博通(Broadcom,代码 AVGO)在 2025-09-04 召开的 2025 财年第三季度财报电话会全文长达 41,767 个字符、7,375 个英文单词、共 60 个发言自然段。在这场电话会里,CEO Hock Tan 不仅披露了全公司合并未交付积压订单(Backlog)创纪录达到 1,100 亿美元、Q3 AI 半导体营收同比增长 63% 至 52 亿美元,还首次确认:在原有的 3 家超大规模云客户之外,第 4 家定制 AI 加速卡(XPU)客户已正式下达超 100 亿美元的生产订单,并将于 2026 财年大规模出货;同时宣布他本人将继续留任博通 CEO 至少到 2030 年。
如果把整篇 4 万多字符的英文会议纪要一股脑塞进大模型(如 DeepSeek、Claude 或 GPT)上下文窗口,不仅每场电话会平白消耗近万个无效 Token,还会因为 41 段华尔街分析师连珠炮式的假设性提问(如追问客户身份或毛利率压力)导致模型把“分析师猜测”当成“高管官方承诺”。更稳健的工程做法是调用 FMP(Financial Modeling Prep)数据源 的 /stable/earning-call-transcript 接口拉取结构化纪要,先在 Python 层按高管角色做精准切片,再交由大模型生成带原文证据的指引表。
一、 构建财报电话会 AI Agent 的两个工程防坑点
在用 Python 编写自动化投研 Agent 拉取并解析 /stable/earning-call-transcript 时,需要先处理好两个数据源细节:
- 先调日期目录接口对齐财年(
fiscalYear),切勿直接拿自然年硬拼参数:- 美股科技公司的财年截止月各不相同(如博通 AVGO 财年于每年 10 月底至 11 月初结束,微软 MSFT 于 6 月底结束,英伟达 NVDA 于 1 月底结束)。
- 工程上应先请求
/stable/earning-call-transcript-dates?symbol=AVGO获取已发布的(fiscalYear, quarter)列表,再映射为正文接口/stable/earning-call-transcript的year与quarter参数,防止因财年错位拉错历史季度。
- 按
发言人: 正文结构先锁高管白名单,剔除 60% 以上的提问与串场噪音:- 接口返回的
content字段以换行符\n分隔自然段,每段格式固定为Speaker Name: Speech Content。 - 实测博通(AVGO)与超威半导体(AMD)2025 Q3 电话会的段落分布,高管发言段数占比均不到四成,其余全是接线员
Operator、IR 主管串场与买方/卖方分析师提问:
- 接口返回的
| 标的代码与所属财季 | 会议日期 | 纪要总字数与词数 | 段落角色占比构成 | 发言管理层名单 |
|---|---|---|---|---|
symbol / year + period |
date |
content_size |
paragraphs_breakdown |
management_speakers |
| AVGO(2025 Q3) | 2025-09-04 | 41,767 字符(7,375 词) | 共 60 段(高管发言 19 段 / 提问与主持 41 段) | Hock Tan (CEO)、Kirsten Spears (CFO) |
| AMD(2025 Q3) | 2025-11-04 | 48,016 字符(8,043 词) | 共 59 段(高管发言 24 段 / 提问与主持 35 段) | Lisa Su (CEO)、Jean Hu (CFO) |
二、 Python 角色过滤与上下文压缩代码:砍掉 70.4% 无效 Token
下面这段代码拉取博通(AVGO)2025 Q3 电话会原文,先按 CEO(Hock Tan)与 CFO(Kirsten Spears)姓名过滤出 19 段高管发言,再通过订单与指引特征词筛出 6 段核心定量语料(仅 2,180 词,节省 70.4% 的上下文开销):
import requests
# 演示环境默认参数(支持直接运行验证 AAPL / NVDA / TSLA / MSFT / AMZN / META / GOOGL / AMD / AVGO 等示范标的)
API_KEY = "7dDuiUEbVsYU2mYf"
# 若请求繁忙可切换备用参数: 1T4BgJwz5yshQMkb
URL = "https://quote-fmp.com/stable/earning-call-transcript"
# 1. 拉取博通 (AVGO) 2025 Q3 财报电话会完整逐字稿
rec = requests.get(URL, params={
"symbol": "AVGO", "year": 2025, "quarter": 3, "apikey": API_KEY}, timeout=20).json()[0]
paragraphs = [p.strip() for p in rec["content"].split("\n") if ":" in p]
# 2. 按高管角色分离管理层发言与分析师提问,并提取含 AI 订单与业绩指引的完整候选段
MANAGEMENT = {
"Hock Tan", "Kirsten Spears"}
KEYWORDS = ("110 billion", "10 billion", "guidance", "forecast", "2030", "fourth quarter customer")
mgmt_all, mgmt_candidates, qa_or_mod = [], [], []
for idx, p in enumerate(paragraphs, 1):
speaker, speech = [x.strip() for x in p.split(":", 1)]
if speaker in MANAGEMENT:
mgmt_all.append((idx, speaker, speech))
if any(k in speech.lower() for k in KEYWORDS):
mgmt_candidates.append((idx, speaker, speech))
else:
qa_or_mod.append((idx, speaker, speech))
total_words = len(rec["content"].split())
cand_words = sum(len(x[2].split()) for x in mgmt_candidates)
