文章目录
- [AI科技热点日报 | 2026年08月02日](#AI科技热点日报 | 2026年08月02日)
-
- [📌 今日摘要](#📌 今日摘要)
- [1、AI 头部观点:马斯克与 Altman 在 AI 奇点上罕见同频](#1、AI 头部观点:马斯克与 Altman 在 AI 奇点上罕见同频)
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- [2、小米宣布多款手机 8 月 2 日起上调售价,旗舰上探千元](#2、小米宣布多款手机 8 月 2 日起上调售价,旗舰上探千元)
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- [3、Apple 内部预警:AI 算力短缺或致产品延期发布](#3、Apple 内部预警:AI 算力短缺或致产品延期发布)
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- [4、长鑫科技上市首周股价涨 523%,市值突破 3.61 万亿元](#4、长鑫科技上市首周股价涨 523%,市值突破 3.61 万亿元)
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- [5、DeepSeek-V4-Flash 正式版 API 开放公测,Agent 能力较预览版暴涨 6 倍](#5、DeepSeek-V4-Flash 正式版 API 开放公测,Agent 能力较预览版暴涨 6 倍)
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- [6、华为昇腾 950 + DeepSeek V4:国产算力 + 国产模型的"组合拳"](#6、华为昇腾 950 + DeepSeek V4:国产算力 + 国产模型的"组合拳")
-
- 事件概要
- [来源 / Sources](#来源 / Sources)
- 行业数据速览
- 媒体人解读
- 参考来源
AI科技热点日报 | 2026年08月02日
综合整理自 证券时报、36氪、TechWeb、PChome、钛媒体、东方财富等科技财经媒体当日报道,聚焦 AI 大模型、机器人、芯片、算力、AI 终端与产业金融动态。

📌 今日摘要
2026 年 8 月 2 日,AI 圈迎来多线共振:马斯克与 Sam Altman 在 AI 奇点判断上罕见同频,OpenAI 同期向内部员工释放"算力短缺或致产品延期"的预警;国产侧,小米宣布多款手机当天起上调售价 300 元起,长鑫科技上市首周股价上涨 523%、市值逼近 3.6 万亿元;模型层,DeepSeek-V4-Flash 正式版 API 昨日开放公测、Agent 能力较预览版暴涨 6 倍,价格仅为 Claude 同档位模型的 1/90。
On August 2, 2026, the AI sector saw multiple coordinated signals: Elon Musk and Sam Altman converged on the AI singularity narrative from opposite ends of the industry, while Apple warned internally that AI compute shortages could delay product releases. On the Chinese side, Xiaomi raised smartphone prices by at least RMB 300 starting today, ChangXin Memory's stock surged 523% in its debut week with a market cap near RMB 3.61 trillion, and DeepSeek opened public beta of its V4-Flash API whose Agent capabilities jumped 6x versus the preview at 1/90 the price of comparable Claude models.

1、AI 头部观点:马斯克与 Altman 在 AI 奇点上罕见同频
事件概要
2026 年 8 月 1 日,海外科技媒体 The News 报道,Elon Musk 与 OpenAI CEO Sam Altman 在"AI 奇点(singularity)"叙事上罕见一致。Altman 7 月 25 日做客 Relentless 播客时表示:"我们现在差不多已经身处奇点时代",并预言人类当前可完成的工作中将有一大批被 AI 接管,已可自动化处理约 30%-40%。Musk 于 7 月 31 日在 X 平台发文称:"AI 在很多方面都已经超越了人类。我们正处于奇点之中。"AI 奇点概念由未来学家 Ray Kurzweil 在 2005 年提出,描述 AI 自我改进速度超过人类监控与控制能力的理论情境。
On August 1, 2026, overseas tech outlet The News reported that Elon Musk and OpenAI CEO Sam Altman have converged on the AI singularity narrative --- a rare alignment between two long-time rivals. Speaking on the Relentless podcast on July 25, Altman said "we are roughly already in the singularity era" and predicted that a large share of work currently done by humans would be taken over by AI, with roughly 30-40% already automatable. Musk followed up on July 31 via a post on X: "AI has surpassed humans in many respects. We are in the singularity." The singularity concept was proposed by futurist Ray Kurzweil in 2005 as a theoretical scenario in which AI improves itself faster than humans can monitor or control it.
来源 / Sources

