微软宣布三年后建成商用级量子计算机:靴子能否落地?

marsbitPublicado em 2026-06-15Última atualização em 2026-06-15

Resumo

微软宣布其新一代拓扑量子芯片Majorana 2的量子比特平均寿命达到20秒,可靠性比上一代提升1000倍,并计划在2029年建成具有商业价值的可扩展量子计算机。这一突破主要得益于其独特的拓扑量子比特技术路线,利用马约拉纳零能模的非局域特性来抵抗环境干扰,以及借助自家AI平台“Microsoft Discovery”的代理型AI大幅加速研发流程,优化材料和工艺参数。 然而,实现商用级量子计算机仍面临挑战:目前芯片仅集成12个量子比特,距离所需的百万量级仍有巨大工程差距;20秒的相干时间虽长,但对于运行复杂量子算法可能仍不足;此外,量子计算的编译成本高且结果验证困难。业内科学家对微软的进展持审慎乐观态度,期待更多经同行评审的数据。量子计算的最终实用化前景依然存在不确定性。

微软最近发布了新一代量子芯片Majorana 2。官方宣称,这颗芯片上的量子比特的平均存活时间达到惊人的20秒,可靠性比上一代提升了1000倍。微软据此放了一句狠话“2029年,我们将拥有一台具有商业价值的可扩展量子计算机。”就在去年,业界对此主流预期还是“十年以后”。现在,微软直接把时间缩短一半。

在这颗芯片的研发过程中,微软大量借助了自家AI平台“Microsoft Discovery”的代理型AI,让AI团队像人类科研小组一样分工协作,自主分析海量实验数据、提出假设、优化制造工艺。一个是量子计算的硬件突破,一个是人工智能的软件助攻,两个最前沿的赛道正在相互成就。

什么是量子芯片?

量子芯片操纵的是量子比特。一个量子比特可以处于0和1的叠加态,测量之前,它像一枚旋转的硬币,同时携带0和1的可能。两个量子比特的叠加态可以包含00、01、10、11四种可能,三个比特则对应八种......以此类推,n个量子比特的量子态是2n维的。再通过精心设计的操作,由量子门(量子计算中的逻辑门)操纵量子态的相位,使得概率幅在叠加中发生干涉,将正确的答案放大出来。这就是人们称量子计算机拥有“指数级算力”的原因。

除此之外,两个纠缠的量子比特还有一种奇特的关联,测量其中一个,另一个的状态会瞬间确定,无论它们相隔多远。利用这些量子特性,量子计算机有望完成经典计算机很难高效完成的任务。

量子芯片就是专门用来产生、操控和测量这些量子比特的处理器。它所用的不是传统的晶体管,而是用超导电路、囚禁离子、光子乃至拓扑材料来捕捉量子态,让它们按照人所设定的逻辑,即量子门,进行计算。

量子芯片的弱点

量子比特虽然强大,但它具有极其敏感、极度脆弱的致命弱点。

一个量子比特的叠加态,稍微受到一点外界干扰,比如温度波动、电磁辐射和宇宙射线,就会瞬间坍缩成一个确定的0或1,从而丧失并行计算的能力。这种现象叫退相干。

在微软的Majorana 1芯片之前,主流超导量子比特的寿命通常只有几十微秒。也就是说,刚刚把它准备好,还没算几步,它就死了。因此,衡量量子芯片好坏的一个关键指标,就是量子比特的寿命,也称相干时间。

微软这次宣称量子比特的寿命达到20秒,在业内引发了地震般的反响。因为对于量子操作来说,20秒已经是天文数字。要知道,执行一个量子门操作只需要一微秒(百万分之一秒)。20秒意味着可以进行两千万次操作,从理论上讲足够运行相当复杂的量子算法了。微软甚至打了一个形象的比方:“这一改进差不多相当于发明一种手机电池,手机原本只能用一天,现在充一次电可用近三年。”

20秒只是平均值,部分量子比特甚至能维持一分钟。而上一代Majorana 1的寿命只能达到毫秒级别,所以微软才说“可靠性提升了1000倍”。

那么,微软是怎么做到的?答案隐藏在它的技术路线里:拓扑量子计算。

微软的秘密武器:拓扑量子比特

大多数主流量子芯片,比如谷歌、IBM,使用的是超导量子比特。它的技术相对成熟,但为了规避环境干扰,需要极低的温度,接近绝对零度-273°C,而且寿命短,容易出错。

