姬永锋

西班牙宣布与以色列断绝外交关系。马德里当局于3月27日召回了其驻以色列大使，并计划关闭其驻特拉维夫大使馆。

【北京启动智能网联新能源汽车商业保险开发应用 统一适配L2至L4全级别】财联社3月29日电，从北京金融监管局了解到，北京已在全国率先启动智能网联新能源汽车商业保险产品开发应用。新产品基本沿用现有的新能源商业车险体系，按照“总体稳定、部分优化”的原则，主要为消费者和汽车企业关心的特定智驾场景、软硬件损失等提供风险保障，可以统一适配L2至L4全级别智能网联新能源汽车。比如，现有车险产品对驾驶人的定义主要基于“人驾”的基本场景，对L3、L4级别的“人机共驾”或者“机驾”情形不能完全适用。再比如，L2级辅助驾驶车辆，消费者购买新车后，有的会自费升级辅助驾驶系统，而现有车险产品没有覆盖这部分损失，需要进一步优化。

李嘉诚在加拿大石油产业的布局，历经油价数次起落与股价几度过山车，却在2026年伊朗局势骤然升级后，再度成为大赢家。最近关注伊朗局势下各大石油企业的走势，又让人想起他四十年前埋下的一步精妙棋局。许多人以为李嘉诚旗下核心业务不过是地产、港口与零售，殊不知他早在四十年前便悄然在加拿大铺开一盘超级大棋——从赫斯基能源到Cenovus，再到2025年低谷加码MEG Energy，一步步构建起一个庞大的能源帝国。
1986年：经典低谷抄底
彼时两次石油危机余波未平，叠加沙特增产，国际油价一度跌破10美元/桶。加拿大赫斯基能源主营重油与油砂，负债累累、奄奄一息，市场纷纷抛售。李嘉诚通过和记黄埔及家族公司，仅以32亿港元拿下52%控股权（1991年进一步增持至95%）。彼时外界纷纷质疑他“接盘”，油砂开采成本高、运输难、环保压力大，看上去是块烫手山芋。
结果如何？
90年代末至2008年前，油价从20多美元一路攀升至140美元，赫斯基扭亏为盈，成功上市扩张，李家通过分红与盈利收回的现金早已远超本金。
金融危机与油价崩盘：逆势坚守
2008年金融危机，加之2014—2016年油价暴跌，油价几度腰斩，赫斯基股价重挫70%—80%，李家账面一度蒸发数百亿港元，外界再次唱衰“夕阳产业”。2020年疫情冲击更甚，布伦特原油一度跌至19.9美元/桶，赫斯基连续亏损。李嘉诚并未退缩，而是推动赫斯基与Cenovus进行对等合并，新公司日产75万桶油当量，其家族持股约29%，仍为单一最大股东，董事会掌控力稳固。
纵观几次起伏，核心逻辑始终围绕商品周期的轮动：油价受地缘、供需等消息面扰动频繁，短期波动剧烈，但只要中长期需求（工业+消费）与供给约束仍在，油企的经营杠杆便能放大利润。赫斯基与Cenovus并未依赖期货炒作，而是通过技术迭代（SAGD蒸汽辅助开采将油砂成本压至40美元/桶左右）与稳定的加拿大政治环境，远离中东热点，资产安全性极高。股价随油价起伏，但李嘉诚着眼长远，以并购锁定产能与现金流。
2025年：再下一城
2025年，油价再度回落至60美元附近，市场高呼“能源夕阳”，投资者争相抛售。李嘉诚家族通过Cenovus以443亿港元（约79亿加元）全资收购英国上市的MEG Energy——同样是阿尔伯塔核心油砂区块加下游炼化资产，协同效应显著，新增11万桶/日低成本产能。
收购完成仅数月，2026年初伊朗局势骤然升级，霍尔木兹海峡航运受阻、中东部分基础设施受损，布伦特油价两周内飙升至112美元以上，高盛等机构将油价中枢上调至95—115美元区间。
当下格局：稳坐赢家
Cenovus 2026年产量指引为94.5万—98.5万桶油当量/日，距离百万桶仅一步之遥。李家合计持股（CK Hutchison约17% + 李嘉诚个人约12%）控制力稳固，按保守的100美元/桶估算，单日税前现金流近1亿美元，年化经营净流入轻松突破300亿美元。更关键的是，所有核心产能均位于加拿大，远离冲突区，原油可稳定出口至亚洲（包括中国市场）。同期Cenovus股价从去年低点10美元附近飙升至25美元以上，市值逼近500亿美元，李家持仓价值超千亿港元。
这场历时四十年的布局，其精髓并非押注单次油价反弹，而是纯粹的周期趋势投资：每逢全球资本恐慌出逃、油价被打至地板价时逆势抄底；待油价回暖，现金流便呈爆发式增长。油价几度起落，股价几番过山车，每一次低谷他都在场。2026年伊朗霍尔木兹海峡危机爆发，油价直奔110美元以上，其家族旗下日产量逼近百万桶，瞬间成为全球能源博弈中最稳的赢家。
这堪称“资本慢炖”与周期洞察的典范：
1） 低谷重仓——1986年出手赫斯基，二十年后推动赫斯基与Cenovus合并，2025年收购MEG Energy，每一次均在市场最恐慌、估值最低时出手；
2） 技术+规模——将油砂从高成本变为低成本，通过并购将日产从单公司的数十万桶提升至近百万桶，充分放大经营杠杆；
3） 优质资产守根据地，地缘乱局迎大风——加拿大资产远离中东，霍尔木兹海峡一乱，别人断供，他稳产高价售。
归根结底，李嘉诚家族这四十年石油棋局，验证了周期趋势投资的本质：不在乎短期油价几起几落、股价几次过山车，而是提前在低谷收集筹码，待需求复苏与地缘因素共振时，静坐岸边，看他人仓皇。2026年的伊朗局势，不过是又一次验证。
周期轮动中，真正赚取大钱的，永远是那些在低谷敢于下注，并能严控风险（不加杠杆、把稳现金流）、耐心等待引爆时机的人。