2、小米宣布多款手机 8 月 2 日起上调售价,旗舰上探千元
事件概要
2026 年 8 月 2 日,小米正式对旗下多款机型执行新一轮价格调整,覆盖小米 17 系列、REDMI K90 全产品线和 Turbo 5 系列共 9 款机型:Redmi Turbo 5 起售价从 2299 元上调至 2599 元,Turbo 5 Max 由 2499 元上调至 2799 元;Redmi K90 标准版从 2799 元提至 3099 元,K90 Max 由 3199 元提至 3499 元,K90 Pro Max 由 4199 元提至 4499 元,K90 至尊版由 2999 元提至 3299 元;小米 17 从 4499 元提至 4799 元,17 Pro 由 4999 元提至 5399 元,17 Pro Max 由 5999 元提至 6499 元,整体涨幅在 300 元至 500 元区间。钛媒体、PChome 等媒体报道指出,本轮调价与 DRAM、NAND 内存以及 SoC、散热模组等核心元器件成本上行直接相关;小米内部高管此前已在公开场合多次谈及内存涨价带来的行业普涨压力。
On August 2, 2026, Xiaomi officially implemented a new round of price hikes covering nine SKUs across the Xiaomi 17 series, the full Redmi K90 lineup and the Turbo 5 family: the Redmi Turbo 5 moved from RMB 2,299 to RMB 2,599, the Turbo 5 Max from RMB 2,499 to RMB 2,799; the Redmi K90 standard from RMB 2,799 to RMB 3,099, the K90 Max from RMB 3,199 to RMB 3,499, the K90 Pro Max from RMB 4,199 to RMB 4,499, and the K90 Ultra from RMB 2,999 to RMB 3,299; the Xiaomi 17 went from RMB 4,499 to RMB 4,799, the 17 Pro from RMB 4,999 to RMB 5,399, and the 17 Pro Max from RMB 5,999 to RMB 6,499 --- a 300 to 500 RMB increase per SKU. Coverage from TMTPost and PChome attributes the move to rising DRAM/NAND, SoC and thermal-module costs, while Xiaomi executives have publicly acknowledged the industry-wide memory-price pressure on multiple prior occasions.
来源 / Sources

3、Apple 内部预警:AI 算力短缺或致产品延期发布
事件概要
2026 年 8 月 2 日,多家英文科技媒体报道,Apple 在近期内部沟通中向员工释放预警:因 AI 算力供应持续紧张,部分新品节奏可能被迫推迟。报道指出,Apple 现有 AI 路线图对 GPU/NPU 资源、HBM 显存以及推理集群的依赖度持续上升,而供应链优先级排序让 Apple 在排队获取最新一代 AI 加速硬件时无法优先满足。Apple 未对外披露具体受影响产品线,但市场普遍关注 iPhone、Apple Intelligence 与 Mac AI 模块的发布窗口。
On August 2, 2026, multiple English-language tech outlets reported that Apple has internally warned staff that AI compute shortages could delay the rollout of some products. The reports note that Apple's AI roadmap is increasingly dependent on GPU/NPU resources, HBM capacity and inference clusters, while the supply chain's priority order prevents Apple from being first in line for the newest AI accelerators. Apple has not disclosed specific product lines affected, but the market is closely watching the release windows for iPhone, Apple Intelligence and Mac AI modules.
来源 / Sources

4、长鑫科技上市首周股价涨 523%,市值突破 3.61 万亿元
事件概要
2026 年 7 月 27 日登陆科创板的国产 DRAM 龙头长鑫科技(688825.SH)股价持续走强。截至 8 月 2 日,长鑫科技上市后第一周股价累计上涨 523%,总市值约 3.61 万亿元,超越 2020 年中芯国际 532 亿元募资规模,成为科创板史上最大 IPO。证券时报报道指出,长鑫科技是目前国内唯一一家完全聚焦内存赛道的上市企业,填补了 A 股市场本土内存制造能力企业代表的空白;Kimi K3 等国产前沿模型的快速放量,也带动了对国产 DRAM、HBM 的需求侧联动。野村证券同期给出 116 元/股目标价、相当于 7.76 万亿元目标市值。
ChangXin Memory (688825.SH), China's leading domestic DRAM maker that listed on the STAR Market on July 27, 2026, has seen its stock price climb 523% in its first week, taking its market cap to roughly RMB 3.61 trillion --- surpassing SMIC's 2020 RMB 53.2 billion raise and making it the largest IPO in STAR Market history. Securities Times notes that ChangXin Memory is currently the only A-share-listed company fully focused on the memory track, filling a structural gap in domestic capital markets, and that the rapid ramp of frontier domestic models like Kimi K3 is also driving demand-side pull for local DRAM and HBM. Nomura simultaneously issued a RMB 116 / share target price implying a RMB 7.76 trillion target market cap.
来源 / Sources