微软花了20年走了另一条更难,但从理论上讲更有优势的路:拓扑量子比特。

在纸上打一个孔或两个孔,把纸揉成一团,纸会变形,孔却始终在那儿,一个孔不会变成两个孔,两个孔不会变成一个孔,纸上有多少个孔就是一种拓扑不变量。再比如将两根绳子编织在一起,绳子彼此交换位置的顺序也是一种拓扑不变量。拓扑量子比特正是利用了拓扑不变性来保护量子比特信息,信息并非存储在特定的粒子上,而是存储在准粒子(粒子系统的某种集体激发)相互交换位置所成的编织纹理之中。这种存储方式是非局域的,也就是说,比如噪声和热量这样的小扰动很难破坏整体的拓扑结构。因此,拓扑量子比特天生就对环境噪声不敏感,稳定性远超其他类型的量子比特。

微软使用的这种准粒子有一个传奇的名字:马约拉纳(Majorana)子。1937年,意大利物理学家埃托雷·马约拉纳预言,有一种奇特的费米子,其反粒子就是它自己。目前这种粒子还没被发现。21世纪初,科学家们开始在凝聚态物理中寻找它的模拟:一种叫做马约拉纳零能模的准粒子。马约拉纳零能模在二维空间中交换位置时,总体的量子态会发生变化,交换的顺序影响最终结果,类似于头发的编法不同,最后的辫子不一样。

1997年,任职于俄罗斯朗道研究所的物理学家Alexei Kitaev首次提出将马约拉纳子用于拓扑量子计算的理论。2005年,微软成立StationQ,Kitaev当时是核心成员之一,微软从此投身于这条技术路线,前后花了将近20年。2025年,微软发布了第一代Majorana芯片,证明拓扑量子比特在原理上的可行性,他们革命性地使用了拓扑超导体,可以创造一种全新的物质状态,从而实现更稳定的量子计算。今天的Majorana 2则是把原理变成了实实在在的性能飞跃。

其中一个关键改进是材料变了:第一代Majorana芯片的拓扑超导体用铝为材料,第二代改用铅。铅本身常用作辐射屏蔽材料,用它做超导体,可以大大加厚量子比特的护盾,保护脆弱的量子态免受宇宙射线干扰。这个听来似乎不算颠覆的改动,辅以AI对数百个工艺参数的优化,最终带来了1000倍的可靠性提升。

然而,目前Majorana 2只集成了12个量子比特。要想实现具有商业价值的通用量子计算机,业内普遍认为至少需要数百万个量子比特。从12到100万,中间还有无数工程和物理难题需要攻克。微软敢说2029年,表明他们对自己的拓扑路线非常有信心,因为从理论上讲,拓扑量子比特的纠错开销远低于其他主流方案,一旦付诸实践,有望比其他方案更快落地。

AI立功:代理型AI如何加速量子芯片研发

微软这次之所以能在可靠性上实现1000倍飞跃,还有一个不可忽视的“助攻”:代理型AI。微软拥有Microsoft Discovery的平台。这个平台的核心能力是部署代理型AI,也就是多个AI智能体可以分别扮演不同的角色,比如数据分析员、实验设计员、文献研究员,在人类科学家的指导下自主完成科研工作流。

事情要从Majorana芯片最核心的材料说起。第一代Majorana使用铝作为超导体,而第二代改用铅。更换材料牵一发而动全身,团队花了数年时间才摸索清楚各种权衡。要找到那个精确的掺杂配方,需要成百上千次实验。而现在,AI首先通过模拟圈出高概率目标,理想情况下,只需要实验一次。

这还只是开始。量子芯片的制造涉及软件、架构、材料堆栈、工艺、测量等无数环节,一个参数的改动可能引发连锁反应。人类工程师很难同时盯住所有变量,但AI代理可以。更关键的是,微软量子团队积累了近二十年的海量实验数据,格式五花八门,分散在不同国家、不同专业背景的科学家手里。AI代理能够重新综合并找到我们人类看不见的关联,因为没有任何一个人拥有那么广阔的视野。

