← All shows

State of AI with Nathan Benaich - Europe_s air defense gap

Published Jul 16, 2026 · Duration 6:53 · Language en · 5 highlights

Summary

本期节目采访了防务初创公司Alta Aries的联合创始人兼CEO Hadrian Cantor,聚焦欧洲的防空缺口问题。他在2022年俄乌全面开战一周后抵达乌克兰,目睹了北约早已遗忘的高强度战争——欧洲和北约首次在现代史上失去了制空权,而夺走制空权的正是像沙赫德136这样极其廉价的自杀式无人机(他坚持称之为“弹药”而非无人机)。Alta Aries用运行在小型GPU、超压缩视觉模型的拦截器,在断网、被干扰的战场边缘实时识别并击落这些低质量画面中仅有一个像素大小的目标,并保持人在决策回路中。他强调仿真只能做概念验证,真正训练系统的是来自真实拦截的真实数据,因为对手(技术多源自中国、经俄罗斯和伊朗)在不断学习、加装后视传感器并以蜂群方式协同。他的核心论点是“数量即质量”:北约过去只买少量昂贵精密的系统,假设自己永远掌握制空权,如今这一假设已崩塌,连以色列的铁穹在开战头十天就用掉了全球35%的爱国者导弹库存。他还批评预算方向错误(如法国斥资100亿欧元造航母,却可能被100艘廉价无人艇击沉),并指出人才因美欧利益分歧正流向欧洲防务。最后他引用罗马格言“若想和平,先备战争”,坦言自己希望战争结束,但认为对手已透明地表明意图,欧洲必须重新武装,因为自由并非免费。

Highlights

  1. I wanted the room to hear what applied AI looks like when the test set shoots back. The continent lost the sky.

    我想让在场的人听听,当测试集会向你开枪时,应用型人工智能是什么样子。这块大陆失去了天空。

    Striking framing of applied AI in war
  2. There is rarely internet at the edge and never a data center. So everything runs on super compressed models on small GPUs inside the interceptor.

    战场边缘几乎没有互联网,更不会有数据中心。所以一切都运行在拦截器内部小型GPU上的超压缩模型里。

    Concrete edge-AI engineering reality
  3. They do not call the things drones. We call them munitions because it doesn't come back. There's an 800-gram warhead in the nose. I don't recommend them to come back.

    他们不把这些东西叫无人机。我们称之为弹药,因为它不会返航。机头里装着800克的战斗部。我可不建议它们回来。

    Dark, memorable insider naming
  4. During peacetime, you have time, but you don't have money. And during war times, you don't have time, but you have money.

    和平时期,你有时间,却没有钱;而战争时期,你没有时间,却有了钱。

    Pithy line on defense budgeting paradox
  5. One of our best talent acquisitors, his name is Donald Trump. When JD Vance told the Munich Security Conference that American and European interests had diverged, well qualified engineers at U.S. companies in London started leaving to build European defense instead.

    我们最出色的招聘者之一,他的名字叫唐纳德·特朗普。当JD·万斯在慕尼黑安全会议上说美国和欧洲的利益已经分道扬镳时,伦敦那些美国公司里合格的工程师们开始离职,转而去打造欧洲的防务。

    Surprising, provocative take on talent flow
Full transcript

Europe's air defense gap with Hadrian Cantor, co-founder and CEO of Alta Arez at Rise 2026. A few days before he walked on stage at Rise, we led the $60 million series A in Hadrian Cantor's company, Alta Aries. Two days before the talk, with the round barely closed, he signed a partnership with Airbus. Alta Aries is a next generation defense prime built around a single, especially critical problem.

shooting cheap flying objects out of the sky before they reach a city. Shaw heads, cruise missiles, the mass-produced munitions that have made the last four years of war look nothing like the wars Europe prepared for. Defense had not come up once in a day of talks about models and medicine and code. I wanted the room to hear what applied AI looks like when the test set shoots back. The continent lost the sky.

Hadrian arrived in Ukraine in the first week of March 2022, a week after the full-scale invasion. What he saw was a kind of war NATO had forgotten how to fight. A high and long-intensity warfare, not the small expeditionary operations in Africa and the Middle East, where the West always owned the air. For the first time in modern European history, he argues, Ukraine, but also NATO, doesn't have the upper hand on the air superiority. The weapon that took it away is almost insultingly cheap.