Just now: The Iranian Revolutionary Guard warned everyone to leave U.S. institutions in the Middle East, as a retaliatory action is imminent.
Targets:
Qatar
Carnegie Mellon University
Georgetown University
Texas A&M University
Northwestern University
Weill Cornell Medical College
UAE
New York University Abu Dhabi
American University in Dubai
American University of Sharjah
Lebanon
American University of Beirut
Lebanese American University
Iraq
Baghdad Iraqi American University
Sulaymaniyah Iraqi American University
Dohuk Kurdish Region American University
Kuwait
Kuwaiti American University
Jordan
Madaba American University

Ten years in a humble room,
finally rising to a tall building.
Ten years ago,
Xiaopeng was just a small workshop in Guangzhou University Town that couldn't afford air conditioning; ten years later,
Xiaopeng has grown into a large group
producing cars, robots, flying cars, and Robotaxis.
From being unnoticed,
to attracting everyone's attention,
Xiaopeng has never stopped moving forward.

一增一减！车载HUD正式进入AR时代车载前装HUD赛道，迎来历史性时刻。
本周，高工智能汽车研究院发布数据显示，2026年1-2月，中国市场（不含进出口）乘用车前装标配W-HUD交付新车34.99万辆，同比小幅下滑1.21%；AR-HUD标配搭载交付21.34万辆，同比增长89.86%，继续保持高增速。
另一组数据显示，去年下半年以来，中国市场配置W-HUD上市新车达到108款，而配置AR-HUD为69款。两组数据反差的背后，凸显终端市场对于HUD选择的倾向性。
尤其是城区NOA的普及加速，对于AR-HUD的需求更为迫切。比如，小鹏主推的AR车道级导航、AR辅助驾驶、AR安全辅助三大场景正在成为刚需，也是弥补现有人机共驾交互的不足。
和传统的W-HUD相比，AR-HUD在信息显示丰富度、车与道路的交互上更加进一步提升驾驶安全性，并且有效减轻驾驶过程的负担，提升驾驶舒适性。同时，大尺寸的显示画幅，更是为后续交互体验升级提供可能性。
在车企端，随着2025年小鹏、零跑两个新势力品牌实现AR-HUD首次上车，头部新能源玩家已经全部实现HUD的上车覆盖。其中，蔚来也首次从W-HUD升级至AR- HUD，在行业内率先采用MINI LED技术的全天候AR HUD。
2025年中国市场（不含进出口）乘用车前装标配W/AR-HUD交付439.19万辆，搭载率升至19.11%，这个数字在2020年仅为3.45%。同时，AR-HUD持续维持高位增长态势。
其中，前装标配AR-HUD交付155万辆，同比增长75.28%，搭载率升至6.74%。相比而言，W-HUD的同比增速仅为7.47%。目前，泽景、华阳多媒体、水晶光电、华为等几家中国本土供应商是AR-HUD的头部玩家。
尤其是华为主推的Lcos方案更是一枝独秀，高工智能汽车研究院监测数据显示，2025年1-12月，基于Lcos技术路径的AR-HUD方案实现新车交付34.98万辆，同比增长117.67%。
同时，在技术路径上，多元化的方案也在为车企提供更多选择。比如，作为AR-HUD行业的头部企业，水晶光电已经实现了TFT、DLP、Lcos、光波导、斜投影等多技术路径的AR-HUD多元化产品矩阵。
新进入者部分，德赛西威公告显示，公司HUD业务取得重要突破，获得东风汽车、上汽通用、广汽本田、东风日产、广汽乘用车等客户订单，并且即将量产首个增强显示AR-HUD。
考虑到座舱交互进一步升级，尤其是20万元以下车型竞争加剧，高工智能汽车研究院预测，2026年AR-HUD搭载率将冲刺15%-20%区间。同时，搭载量也将超过W-HUD，成为市场主力。（高工智能汽车）

Middle East's Largest Aluminum Company: Factory Severely Damaged by Iranian Missile and Drone Attacks

Excerpt from Shi Zhengcheng, Science and Technology Innovation Board Daily

EGA (Emirates Global Aluminium) announced that its Tawira Tawira production facility in Abu Dhabi suffered severe damage.