AI的另一个杀手锏是加速实验。创建拓扑量子态需要同时调整几百个电压参数,然后进行测量,而测量恰恰是量子计算中最耗时、最精细的环节。以前,一个科学家手动完成一轮测量可能要花几周。团队曾尝试用早期的机器学习方法进行自动化,但没能成功。直到他们利用Microsoft Discovery平台训练出一个专用AI代理,将整个周期缩短了若干个数量级。AI可以并行扫描整个参数空间,自动判断哪个地方才是一切可以正常运转的最低点,然后精准定位。

最后,AI还帮团队解决了“幽灵噪声”问题。有一次,实验数据总是不对劲,科学家们排查了很久都没有头绪。后来一个AI代理综合了物理模型、设备日志和工艺知识,从原始数据中揪出了一个未经校准的温度传感器,它一直在悄悄破坏测量结果。

可以说,没有AI的参与,Majorana 2的1000倍性能飞升可能要再花好几年才能实现。这也印证了一个正在形成的共识:量子计算和人工智能可以相互成就。AI加速量子计算硬件的研发,量子计算机未来又反哺AI,为机器学习提供指数级算力。

靴子能落地吗?

在这个竞技场中,微软不是唯一的选手。通往“量子彼岸”的道路不止拓扑量子芯片一条,还有超导量子芯片、囚禁离子芯片、光量子芯片和硅自旋量子比特。各国政府也在加码投入。中国在量子通信和量子计算领域都有大规模布局;美国向量子计算公司大量拨款;欧盟也启动了“量子旗舰”计划。

2029年,微软真的能做出商用级量子计算机吗?英国萨里大学的物理学教授Paul Stevenson评价道,在制造可靠量子比特方面,微软看来已经取得了突破,如果成果经得起检验,这个时间听来合理。但与此同时,也有不少科学家希望看到更多经过同行评审的详细数据,因为微软此次发表的相关论文尚未完成同行评审。

当然,在微软的高调宣言和狂欢背后,也有几个问题值得冷静思考。第一,20秒够用吗?20秒的量子比特寿命,较之几十微秒而言,确实是惊人的飞跃。但实用级别的量子算法需要数以亿计的量子门操作。即便按一微秒一次算,20秒也只能跑两千万步,距离破解RSA密码、精确模拟药物分子所需要的数字还有好几个数量级的差距。须知退相干是物理定律定下的限制,是工程技术永远无法彻底摆脱的。第二,编译成本问题。每次用量子计算机解决一个问题,都先要在经典计算机上做一套编译,把问题翻译成特定的量子电路,再根据量子芯片的参数求解方程以得到量子门对应的电磁脉冲序列。这套编译过程不具有普适性,一次一编译,而且编译过程本身消耗的经典计算机算力可能接近,甚至会超过直接用经典方法求解的成本。第三,万一量子计算机报出的答案错了呢?人无法用经典计算机加以验证,假如能验证,也就不需要量子计算机了。最后答案错了,也不知错在何处。

建成商用级量子计算机的美梦,犹如一只悬空的靴子,迟迟不能落地。有朝一日纵使落地,可能不过是一声闷响。纵观科学史,科技的进步有时更像“无心插柳柳成荫,有心栽花花不成”。被人寄予厚望的,不一定能实现,而出路和突破,或许偏偏就在意想不到之处。

参考文献

https://news.microsoft.com/source/features/innovation/majorana-2-microsoft-discovery-agentic-ai/

https://www.bluequbit.io/blog/quantum-chips

https://www.bbc.com/news/articles/cj4p7gyvp52o

https://zhuanlan.zhihu.com/p/2035004303467917427?share_code=14f9XN3e5wlBq&utm_psn=2035105136662553502&utm_source=wechat_session&utm_medium=social&s_r=0&wechatShare=1

本文来自微信公众号: 心智观察所 ,作者:心智观察所

Perguntas relacionadas

Q微软宣布将在哪一年建成具有商业价值的量子计算机?

A2029年。

Q微软最新发布的Majorana 2量子芯片,其量子比特的平均寿命达到了多少?

A20秒。

Q相比主流的超导量子比特路线,微软在量子计算领域主要押注的技术路线是什么?

