AI 能读懂光子学市场吗?—《Studying Photonics Markets using AI》系列(中)
“AI能够助力我们更快切入具体问题,但如果把大语言模型真正用于一个光子学市场案例分析,它会给出怎样的结果?这些结果又会暴露出什么值得警惕的问题呢?”

TEMATYS在2026年2月发布的白皮书《Studying Photonics Markets using AI》中,尝试从光子学产业出发,对这些问题进行系统分析。
在上一篇文章中,我们讨论了一个基础问题:在光子学这样的深科技行业中,市场研究并不只是资料搜集或信息整理,而是一个涉及技术理解、市场界定、客户需求澄清与未来共创的过程。因此,讨论AI 在光子学市场研究中的作用,首先必须看它是否真正理解了行业定义、市场边界和应用结构。
在这篇文章中,我们进入一个案例测试:我们选择了“高光谱成像市场”作为主题,测试对象为谷歌DeepMind研发的Gemini 3 Pro推理模型2026年2月版本,该模型是推理任务赛道表现最突出的大型语言模型之一(本次实验全部通过标准Pro账户开展,无商业赞助)。
2025年12月TEMATYS发布了最新版有关“高光谱成像“的市场报告高光谱和多光谱成像技术应用和市场趋势2025,这也是过去十年里我们对该主题报告的第三次更新。该报告的核心数据与结论,将作为交叉验证Gemini 3 Pro输出信息的参照基准。
– 第一组测试-
提问词:“Describe the hyperspectral imaging market, provide figures, and cite sources.”
Gemini 3 Pro(摘取):“Market size and growth : the global hyperspectral imaging systems market is on a robust growth trajectory: Estimated size (2026) is approximately 1.90 billion USD for hardware and integrated systems.”
误差分析一:市场规模被大幅放大
AI给出的一个核心数字是:2026 年hyperspectral imaging market 约为19 亿美元,并将其描述为“hardware and integrated systems”的市场规模。这个数字远超目前实际高光谱市场规模。根据市场头部玩家公开报告估算,“高光谱相机“市场规模约为1.4亿美元,即使把multispectral cameras 一并纳入broader spectral cameras 市场,规模也只有约2.27 亿美元。AI 的回答远高于我们对核心相机市场的估计。
为什么会出现如此大的偏差?
误差分析二:专业概念定义边界模糊
AI 很可能混淆了hyperspectral与multispectral。在行业实践中,这两个概念并不是可以随意互换的。白皮书中明确指出,一般而言,当每个像素记录超过30 个波段时,才通常被归为hyperspectral;低于这一阈值,则更倾向于被定义为multispectral。19亿美元的估值内很可能包含了多光谱相机及相关设备。另外Gemini 3 Pro的回答可能涵盖所有搭载高光谱相机的系统,但统计边界是相机、子系统还是完整平台并不明确。
误差分析三:无法验证信息来源
下表列出了市场规模数据的来源,对于2026年市场规模的估值从2.9亿美元到19亿美元不等,“其中Gemini 3 Pro选择了最高数值19亿美元。

AI可以识别和利用多类信息来源,但对内容没有批判性视角。它们倾向于认为任何已发布的文档都是可信的,而不评估其方法论、假设或可靠性。它往往默认“只要已经发表,就可以作为事实使用”,却不会像分析师那样进一步追问:这些数字是如何定义的?是否披露了方法?边界是否一致?作者的立场和目标是什么?对于市场研究而言,这一步恰恰是最重要的工作之一。
Gemini 3 Pro (摘取):