Author | Shi Zhengcheng

As the conflict between the US, Israel, and Iran continues, the focus of the capital market has shifted from disrupted shipping to whether the region's industrial output can survive until the end of the war.

Late Saturday night Beijing time, EGA, the Middle East's largest aluminum producer, issued a statement saying that earlier that day, its Tawira production facility in the Khalifa Economic Zone of Abu Dhabi was attacked by Iranian missiles and drones, suffering severe damage.

As background, on Friday local time, following attacks on two large Iranian steel mills and their associated power facilities, as well as a "yellowcake" factory, Iran listed six steel mills in Israel and related industrial facilities in five other countries in the region as new targets for retaliatory strikes, warning that "industrial enterprises and heavy industry workers in the region with ties to the US and Israel should immediately leave their workplaces to avoid endangering their lives."

Emirates Global Aluminium stated that several employees were injured in the attack, but none are in life-threatening condition. The assessment of the damage to the facilities is still underway.

The Abu Dhabi Media Office disclosed earlier on Saturday that three fires caused by ballistic missile debris near the Khalifa Economic Zone injured six people.

Emirates Global Aluminium is one of the largest aluminum producers in the Middle East and a significant supplier to the global market. The Middle East accounts for approximately 9% of the global aluminum market (8% excluding Iran).

According to the company's website, one ton of every 25 tons of aluminum produced globally comes from Emirates Global Aluminium. This equates to 4% of global capacity.

The attacked facilities include a smelter with a production capacity of 1.6 million tons of cast aluminum by 2025 and a refinery that supplies alumina (the main raw material for this metal) to the smelter.

The impact of the attack on production capacity is currently unclear. The company only stated that it had substantial metal inventories in its sea and overseas warehouses at the time of the Israeli and US-led war against Iran last month, and is currently utilizing these overseas inventories to meet customer demand.Besides Emirates Aluminium, another major Middle Eastern aluminum producer, Alba, also announced earlier this month that it was reducing production due to cargo being unable to pass through the Strait of Hormuz. Norwegian company Hydro's Qatalum aluminum smelter in Qatar has also reduced output.

The United States has just discovered that it has been badly deceived by Israel!
Excerpt from Historical Attitude by Tao Ran
On March 27, 2026, AIXOS News reported that Vance had a heated argument with Netanyahu during a phone call.
Vance criticized Netanyahu, saying he was "too optimistic in predicting the U.S.-Israel-Iran conflict and exaggerated the possibility of regime change in Iran."
This means Israel is in trouble, and the White House is starting to realize whether it has been tricked by Netanyahu.
Regarding this war, Vance has long been recognized as an "anti-war faction" and strongly opposes going to war with Iran.
He was even reprimanded by Trump for this.
However, on March 25, Trump suddenly agreed to send Vance to Pakistan to lead talks with Iran. That’s how the phone call between Vance and Netanyahu came about; the U.S. and Israel needed to synchronize their messages before the negotiations.
On the surface, it was Vance leading the talks at Iran's request.
In reality, if Trump had disagreed, Vance could not have led the talks, and that phone call wouldn't have happened.
Iran chose Vance over Whitakoff and Kushner because they valued Vance’s anti-war stance. Trump knew this well; agreeing to Iran’s request indicated a problem.
We know that Israel's external propaganda is very powerful, leading the whole world to believe that Iran is weak and just a crumbling house that will collapse with a kick.
Trump believed it too, thinking he could easily rake in a lot of spoils, just like in Venezuela.
No one expected such big trouble; U.S. military bases in the Gulf region suffered heavy losses, and decades of infrastructure were wiped out, leaving imperial dignity in tatters.
There have been too many reports proving that Trump wants to extricate himself from this situation, which aligns with his business mindset; if he can't profit, he quickly minimizes losses, which is why Vance can lead the talks.
Besides, Israel has even worse news.
This war has had a profound impact within the United States.
More and more people are dissatisfied with Israel's blatant dragging of the U.S. into the water; even a fool can see that the war has nothing to do with U.S. national interests and is purely for Israel's sake.
Recently in the U.S., a phenomenon has emerged where many political figures publicly cut ties with Jewish capital.
For example, Ruben Gallego, a Democratic Senator from Arizona, has pledged not to take sponsorship from the "American Israel Public Affairs Committee."
For instance, Brandt Herrera, a Republican candidate for Texas's 23rd district, who was promoted by Trump and is part of the MAGA faction, has openly stated he will not take sponsorship from the "American Israel Public Affairs Committee."
Those familiar with American politics know that the "American Israel Public Affairs Committee" (AIPAC) holds significant weight in American politics.
Many American politicians take money from this organization. In the past, few dared to oppose AIPAC; now, opposition to AIPAC is becoming public, even if "anti-Semitism" is politically correct, Israel's actions this time are too much.
Vance's attitude is merely a reflection of the overall public sentiment in the U.S.
So, is the U.S. going to withdraw from the war?
That’s wishful thinking, as Netanyahu has explicitly stated that they have prepared for this war for over forty years.
Forty years of manpower, resources, finances, and energy invested—how can they easily let the "cooked duck" that is America fly away? Israel and Jewish capital will certainly take more drastic actions.
There is not just one Loli Island. Israel's years of careful planning may very well come into play at this critical moment.
More importantly, although this war does not align with U.S. interests, an easy withdrawal is even less in line with U.S. interests. The core of American hegemony is dollar hegemony, and the pillar of hegemony is oil, commonly known as "petrodollars."
The princes agree to settle oil in dollars; in exchange, the U.S. military must protect the princes.
If Iran effectively closes the Strait of Hormuz, the princes' oil cannot be sold, and U.S. protection is minimal; having collected protection fees without delivering, will the princes still willingly pay in the future?
All of this could severely impact petrodollars.
Without giving Iran a heavy blow, the "petrodollar" system cannot be maintained. Once things reach this point, even if Trump wants to back out, many domestic interest groups related to petrodollars will oppose it.
So we see that despite overall public opposition to war, every vote in Congress to limit the war after it started, Trump has managed to pass, and the anti-war faction does not hold the upper hand in politics.
Perhaps this is exactly what Israel hopes for.