A拓扑量子计算(或拓扑量子比特)。

Q在研发Majorana 2量子芯片的过程中,微软大量使用了哪种人工智能技术来加速进程?

A代理型AI(部署在Microsoft Discovery平台上)。

Q根据文章,距离实现具有商业价值的通用量子计算机,业内普遍认为至少需要多少个量子比特?

A至少需要数百万个量子比特。

Leituras Relacionadas

M&A Deals in the Crypto Market Are Unusually Active

Title: M&A Activity in Crypto Market Becomes Unusually Active A rare signal is emerging in the crypto primary market: mergers and acquisitions (M&A) are nearing half of all financing deals. According to RootData, this month, M&A cases in the crypto industry reached 10, while financing rounds numbered only 14, meaning M&A accounts for approximately 42% of primary market transactions—the highest level in history. This does not signal a sudden industry boom. Instead, the rapid rise in M&A share primarily reflects the continued downturn in the financing market. Since November 2024, monthly crypto M&A deals have remained between 10-20, while financing deals have plummeted from around 100 to about 50, possibly hitting a new low this month. For project teams, this means the traditional path of relying on narratives, token expectations, and ecosystem subsidies to maintain valuations is narrowing. For leading companies, it presents a rare window to acquire teams, licenses, technology, liquidity, and market access at lower prices, with less competition and stronger bargaining power. Key active buyers include Coinbase, Kraken, Ripple, MoonPay, Polymarket, Kaiko, Sol Strategies, GSR, Keyrock, Jupiter, Paxos, and Ondo Finance. Their M&A logic is consistent: acquiring key capabilities at lower costs during the industry downturn. This is driven by more attractive valuations, reduced time and trial-and-error costs, the acquisition of licenses and compliance resources, and the integration of industry upstream and downstream segments. Current M&A focuses are concentrated in four areas: trading infrastructure (e.g., Coinbase acquiring Deribit, Kraken acquiring NinjaTrader), payments and stablecoins (e.g., MoonPay, Ripple expanding payment networks), compliance licenses, and asset issuance/distribution (e.g., acquisitions related to RWA and token issuance platforms like Coinbase's purchases of Liquifi and Echo). The rise in M&A is altering the primary market's exit logic. It provides an alternative path to the token-dependent model, encouraging teams to build tangible products, revenue, and strategic value that can be integrated. This could inject confidence into the market, showing that asset buyers and exit possibilities still exist, albeit with a stricter focus on real utility. However, this trend also indicates the crypto industry is becoming more centralized. As asset issuance, trading, market-making, custody, payments, and data gradually consolidate in the hands of a few major players, the industry's initial emphasis on openness and anti-monopoly is being reshaped by commercial realities. Coupled with rising compliance barriers, this signals the end of the low-barrier era for crypto entrepreneurship.

链捕手Há 7m

M&A Deals in the Crypto Market Are Unusually Active

链捕手Há 7m

M&A Deals Are Exceptionally Active in the Crypto Market

Mergers and acquisitions (M&A) activity in the cryptocurrency primary market has reached a historic high, accounting for approximately 42% of total deals in the current month, nearly matching the number of financing rounds. This shift does not signal a new boom cycle but rather reflects a severe contraction in the venture capital funding environment. As financing dwindles, established industry giants—including major exchanges, payment firms, and infrastructure providers—are seizing the opportunity to acquire strategic assets at lower valuations. Key drivers behind the surge in M&A include depressed project valuations, the need to quickly acquire talent and technology to capture short market windows, the pursuit of crucial regulatory licenses, and the strategic expansion into adjacent business verticals such as derivatives, payments, stablecoins, and real-world asset (RWA) issuance. Major acquisitions, like Coinbase's purchase of Deribit and Kraken's acquisition of NinjaTrader, exemplify the push to expand into high-margin areas like derivatives and multi-asset trading. This trend is reshaping the industry's exit landscape, offering an alternative to token-based exits and incentivizing startups to build tangible products and revenue streams with inherent strategic value for acquisition. However, it also points toward increasing centralization, as critical functions—trading, custody, payments, compliance—become concentrated within a few large, well-capitalized platforms, potentially raising barriers to entry for new ventures.