saved_pct = (1.0 - cand_words / total_words) * 100.0
print(
f"标的: {rec['symbol']} | 财季: {rec['year']} {rec['period']} | 召开日: {rec['date']} | "
f"全文: {len(rec['content'])} 字符 ({total_words} 词, {len(paragraphs)} 段)"
)
print(
f"角色切分: 管理层发言 {len(mgmt_all)} 段 (命中 AI 订单与业绩指引候选段: {len(mgmt_candidates)} 段, 共 {cand_words} 词, 压缩无效 Token {saved_pct:.1f}%) | "
f"提问与主持: {len(qa_or_mod)} 段\n"
)
for idx, spk, txt in mgmt_candidates[:4]:
print(f"=== [候选段 #{idx} | 高管: {spk} | 词数: {len(txt.split())}] ===")
print(txt + "\n")
终端实跑输出(2026-11-10 实测,下方节选前 4 个核心候选段关键句):
标的: AVGO | 财季: 2025 Q3 | 召开日: 2025-09-04 | 全文: 41767 字符 (7375 词, 60 段)
角色切分: 管理层发言 19 段 (命中 AI 订单与业绩指引候选段: 6 段, 共 2180 词, 压缩无效 Token 70.4%) | 提问与主持: 41 段
=== [候选段 #2 | 高管: Hock Tan | 词数: 905] ===
Thank you, Ji. And thank you everyone for joining us today. In our fiscal Q3 2025, total revenue was a record $16 billion, up 22% year on year... And our current consolidated backlog for the company hit a record of $110 billion... AI Semiconductor revenue of $5.2 billion, which was up 63% year on year... Last quarter, one of these prospects released production orders to Broadcom. And we have accordingly characterized them as a qualified customer for XPUs. And in fact, has secured over $10 billion of orders of AI regs based on our XPUs... For Q4, 2025, we forecast AI semiconductor revenue to be approximately $6.2 billion, up 66% year on year...
=== [候选段 #3 | 高管: Kirsten Spears | 词数: 674] ===
Thank you, Hock. Let me now provide additional detail on our Q3 financial performance... Free cash flow in the quarter was $7 billion and represented 44% of revenue... Now moving to guidance. Our guidance for Q4 is for consolidated revenue of $17.4 billion up 24% year on year. We forecast semiconductor revenue of approximately $10.7 billion up 30% year on year. Within this, we expect Q4 AI semiconductor revenue of $6.2 billion up 66% year on year. We expect infrastructure software revenue of approximately $6.7 billion up 15% year on year... We expect Q4 adjusted EBITDA to be 67%...
=== [候选段 #4 | 高管: Hock Tan | 词数: 66] ===
I don't know about exciting, Kirsten, but I do. I thought before we move to questions, I should share an update. The board and I have agreed that I will continue as the CEO of Broadcom through 2030 at least. These are exciting times for Broadcom, and I'm very enthusiastic to continue to drive value for our shareholders...
=== [候选段 #7 | 高管: Hock Tan | 词数: 84] ===
I think it's both, Ross. But to a large extent, it's the fourth quarter customer that we now add on to our roster. Which we will ship pretty strongly in 2026, I should say... And the addition of a fourth customer with immediate and fairly substantial demand. Really put our really, changes our thinking of what '26 would be starting to look like. Thank you.