“When the energy is strong, one can go to the sea; when the energy is weak, one must go to the mountains.” This reflects the ancient understanding of the rise and fall of “energy.”
“Energy” can be understood as life force or the state of fortune.
Why go to the mountains when energy is low?
The mountain represents yang, strength, and stillness. The mountain's towering and steady presence allows energy to rise and disperse outward.
When a person is low in energy and depressed (weak energy), what is needed is uplift and support. Climbing high to see far and absorbing the strong energy of the mountains helps stabilize the mind and regain upward strength.
If one were to go to the sea at this time, the sea represents yin, softness, and entrapment. The vastness and depth of the sea might amplify feelings of powerlessness and confusion within, hence the metaphor “it is easy to fall into the water when observing the sea with weak energy.”
Why go to the sea when energy is strong? When a person is full of energy and feels accomplished (strong energy), what is needed is restraint and settling. Facing the vastness of the sea instills a sense of awe, teaching one to be tolerant and humble, redirecting and collecting the overly extroverted “strong energy.” At this time, if one were to climb the mountains, it becomes easier to foster pride and impulsiveness at a high place; the saying goes, “the heights are hard to bear,” and one might carelessly “fall off the cliff” due to excessive confidence.

In 2018, before the United States cut off supplies to Europe, Japan, and South Korea, China's integrated circuit export value was only $84.67 billion. However, seven years after the cutoff, in 2025, the export value soared to $201.9 billion, breaking through the $200 billion mark for the first time.
It may sound a bit exaggerated, but many countries should be developing new energy.

华为盘古大模型负责人王云鹤离职，被曝Agent创业
摘自量子位
王云鹤在北大攻读博士期间，就进入了华为诺亚方舟实验室实习，2018年博士毕业后，正式入职。
8年间，王云鹤历任高级工程师、主任工程师、技术专家，2021年底升任算法应用部部长，后于2025年接棒姚骏，成为诺亚方舟实验室主任，负责盘古大模型的研发。
从华为实习生到90后实验室主任
王云鹤出生于1991年。本科毕业于西安电子科技大学数学与应用数学专业，后进入北大攻读博士学位。
博士期间，王云鹤的专业方向就是人工智能，师从许超教授和陶大程教授，
2017年，王云鹤进入诺亚方舟实验室实习，主攻方向是机器学习和计算机视觉。
博士毕业后，他正式进入诺亚方舟实验室，直到现在宣布离职。
学术成果方面，根据Google Scholar，王云鹤的被引数为33109，h-index为68。
他与许超教授团队合作的《Image Processing GNN: Breaking Rigidity in Super-Resolution》获得了CVPR 2024的最佳学生论文奖提名。
这篇论文提出了一种嵌入图神经网络的超分成像算法，在超分成像实验中达到了当时的SOTA。
论文一作是北大智能学院2022级博士研究生田雨川，王云鹤为该论文的通讯作者。
被引数最高的一篇论文，则是与韩凯、田奇等人合作的《Ghostnet: More features from cheap operations》。
这篇论文要解决的，是在内存和计算资源有限的情况下，在嵌入式设备上部署卷积神经网络的难题。
王云鹤还曾获华为“十大发明”奖，这个奖项旨在奖励那些未来有潜力开创新的产品系列、成为产品重要商业特性、为华为和行业带来巨大商业价值的发明或专利技术。这一获奖成果还被应用于中国天眼FAST，协助发现数百个快速射电暴样本。
而进入华为之后，王云鹤的升职路径也可以说十分顺遂。2021年任算法应用部部长前，基本上是每年一升。2025年，在原诺亚方舟实验室主任姚骏内部调岗之后，王云鹤这位“90后”接任诺亚方舟实验室主任一职，成为盘古大模型负责人。
在朋友圈的离职官宣中，王云鹤并未透露接下来的去向。
量子位获悉，此番出走，王云鹤将投身Agent创业，已在进行水下融资。

