Sam Winter-Levy
@SamWinterLevy
Fellow @CarnegieEndow, Technology + International Affairs. Previously poli sci PhD @ Princeton, @ForeignAffairs, @TheEconomist.
Drone threats make the UAE and Saudi Arabia the wrong homes for critical AI data centers, writes @CarnegieEndow's @CChivvis of @ceipStatecraft and @SamWinterLevy. wapo.st/4f8El44
Re-upping this new report I wrote on how to think about major power competition over "swing" states and what the US should do about it, with a focus on South Asia:
How does major power rivalry play out for South Asia’s “swing states” – and what does that mean for U.S. policy in the region? @pstanpolitics highlighted a few policy recommendations in his new @CarnegieSAsia paper Read the full paper here: carnegieendowment.org/research/2025/…
Yes, selling H20s to China will let them run reasoning models at scale and free up other chips for training But there's a bigger problem: Trump has linked AI chips to the trade talks. Now, any tough controls risk blowing up his deal How to lose your AI edge, with @SamWinterLevy
In @ForeignPolicy, @alasdairpr and I wrote about the Trump administration’s mismanagement of export control policy. In its desperation to reach a deal in its trade war with China, the administration has hamstrung its ability to impose new chip restrictions [link below]
Our job ad closes August 3rd! We're still looking for technical folks (ML, CS, EE, semiconductors) who're keen to apply their knowledge to ground policy in technical details. Also, strong generalists who help me run the show! Please apply, even if you're in doubt.
My team at RAND is hiring! Technical analysis for AI policy is desperately needed. Particularly keen on ML engineers and semiconductor experts eager to shape AI policy. Also seeking excellent generalists excited to join our fast-paced, impact-oriented team. Links below.
Good piece. It’s interesting how the public debate shifts between cyber and physical. Reliability, and survivability to a lesser extent, is a priority for any hyper scaler. Perhaps we think no sane person would destroy such a valuable economic resource. Our collective memory is…
Trump's Gulf data center plan will put a central node of the world’s AI infrastructure in locations highly vulnerable to drone and missile attack. Yet despite these obvious vulnerabilities, physical security has barely registered in the public debate. That needs to change.
Trump's deals in the Gulf put AI infrastructure in locations vulnerable to attack — at a time when the risk from cheap drones is intensifying, @CChivvis and @SamWinterLevy write in the @washingtonpost.
China's approach to AI governance follows a predictable cycle: when they feel technologically strong → more control. When they feel behind → prioritize growth. DeepSeek just broke this pattern. In our latest paper for @CarnegieEndow, @mattsheehan88 and I explain how we got…
New Post: Datacenter Delusions Among middle powers, the UAE has the clearest AI strategy: become an 'AI oil state' with meaningful global compute share. Most other countries are building datacenters without targeting viable sovereignty thresholds. 🧵
🚨 The Center for a New American Security (CNAS) is hiring for *five* new AI positions! 🚨 This is a terrific opportunity to work on some of the most fascinating, fast-moving, and consequential policy issues of our time with the best colleagues around.
There are big potential new AI laws in California, New York, and Michigan. @Scott_R_Singer and I laid out what’s in them and compared them to last year’s SB-1047. Transparency requirements, incident reporting, and whistleblower rules—yes; new liability, kill switches—no 1/
The definitive (read: very long) history of how America went to war in Iraq, a decision no one actually ever made.
Submit to the Irregular Warfare Initiative (@irregwarfare) Writing Contest, and share this opportunity widely! Short submissions due August 1st! @Jacob_A_Ware @SamWinterLevy @z_e_griffiths buff.ly/3pd3WMI
Over the last year, those of us who follow China's AI governance have been carefully watching whether China would establish an AI Safety Institute (AISI) to match those in the UK, US, and globally. That institution has now emerged, and it tells us a lot about the state of debate…
If China produces 200k Ascend 910Cs annually, filling the UAE's planned 5GW would require 15 years of production (3.1M chips) while delivering only ½ the performance. Or match performance with 7M chips consuming 2.2x more energy. China cannot backfill projects at this scale. 1/
BIS Undersecretary Kessler to Congress, echoing Secretary @howardlutnick, "Our assessment is that Huawei's Ascend chip production capacity for 2025 will be at or below 200,000, and we project that most or all of that will be delivered to companies within China."
Big open questions after the Gulf AI deals (link below) 1. Will they stick? A lot can change by 2030 2. Who wins and loses? OpenAI and Nvidia beat out Anthropic. What about AWS/GOOG/MSFT? 2. How do China, EU, India respond? Is the EU now fourth in AI? 3. More deals to come?