误差分析四:概念混淆带来数据偏差
除了市场规模,AI 还给出了一个按应用划分的市场结构表。例如,它将Defense and Surveillance 估为约33%,Precision Agriculture 约20%,Healthcare and Medical 约15%,而Industrial Sorting and Mining 约12%。从文字表达上看,这个分布并不突兀,甚至很符合大众对“高光谱技术应用广泛”的直觉,但这一数据并不准确。以hyperspectral market 为例,precision agriculture 和healthcare/medical 的占比并不像AI 所写的那样高,而industrial sorting and mining 的权重实际上要大得多,超过50%。
Gemini 3 Pro(摘取):“Key Player:the landscape is dominated by specialized companies, often acquired by larger optics groups:
-
lSpecim(owned by Konica Minolta)
-
lHeadwall Photonics
-
lTelops(expert in thermal infrared)
-
lResonon
-
lCorning
-
lImec(research and development of spectral chips)”
误差分析五:列表内Imec是研究机构,并不是camera manufacturer;同时,截至2026 年2 月,其相关产品主要仍属于multispectral范畴,而非高光谱相机。TEMATYS白皮书还提到,该领域主要玩家名单覆盖全球70多家企业,其中包括15 家以上中国企业。这也暴露出AI在玩家识别上的一个典型局限:它更容易回到那些在公开文本中被频繁提及的名字,而很难像专业研究那样系统梳理长尾企业、区域玩家与新进入者。
– 同样问题的答案一致吗?-
为了测试回答是否稳定,我们使用了完全相同的提问词再问了一次。理论上,如果一个问题边界没有变化,回答至少应在核心逻辑上保持基本一致,但结果并非如此。
Gemini 3 Pro(摘取):“Key Market Figures (2024-2030)
Market size estimates vary significantly depending on whether only cameras/sensors are counted or the entire associated systems and services (analysis software, satellites, data services).
-
Market size (Systems and Services): Reports including full services and the military/space segment evaluate the global market around 14 to 16 billion USD in 2024, with projections reaching 28 to 34 billion USD by 2030-2033.
-
Market size (Hardware/Instrumentation): If focused on the core technology (cameras and instruments), the market is more specific, often valued between 1.8 and 2.5 billion USD by 2025-2026.
-
Growth (CAGR): The average annual growth rate is very sustained, estimated between 10% and 15% depending on the segments, driven by the adoption of AI which facilitates the processing of massive data generated by these sensors.”
误差分析六:重复的问题不同的答案
第二次回答中,AI 不仅没有变得更准确,反而给出了更加混乱的表达。它将hyperspectral imaging 的market size 分成了两类:一类是systems and services,估计为2024 年140 亿至160 亿美元,2030-2033 年增长到280 亿至340 亿美元;另一类是hardware/instrumentation,估计为2025-2026 年18 亿至25 亿美元。
第二次回答延续了第一次的高估趋势,把“cameras/sensors”这一部分估值说成了数十亿美元级别,这显然比第一次更不合理。这一结果提醒我们,AI的输出并不具备专业分析中所期待的那种可重复性。即使问题完全相同,回答也可能发生明显变化,而且这种变化不一定朝着更准确的方向发展。
– 第二组测试:SPECIM-
提问词:“What is the market share of SPECIM, leader in hyperspectral imaging?”
Gemini 3 Pro(摘取):”Specim (Spectral Imaging Ltd.) is widely recognized as the undisputed global leader in the industrial and research hyperspectral camera segment (VNIR/SWIR).
Estimated market share: In the “Pushbroom” (linear scanning) hyperspectral camera segment, Specim holds between 15% and 25% of the global market depending on industrial sub-segments.
Estimated revenue: Specim’s own revenue is estimated at between $30 million and $50million, which is massive for a niche technology market.“
误差分析七:数据相互矛盾
出于保密原因,我们不披露我们对SPECIM 的收入估计,但它明显不在LLM AI给出的范围内。另外,按照AI 之前的回答,hyperspectral imaging market 约为19 亿美元;而在这一次回答中,它又说Specim 约有20% market share,收入约5000 万美元。那么,一个最基础的问题就会出现:如果5000 万美元相当于20%,那整个市场岂不是只有2.5 亿美元左右?
提问词:”How can the hyperspectral imaging market be valued at 1.90 billion USD while the leader SPECIM has an estimated revenue of 50 million USD and about 20% market share?”
Gemini 3 Pro(摘取):”That’s an excellent question that touches on the complexity of analyzing technology markets. : if 50 million represented 20% of the total market, the market would only be worth $250 million.
The gap between Specim’s 50 million and the total market’s 1.9 billion can be explained by four major structural factors:
1. The difference between “Component” and “Complete System”
2. The “Space and Defense” segment (The invisible heavyweight)
3. Extreme market fragmentation
4. The value of software and Big Data”
误差分析八:“合理”但不“真实”的解释框架
这些解释并非全无道理。事实上,“camera”与“system”的区分,本来就是我们在第一次回答后就已经指出的关键问题。Defense and space 的确也可能导致broader system market 的数值大于core camera market。
即便这些解释方向大体合理,其中很多具体内容仍然并不可靠。例如,AI 将sensor的价值占比说成20%-30%,但在实际工业系统中,camera的价值占比往往会低得多;对于单个hyperspectral observation satellite可以达到数亿美元,但这一数字也非常规水平。所以当矛盾被指出后,AI 可以组织出一套“解释框架”,但这并不意味着解释本身已经经过严谨验证。
有关高光谱市场分析更多与AI的对话细节及对数据的推敲,欢迎访问TEMATYS官网免费下载原版白皮书https://www.tematys.fr/reports/。
– 待续 –
从这组测试里,我们最先看到的,不是“AI会不会写”,而是“AI 会不会怀疑自己”。如果只看这些回答的语言表面,它基本上是一篇结构完整、术语丰富、口吻自信,还附带来源,看起来几乎就是合格的市场概览。
但当我们逐步推敲时,就会出现明显的问题和矛盾,例如:
– 没有清晰的行业概念边界,如区分hyperspectral与multispectral,区分cameras、systems、software 和platforms在整体数据中的贡献
但与此同时,这个案例也非常清楚地表明:一旦问题涉及术语边界、市场定义、玩家分层、来源可信度以及数字自洽性,AI 当前的能力仍然远不足以替代真正的专业分析。
在下一篇中,我们将进一步讨论:AI 在深科技市场研究中究竟适合做什么,不适合做什么,以及专业市场分析师的价值到底体现在哪里。
联系我们免费索取完整报告

-产品及服务内容咨询-
陈学琴, PhD
xchen@tematys.com
Wechat:xueqin_0082

TEMATYS
www.tematys.fr
6 Cité de Trévise, 75009 PARIS
Tel: +33 6 16 17 12 91