marsbitHá 8m

M&A Deals Are Exceptionally Active in the Crypto Market

marsbitHá 8m

Solana Privacy Ecosystem Panorama: A Complete Privacy Stack from Compute to AI

**Title: The Solana Privacy Ecosystem: A Full-Stack View from Compute to AI** **Summary:** This article provides a comprehensive overview of the emerging privacy landscape on the Solana blockchain, characterizing it as still in early development. It identifies two primary verticals—Neobanks and Private DeFi—as key drivers, while noting gaps in tooling and user experience. The discussion centers on two main approaches to private computation: Arcium, which utilizes Multi-Party Computation (MPC) networks (Multi-Party eXecution Environments) to process encrypted data with final settlement on Solana; and Magic Block, which leverages Trusted Execution Environments (TEEs) via its Private Ephemeral Rollup (PER). Both enable confidential applications like dark pools and private DeFi with minimal code changes. Building on this infrastructure, projects are creating privacy-focused applications. Umbra, built on Arcium, offers Encrypted Token Accounts (ETAs) for private balances, transfers, and selective disclosure for compliance. Other wallets like Privacy Cash and Hush provide mixer-like functionality for SOL. For private trading, encifherio uses TEEs to encrypt swap details routed through Jupiter, while VanishTrade and Darklake focus on shielding transaction intent and liquidity routing, with Darklake introducing a "blind slippage pool" to prevent front-running. Further applications include private prediction markets (e.g., Melee Markets using Arcium's encrypted order books) and private AI. Loyal exemplifies the latter, using both Magic Block and Arcium to enable decentralized AI agents that store user data, conversations, and transactions confidentially on-chain. The article concludes by framing privacy not as a single technology but as an evolving "ultimate privacy stack," with experts like Helius's Mert envisioning a future combination of Fully Homomorphic Encryption (FHE) and Zero-Knowledge proofs (ZK). Helius Privacy itself is developing a ZK-based UTXO privacy layer for Solana.

Foresight NewsHá 14m

Solana Privacy Ecosystem Panorama: A Complete Privacy Stack from Compute to AI

Foresight NewsHá 14m

Trading

Spot
Futuros

Artigos em Destaque

Como comprar CHIP

Bem-vindo à HTX.com!Tornámos a compra de USD.AI (CHIP) simples e conveniente.Segue o nosso guia passo a passo para iniciar a tua jornada no mundo das criptos.Passo 1: cria a tua conta HTXUtiliza o teu e-mail ou número de telefone para te inscreveres numa conta gratuita na HTX.Desfruta de um processo de inscrição sem complicações e desbloqueia todas as funcionalidades.Obter a minha contaPasso 2: vai para Comprar Cripto e escolhe o teu método de pagamentoCartão de crédito/débito: usa o teu visa ou mastercard para comprar USD.AI (CHIP) instantaneamente.Saldo: usa os fundos da tua conta HTX para transacionar sem problemas.Terceiros: adicionamos métodos de pagamento populares, como Google Pay e Apple Pay, para aumentar a conveniência.P2P: transaciona diretamente com outros utilizadores na HTX.Mercado de balcão (OTC): oferecemos serviços personalizados e taxas de câmbio competitivas para os traders.Passo 3: armazena teu USD.AI (CHIP)Depois de comprar o teu USD.AI (CHIP), armazena-o na tua conta HTX.Alternativamente, podes enviá-lo para outro lugar através de transferência blockchain ou usá-lo para transacionar outras criptomoedas.Passo 4: transaciona USD.AI (CHIP)Transaciona facilmente USD.AI (CHIP) no mercado à vista da HTX.Acede simplesmente à tua conta, seleciona o teu par de trading, executa as tuas transações e monitoriza em tempo real.Oferecemos uma experiência de fácil utilização tanto para principiantes como para traders experientes.

348 Visualizações TotaisPublicado em {updateTime}Atualizado em 2026.06.02

Como comprar CHIP

Discussões

Bem-vindo à Comunidade HTX. Aqui, pode manter-se informado sobre os mais recentes desenvolvimentos da plataforma e obter acesso a análises profissionais de mercado. As opiniões dos utilizadores sobre o preço de CHIP (CHIP) são apresentadas abaixo.

活动图片