三、 大模型结构化提取 Prompt 与博通(AVGO)指引拆解成品表
将上述脚本筛出的高管候选段投喂给 DeepSeek 或其他大模型,并使用如下结构化提取 Prompt(强制要求模型区分「当季已实现业绩」与「未来业绩/订单指引」,并附带英文原句证据防幻觉):
你是一位资深半导体与美股量化投研工程师。以下文本来自 FMP 数据源 `/stable/earning-call-transcript` 接口返回的博通(AVGO)2025 Q3 财报电话会(召开日:2025-09-04)管理层完整发言段落。
请严格区分【当季已实现业绩(Actuals)】与【面向未来的管理层指引/承诺(Forward Guidance)】,输出一张包含以下 5 列的 Markdown 表格:
1. 发言人及职务(类别:当季已实现 / 未来指引承诺)
2. 核心事项与业务线
3. 目标兑现时间表(结合 2025 Q3 电话会召开时点折算具体财季或年份)
4. 披露具体数量或金额(严禁编造原文未提及的数字,金额统一用中文“亿美元”表示)
5. 电话会英文证据原句(逐字引用原文关键句)
大模型实际提取的博通(AVGO)2025 Q3 核心业绩与前瞻指引表
| 高管姓名与口径分类 | 涉及业务与关键事项 | 预计兑现周期 | 定量金额或比例 | 纪要原文引用句子 |
|---|---|---|---|---|
speaker (category) |
business_item |
target_timeline |
disclosed_amount |
verbatim_evidence |
| Hock Tan, CEO(当季已实现) | 全公司合并积压订单(Backlog)与 Q3 AI 半导体营收 | 截至 2025 FY Q3 末 | 积压订单 1,100 亿美元;Q3 AI 芯片营收 52 亿美元(同比 +63%,XPU 占比 65%) | "our current consolidated backlog for the company hit a record of 110 billion USD... AI Semiconductor revenue of 5.2 billion USD, which was up 63% year on year" |
| Hock Tan, CEO(未来订单指引) | 新增第 4 家合格定制 AI 加速卡(XPU)客户量产大单 | 2026 财年强劲出货 | 已斩获 超 100 亿美元 XPU AI 机柜订单(客户具体名称未披露) | "characterized them as a qualified customer for XPUs. And in fact, has secured over 10 billion USD of orders of AI regs based on our XPUs... Which we will ship pretty strongly in 2026" |
| Kirsten Spears, CFO(未来业绩指引) | 2025 财年 Q4 合并总营收与分业务营收指引 | 2025 FY Q4 | 合并总营收 174 亿美元(同比 +24%):其中半导体 107 亿美元、软件 67 亿美元 | "Our guidance for Q4 is for consolidated revenue of 17.4 billion USD up 24% year on year. We forecast semiconductor revenue of approximately 10.7 billion USD" |
| Hock Tan & Kirsten Spears(未来业绩指引) | 2025 财年 Q4 AI 半导体营收与调整后 EBITDA 利润率指引 | 2025 FY Q4 | Q4 AI 半导体营收约 62 亿美元(同比 +66%);Q4 调整后 EBITDA 占营收 67% | "For Q4, 2025, we forecast AI semiconductor revenue to be approximately 6.2 billion USD, up 66% year on year... We expect Q4 adjusted EBITDA to be 67%." |
| Hock Tan, CEO(治理承诺) | CEO 留任期限正式承诺 | 至少留任至 2030 年 | 任期延续至 2030 年及以后 | "The board and I have agreed that I will continue as the CEO of Broadcom through 2030 at least." |
四、 核心接口参数与工程小结
在实际投研管线中,可将上述两个接口组合为自动化定时任务:
- 电话会排期目录接口(
/stable/earning-call-transcript-dates):传入symbol(如AVGO),返回历史所有已发布纪要的(fiscalYear, quarter, date)列表; - 单季电话会逐字稿接口(
/stable/earning-call-transcript):传入symbol、year(财年)与quarter(1~4),返回含完整多角色对话文本的content字段。