionally drive the cars: “There’s no ability for anyone other than the vehicle to drive.” Waymo performs no teleoperation or remote driving; rider-support staff can speak with passengers, but they have no role in the driving function.

• Remote assistance is narrower. If flooding makes a road disappear, the vehicle may stop and ping that “the road went away”; a dispatcher can send a coded cue, but the onboard system must choose and execute a safe route. Waymo rejects the “gamification of driving” remotely because an operator lacks sufficient context.

10. Trump-family access is becoming a technology business model

• Melania Trump promoted a National Artificial Intelligence Challenge for K–12 students, asking them to complete AI-related projects, with participation trophies, cloud credits, and cash prizes of up to $10,000 per team member. Newton’s objection was specificity: children will discover AI because it does homework, but the announcement never made clear what they should build beyond “something AI America.”

• Donald Trump Jr. joined Polymarket’s advisory board as 1789 Capital, where he is a partner, invested tens of millions—despite his existing advisory role at rival Kalshi. Polymarket has bought a licensed U.S. derivatives exchange, and a federal investigation into Americans accessing its markets recently ended; Roose expects prediction markets to be legalized during the Trump administration.

• Roose worried that turning prediction markets into a coded right-wing activity could reduce their forecasting value by narrowing the ideological mix of participants. Newton focused on the contrast with years of scrutiny over Hunter Biden: he sees far less attention on a president’s son helping “kick the doors open” for Polymarket in America.

• Trump Media separately plans to buy about $105 million of CRO and form Trump Media Group CRO Strategy, Inc., trading as MCGA—“Make CRO Great Again.” Truth Social “gems” may eventually convert into CRO, while Airbnb co-founder Joe Gebbia, now U.S. chief design officer, aims to make federal websites “as satisfying to use as the Apple Store.” Roose noted that government sites support complex services and benefits and cannot simply be broken and fixed later.