照片显示，美国空军E-3哨兵预警机#AE11EA 81-0005在昨天沙特苏丹王子空军基地遭受的无人机和导弹袭击中严重受损，同场袭击还损坏了几架KC-135加油机。
该机型现役机队现已缩减至15架，另有5架部署在中央司令部辖区，且目前尚无替代机型在生产中。

2026年3月27日，XPeng Inc.（港股代码：9868）宣布，自2026年4月1日起，公司中文名称由“小鹏汽车有限公司”正式变更为“小鹏集团”，英文名称XPeng Inc.维持不变。同期，该公司于香港联合交易所有限公司主板交易的A类普通股，其中文股份简称将由“小鹏汽车 – W”变更为“小鹏集团 – W”。英文股份简称“XPENG – W”及股份代号“9868”维持不变。本次更名不影响公司股东的任何权利，股东无需交回现有股票，亦不设任何股票兑换安排。
对此，何小鹏表示：从小鹏汽车到小鹏集团，十二年，一个轮回，全新起点。从智能电动汽车的征程开始，到飞行汽车、图灵AI芯片，VLA自动驾驶模型，人形机器人IRON，Robotaxi等，我们正在物理AI的全球征途上，一步步把梦想变成现实。
#小鹏汽车更名小鹏集团#

2026 Top 100 Potential New Energy Storage Companies in China
Excerpt from Baihua Season DBC Deben Consulting
Energy storage is transitioning from a 'cost item' to a 'strategic asset' in the AI era.
If there is one thing that is most important in the energy storage industry from 2025 to 2026, it is: energy storage is evolving from a passive supporting device into a core infrastructure of the AI era.
This sounds like a slogan, but it hides a profound paradigm shift.
For the past decade, the story of energy storage has been simple: it is an 'appendage' of renewable energy. Photovoltaics and wind power are unstable; we pair them with energy storage to store excess electricity and smooth out the output. At this time, energy storage is a cost item. The core logic of the industry is to reduce the cost of electricity and increase the cycle life.
However, this logic is being disrupted by a 'too romantic' force, which is, of course, AI.
The demand for electricity from AI has already exceeded the design boundaries of traditional power systems. The annual electricity consumption of a large intelligent computing center is comparable to that of a small city with hundreds of thousands of people. Moreover, the requirements of AI computational power on power quality are nearly harsh. Voltage fluctuations, frequency deviations, and even millisecond-level power interruptions can render weeks of model training worthless.
More critically, the power load of AI is not flat. It fluctuates wildly with the peaks and valleys of model training and inference requests. Traditional power grids have never faced such a dual impact of 'digitalization + electrification'.
In the future, the role of energy storage will change. It will no longer be 'batteries that store electricity', but 'systems that regulate power'.
A number of 'new faces' are rising. They may not have the largest battery cell production capacity, but they definitely understand 'power systems + AI loads' the best. For example, companies like Kehua Data and KSTAR, which are rooted in data center UPS, inherently understand what 'high reliability, zero interruption' means; companies like Huawei Digital Energy, leveraging intelligent string architecture, are enabling energy storage systems to possess the ability to 'actively respond to the grid' for the first time; companies like Guoneng Rixin and Hengshi Technology are not making batteries, but energy storage EMS and virtual power plant platforms—using algorithms to predict load, optimize charging and discharging, and participate in electricity market transactions.
Regarding the core capabilities of energy storage in the AI era, it should not only be about how much electricity is stored, but also how to make electricity 'obey'.
Another change that many people overlook is that long-duration energy storage is shifting from 'technical reserve' to 'essential need'.
AI computing centers, large manufacturing bases, and even entire cities' zero-carbon parks require not just 4 hours of peak-valley arbitrage, but a stable supply of green electricity for continuous hours or even across days. Lithium batteries can do this, but cost, safety, and resource constraints will become increasingly tight. Thus, we will see that all-vanadium flow batteries from Dalian Rongke, zinc-iron flow batteries from Weijing Storage, and compressed air storage from China Energy Storage are transitioning from 'marginal technologies' to 'mainstream alternatives'. What they are solving is the Achilles' heel of 'long-cycle stable power supply' in the AI era.
At the same time, the business model of energy storage is being reshaped by AI.
In the past, energy storage made money through 'peak-valley price differences'. Charge during the day, discharge at night, and earn the price difference. Now, with the deepening of electricity market reforms, the sources of income for energy storage have diversified. Capacity leasing, ancillary services, demand response, and even participation in the spot market. And all of this relies on one core capability: forecasting and scheduling.
AI not only creates demand but also provides solutions.
Companies like Meike Shengen and Lechuang Energy in the list represent the use of AI algorithms to optimize energy storage operations. They do not produce batteries but can help users increase the revenue of each kilowatt-hour by several percentage points. In the future of energy storage asset securitization, this 'operational capability' will be more scarce than 'manufacturing capability'.
Therefore, when looking ahead to the next few years, the most important factor might not be which company has the largest capacity, but who can transform energy storage from 'equipment' into 'service'. Who can turn battery stacks into intelligent nodes will occupy a core position in the energy system in the AI era.
Behind this is no longer just competition over materials, processes, and production capacity; the focus is shifting to software, algorithms, and system integration capabilities.
The answers to this transformation are likely hidden in those companies that are 'currently not yet seen'. They may not be located in industrial parks of first-tier cities, and they may not produce battery cells or inverters, but they are quietly building the next generation of energy infrastructure using code, algorithms, and a profound understanding of the electricity market.
AI is too romantic, so romantic that it needs a new power world to support it. And energy storage is becoming the 'foundation' of that world. This is what we see as the most important thing.
(Written by Baihua Season)