As Hadrian traces its lineage, the Shahad 136 is a slow, low-flying design, first drawn up by German engineers, then copied in turn by the Israelis, the Iranians, the Russians, and now the Americans. Nothing changes through the years, he said. The economics did. Most of the time it misses. I asked him to walk through the data loop, the thing every speaker that day had described from behind a laptop. Get representative data, run the model, grade the output, go again.

Alta Aries runs it in a field at night. Most of the time it doesn't hit, he said. And that is the whole problem. There is rarely internet at the edge and never a data center. So everything runs on super compressed models on small GPUs inside the interceptor. A mission has three phases. Radar detection sorts friendly from hostile. Then an interceptor flies into a kill zone roughly two kilometers wide, where an RGB and an infrared camera hunt the sky for a single pixel.

that is genuinely hard to find because the feed is low quality and the airspace is being jammed. Then a human, still in the loop, commits to the intercept. One target is tractable. A swarm is not. The loop only tightens if you close it fast. Alta Aries keeps around 20 engineers in Ukraine as close to the front as Hadrian can put them and combat tests every week or every two weeks. The hardware turns over constantly.

The vision models update remotely as more data comes back. They do not call the things drones. We call them munitions because it doesn't come back," he said. There's an 800-gram warhead in the nose. I don't recommend them to come back. What no simulation can reproduce. Simulation, he allowed, is good for a proof of concept and little more, and the reason is human rather than technical. The engineer who runs a test slept well, had breakfast, had a good coffee.

Almost as good as the tea in England. The operator who fires the system in Ukraine is on their fifth coffee that doesn't wake you up anymore at 3 a.m. Freezing running on adrenaline and far less trained. No amount of lab time closes that gap and no simulator renders the adversary. The Shaheads his team intercepts have started carrying rear-facing sensors that trigger evasive maneuvers when an interceptor closes in. Others fly in a mesh so when one is shot down another inherits the firing solution.

Much of the underlying technology he said is Chinese, routed to Russia and Iran. The lesson Hadrian draws is that classical computer vision is not enough against an opponent that keeps learning. And the only thing that trains a system to beat it is real data from real intercepts, which no simulation can manufacture. Quantity is the quality. So why isn't modern air defense everywhere? Because NATO and Europe bought small numbers of exquisite systems on the assumption that they would always own the air, and that assumption is gone. The new reality is that the quantity is the quality," he said. A sophisticated interceptor fielded in low numbers gets outgunned. Israel's iron dome works because Israel is small to defend, and even so, by his account, Israel used 35% of the world's stock of Patriot missiles in only the first 10 days of the war. Interception is never a sure thing. Never 100%, he said.

somewhere between 25% and 75% on the best systems. The fix is as industrial as it is technical, and the budgets are still pointed the wrong way. France is spending more, he noted, but just committed 10 billion euros to an aircraft carrier that he argued takes a decade to build and that 100 cheap unmanned boats could sink. He put the politics of it in one line. During peacetime, you have time, but you don't have money. And during war times, you don't have time, but you have money.

The talent turned toward European defense. Talent has become his strange dividend of the moment. One of our best talent acquisitors, his name is Donald Trump, he said. When JD Vance told the Munich Security Conference that American and European interests had diverged, well qualified engineers at U.S. companies in London started leaving to build European defense instead. What he sells them is accountability and proximity. Write something in the morning, watch it deployed on the front by night.

Alta Aries does no offensive weapons so far, and in his telling, this work is the precondition for everyone else's. The clinical trials, the healthcare, the AI for good that had filled the rest of the day are possible because we live in a country at peace, bomb the data center, and none of it ships. Pressed by an audience member to separate the founder from the person, he couldn't and didn't try. He reached for the old Roman Maxim, Sivis Pachem, Parabellum. If you want peace, prepare for war.

Europe stopped preparing, and the use of force is being democratized faster than its institutions have noticed. You can annex territory now, he said, and no one does anything about it. He was honest about the discomfort of his own business. I would sleep way better if tomorrow the war in Ukraine stops. But he doesn't think it will, and his case for rearming rests on a grim simplicity. The people on the other side keep telling us what they intend, so we should listen to them, because at least they are being transparent.

Freedom doesn't come for free.

Delete this episode?

This removes the episode page and its saved audio from this library.