AI 士兵崛起？俄乌战场惊现持枪人形机器人
摘自：IT之家
近日，一则关于“俄乌战场惊现持枪人形机器人”的消息在国内网络平台引发热议，机器人士兵真的要来了吗？
经查询发现，这一消息最早来自美国《时代》杂志 3 月 10 日的报道，该文章以《Rise of the AI Soldiers（人工智能士兵的崛起）》为题，介绍了一款由旧金山初创公司 Foundation 生产的 Phantom MK-1 人形机器人。
Phantom MK-1 全身包裹在深黑色钢壳中，面部为有色玻璃面罩，其给人的直观恐惧感远超普通的人形机器人。在今年 2 月，它手持各种高火力武器：手枪、霰弹枪，以及 M-16 步枪的复制品进行了测试。
“我们认为，将这类机器人投入战争而非真人士兵，具有道德上的必要性，”Foundation 联合创始人 Mike LeBlanc 说。这位拥有 14 年海军陆战队服役经历、多次在伊拉克和阿富汗执行任务的退伍军人表示，他们的目标是让机器人能够“使用人类能使用的任何武器”。
如今，Phantom 正在进行测试，其宣称将成为世界上首款专门为国防应用开发的人形机器人。该公司已经与美国陆军、海军和空军签订总额达 2400 万美元（IT之家注：现汇率约合 1.66 亿元人民币）的研究合同，使其成为获准的军事供应商。
此外，该公司还计划与海军陆战队“methods of entry”课程合作开展测试，训练 Phantom 在门上放置爆炸物，以帮助部队更安全地突破目标区域。
报道还提到，今年 2 月，两台 Phantom 人形机器人被送往俄乌战场 —— 最初用于前线侦察支援。该公司表示，五角大楼规定自动化系统在未经人类批准的情况下不能攻击，这也是该公司的设计意图。
“这是一条危险的道路，”总部位于华盛顿的战略研究中心 Defense Priorities 的军事分析主管 Jennifer Kavanagh 说。“自动化事物并让人类脱离控制的做法极具诱惑力。任何冲突双方之间缺乏透明度会引发更多担忧。”
对于曾为美国海军陆战队执行了 300 多次战斗任务的 Foundation 联合创始人 LeBlanc 来说，他将 Phantom 机器人派往俄乌战场后所发现的情况“非常令人震惊”。他表示：“这是一场完全由机器人进行的战争，机器人是主要战斗力量，而人类则提供支援。这与我在阿富汗时的情况完全相反，当时人类是全部，而我们只有辅助工具。”
去年 10 月首次亮相的 Phantom MK-1 身高 180 厘米，体重 80 公斤，可携带 20 公斤的货物或设备，以每小时 6 公里的速度移动，相当于人类步兵的行进速度；配备电动驱动执行器，它能活动手臂和手部，并使用躯干上的多个摄像头探测可见光，观察周围物体或人员。它还搭载了人工智能（AI），可快速评估战场环境并自主做出决策，例如确定移动方向。
Foundation 公司正在加速军事人形机器人的开发，目标是在 2027 年底前生产高达 5 万台。该公司计划今年部署数十台，并将年产量扩展至数千台。这些机器人将租赁而非出售，每台每年的租赁价格估计约为 10 万美元（现汇率约合 69.1 万元人民币）。

AI顶会！NeurIPS 道歉！
摘自 深科技首席 深科技
重磅！NeurIPS致歉，4天就被中国学术圈“打服”！
AI顶会NeurIPS，正式道歉了！
3月27日，NeurIPS官方在X平台发文致歉。
原因是其2026年手册新增条款，封杀873家中国机构。
华为、商汤、海康威视等，均在列。
条款称，OFAC制裁名单上的机构，禁止投稿。
官方辩解：沟通失误，链接覆盖范围超标。
目前已删违规链接，恢复往年政策。
这场对抗，仅用4天就分出胜负！
引爆学术圈！NeurIPS 制裁条款遭全球抗议
3 月 23 日，NeurIPS2026 主赛道手册发布，制裁合规条款曝光，瞬间引爆学术圈。
3 月 24-25 日，海内外多位学者公开拒任领域主席、审稿人。西湖大学修宇亮婉拒 AC 邀约，UIUC 姜楠终止五年 AC 服务，更有研究者发公开信痛斥政策荒谬。
3 月 25 日，CCF 强势发声，强烈反对该政策，倡议中国学者暂停投稿审稿，警告或将移出推荐目录。
3 月 26 日，中国科协叫停所有相关资助申请，相关论文不再纳入项目评价。
3 月 27 日，迫于巨大压力，NeurIPS 官方致歉认错并紧急更新政策。
历史重演！2019年IEEE曾因类似操作被抵制，最终妥协。
中国学者，早已是NeurIPS核心力量。
NeurIPS 2025中，清华论文数全球第一。
学术中立是底线，绝非政治筹码！
跨国界开放，才是顶会的立身之本。

狂飙！小马智行联手Uber登陆欧洲，深圳实现盈亏平衡，目标3000台Robotaxi
摘自汽车产业前线观察
无人出租车赛道杀出重磅玩家！
小马智行一月连放大招，牵手Uber进军欧洲、深圳实现盈亏平衡、年内冲刺3000台车队。
低调领跑的它，正在改写全球自动驾驶格局。
欧洲首单！联手Uber落地克罗地亚，开创商业化先河
小马智行最新官宣，携手Uber与Verne，正式落地欧洲首个商业Robotaxi服务。
服务落户克罗地亚首都萨格勒布，道路测试已启动，收费运营箭在弦上。
合作分工清晰：Uber负责流量入口，Verne负责车辆运维，小马智行提供第七代自动驾驶系统。
在监管严苛的欧洲拔得头筹，小马智行的技术实力得到重磅认可。
四大一线全覆盖！中国无人出行领跑者实锤
很多人不知道，小马智行早已是国内无人出租的隐形冠军。
它是唯一在北京、上海、广州、深圳四大一线城市均落地全无人Robotaxi服务的企业。
近期更接入微信出行服务，十亿级流量加持，用户一键呼叫无人车。
同时与如祺出行组建超百台车队，国内商业化规模持续扩大。
核心突破：第七代Robotaxi成本大降70%，体验升级蓝牙解锁、语音交互、空调预调，行驶更平顺更舒适。
盈利曙光！深圳实现盈亏平衡，两座一线验证商业模式
行业最关心的盈利问题，小马智行率先给出答案。
继广州之后，其第七代车型在深圳实现单车盈亏平衡，成为全球少有的实现双城市盈利跑通的企业。
成本下降、算法优化、车队效率提升，共同支撑起健康的商业模型。
这意味着：自动驾驶不再只烧钱，已经具备规模化可持续盈利能力。
狂飙扩张！2026目标：20城、3000台Robotaxi
小马智行CEO彭军宣布，2026年进入全面加速增长期。
年内目标：将车队规模扩至3000台以上，布局全球超20座城市。
海外将进入新加坡、卡塔尔、阿联酋、克罗地亚，国内继续深耕四大一线。
从技术验证到商业扩张，小马智行彻底驶入快车道。
低调却硬核：它才是全球无人出租的隐形王者？
相比百度、Waymo、特斯拉的高调曝光，小马智行一直低调潜行。
但它却在商业化落地、城市覆盖、成本控制、盈亏平衡上全面领跑。
全无人运营、欧洲首单、双城市盈利、千亿级流量生态接入，多项指标行业第一。
当行业还在谈未来，小马智行已经把无人出行变成了现实生意。
行业拐点已至：自动驾驶从烧钱走向赚钱时代
小马智行的密集动作，标志着Robotaxi行业迎来关键转折。
技术成熟、成本下降、政策放开、流量接入、盈亏跑通，五大条件全部具备。
中国企业凭借场景、数据、效率优势，在全球赛道上率先突围。
这不仅是一家公司的胜利，更是中国自动驾驶产业的里程碑。

The Chairman of the Civil Committee of the Islamic Parliament of Iran stated, "We are seeking a bill that can legally uphold Iran's sovereignty, dominion, and regulatory rights over the Strait of Hormuz, while also generating revenue for the country through tolls." The draft of the bill has been formulated, but it has not yet reached the stage of a complete proposal and will be submitted to the Parliament Research Center next week, to be improved with the participation of the legal team, and will be followed up after being submitted to the Parliament once it convenes. According to this bill, Iran will impose tolls on ships passing through the Strait of Hormuz.

Has Xiaopeng's intelligent driving surpassed Huawei? A comprehensive understanding
Excerpt from Yuan Guoqing

Yu Chengdong said: We are the first.
He Xiaopeng said: We are 5 times ahead of the industry leader.
Who is bragging? Who is telling the truth?
To clarify this issue, I had dozens of conversations with AI. Today, I will explain it thoroughly from seven dimensions.
1. Route Differences
Huawei insists on a safety redundancy route of "LiDAR + multi-sensor fusion," while Xiaopeng bets on a physical AI route of "pure vision + end-to-end large model."
Behind this is actually two different sets of "brains" at work.
Huawei is equipped with a "world model," which first reconstructs everything seen by LiDAR and cameras into a 3D virtual world, and then makes decisions. The advantage is a very high baseline, ensuring no "hallucinations" of non-existent things.
Xiaopeng uses a "physical AI model," where the camera sees the scene, and the large model directly converts it into steering wheel and pedal operation commands. The advantage is extremely fast response, with a driving style more akin to humans, and the iteration speed is exponential.
In simple terms, Huawei adopts an engineer's mindset—first calculating the world clearly, then controlling it, pursuing certainty and safety boundaries; Xiaopeng adopts a biomimetic mindset—mimicking the direct pathway of human "eyes-brain-hands and feet," pursuing efficiency and human-like feel.
2. How to Perceive the World
LiDAR actively emits lasers and measures the time of echoes, obtaining physical 3D true values.
Cameras passively receive light and output 2D images, only providing width and height, with no knowledge of depth.
In the past, pure vision could not distinguish between murals and real obstacles; now the solution is "BEV + occupancy network":
BEV stitches together images from multiple cameras around the vehicle into a bird's-eye view (similar to a satellite map).
The occupancy network divides the space around the vehicle into countless tiny 3D voxels (small cubes) and uses AI algorithms to determine whether each space is occupied or has obstacles. Its core logic is "constructing 3D through reasoning rather than measurement."
This solution has extremely high requirements for models and computing power.
3. What is the perception boundary?
Mainstream 8 million pixel cameras can capture objects 200 meters away on sunny days, but in environments like heavy rain or nighttime, the effective distance is reduced.
4D millimeter-wave radar has a perception range of 200-350 meters, significantly improving the recognition ability of stationary objects compared to traditional millimeter-wave, but still has limitations for objects with extremely low reflectivity.
Common LiDAR has a perception range of 200-300 meters, with accuracy far exceeding other sensors, able to accurately identify obstacles 200 meters away, providing precious time for lane changes or braking.
4. Safety Redundancy: Will it cause confusion?
In 99% of everyday scenarios, pure vision is completely feasible.
But in 1% of extreme scenarios, such as backlighting/heavy rain/fog, the camera image blurs, making LiDAR safer;
Many people worry: Will multiple sensors conflict?
For car companies with average algorithm levels, problems may indeed arise. But Huawei's core logic is: vision dominates decision-making, and LiDAR only provides physical backup. In the vast majority of everyday scenarios, LiDAR remains "silent," only intervening when vision fails, and generally does not create confusion in daily scenes.
5. Who has faster iteration speed?
Pure vision has an unmatched advantage that LiDAR cannot compare: iteration efficiency.
Pure vision relies on software and models, and can be upgraded via OTA, with iteration cycles measured in "weeks"; LiDAR relies on hardware, requiring redesign, tape-out, and mold opening, with cycles measured in "years."
More critically, data: pure vision has a unified data format, with low collection, annotation, and training costs, forming an exponential flywheel of "more data → stronger models → higher sales → more data."
LiDAR point cloud formats are not unified, annotation costs are several times to ten times higher than images, making it difficult to form a universal data closed loop.
In simple terms: AI evolves exponentially, while hardware grows linearly. The longer the time, the more apparent the generational advantage of pure vision.
6. Who has greater market value?
I asked AI a question: If you were an investor with 100 billion funds, which route would you invest in?
AI's answer was: As an investor with 100 billion funds, the final decision is not "either/or," but "layered layout":
1. Invest 70% in the pure vision route: betting on scalability and long-term monopoly. Pure vision can achieve universal adoption of intelligent driving, occupy the largest market share, and create dual barriers of data and models, possessing platform-level value. More importantly, the pure vision model can also be applied to robots, flying cars, and other physical mobile devices.
2. Allocate 30% to the LiDAR fusion route: hedging risks and earning high-end profits. The LiDAR route follows the path of "high-end → safety reputation → brand premium," while also avoiding the risk of future policies mandating redundant hardware.
7. How should ordinary people choose?
For consumers, the choice logic is:
For urban commuting, focusing on cost-effectiveness and believing in the speed of algorithm evolution, choose Xiaopeng.
For frequent long-distance highway driving, facing complex weather, with extreme safety requirements, and not concerned about money, Huawei is the more reliable choice.
The competition between pure vision and LiDAR is fundamentally a dialogue between two technical philosophies: one side believes that AI's capabilities will eventually cover everything, while the other side believes that the physical world requires physical redundancy as a backup.
Looking back at this competition, we find that Huawei's stability and He Xiaopeng's aggression are different paths to the same peak—one paves the way with engineering, while the other climbs with algorithms.
The true victory or defeat is not about who replaces whom, but about who can bring L4 to ordinary people's lives faster, safer, and more inclusively.
Which route do you favor more? Let's discuss in the comments
#Xiaopeng Motors #Huawei #Intelligent Driving #Yu Chengdong #AI

You must learn to adjust the tense pace of research,
Well,
Before going to the next listed company,
I did a foot spa with Old Ye.