5、DeepSeek-V4-Flash 正式版 API 开放公测,Agent 能力较预览版暴涨 6 倍
事件概要
2026 年 7 月 31 日深夜,DeepSeek 悄然上线 V4-Flash 正式版 API 并于 8 月 1 日开放公测。官方基准数据显示,在 Agent 智能体终极测试中,V4-Flash 正式版得分由 V4-Pro 预览版的约 15.8 跃升至 25.2,相对预览版暴涨约 6 倍,且仅依靠后训练迭代、模型骨架未做改动(284B 总参 / 13B 激活 MoE、1M 上下文与 Preview 完全一致)。V4-Flash 输入价 1 元/百万 tokens、输出价 2 元/百万 tokens;V4-Pro 输入 12 元、输出 24 元,对比同期 Claude 同档位模型价格优势超过 100 倍。Artificial Analysis 与 Arena.ai 8 月 1 日同步更新跑分,V4-Flash 智能指数 50 分,Arena 前端代码竞技场 1586 分。DeepSeek 官方表示,下半年华为昇腾 950 超集群批量上市后,Pro 版本价格将进一步下调。
On the night of July 31, 2026, DeepSeek quietly pushed the production V4-Flash API and opened public beta on August 1. The company's benchmarks show that on the Agent end-to-end test, V4-Flash production scored 25.2 versus the V4-Pro preview's 15.8 --- a roughly 6x jump --- achieved purely through post-training iteration with no change to the model backbone (284B total / 13B active MoE parameters, 1M context window, identical to the preview). Pricing stands at RMB 1 / 1M input tokens and RMB 2 / 1M output tokens for V4-Flash, with V4-Pro at RMB 12 / 24 per million tokens --- a 100x+ price advantage versus comparable Claude models. Artificial Analysis and Arena.ai updated scores on August 1, putting V4-Flash at 50 on the intelligence index and 1586 on the Arena frontend-code leaderboard. DeepSeek says V4-Pro pricing will be cut further once Huawei Ascend 950 super-clusters ship at scale in H2.
来源 / Sources
- TechWeb:DeepSeek-V4-Flash 正式 API 上线,Agent 能力暴涨 6 倍,价格仅为 Claude 的 1/90
- DeepSeek 官网:DeepSeek-V4-Flash 正式版 API 已上线公测

6、华为昇腾 950 + DeepSeek V4:国产算力 + 国产模型的"组合拳"
事件概要
2026 年 8 月初,腾讯云开发者社区综合披露 DeepSeek V4 系列与华为昇腾 950 平台深度协同进展:V4-Flash 在缓存命中场景下输入价低至 0.2 元/百万 tokens,V4-Pro 输入 1 元、输出 12 元/百万 tokens。DeepSeek 明确表示,目前的低价是在"算力受限"前提下给出的,待昇腾 950 超节点批量交付后,价格还将进一步大幅下调。从底层算力到顶层模型,DeepSeek 与华为昇腾正在打通一条从训练、推理到应用部署的国产 AI 全栈路径。
In early August 2026, the Tencent Cloud developer community reported on the deep synergy between DeepSeek's V4 family and Huawei's Ascend 950 platform: V4-Flash hits RMB 0.2 / 1M input tokens on cache hits, while V4-Pro is priced at RMB 1 / 1M input and RMB 12 / 1M output tokens. DeepSeek explicitly noted that current pricing is set under "compute-constrained" conditions and will drop further once Ascend 950 super-nodes ship at scale. From silicon to foundation models, DeepSeek and Huawei Ascend are stitching together a full domestic AI stack covering training, inference and deployment.
来源 / Sources
行业数据速览
| 指标 | 数据 | 备注 |
|---|---|---|
| 长鑫科技上市首周涨幅 | +523% | 总市值约 3.61 万亿元,刷新科创板 IPO 纪录 |
| DeepSeek-V4-Flash Agent 基准分 | 25.2 | 较 V4-Pro 预览版 15.8 提升约 6 倍 |
| DeepSeek-V4-Flash 输入价 | 1 元 / 百万 tokens | 缓存命中可低至 0.2 元 / 百万 tokens |
| DeepSeek-V4-Flash 输出价 | 2 元 / 百万 tokens | 对比同档 Claude 价格优势约 100 倍 |
| Redmi K90 Pro Max 涨幅 | +300 元 | 由 4199 元上调至 4499 元 |
| 小米 17 Pro Max 涨幅 | +500 元 | 由 5999 元上调至 6499 元 |
| 小米三年 AI 投入规划 | 超 600 亿元 | 雷军 3 月春季发布会公开口径 |
| Kimi K3 总参数 | 2.8 万亿 | 全球首个开源 3 万亿级别模型 |
媒体人解读
今天的几条主线放在一起看,能看到一个清晰信号:AI 已经从"训练竞赛"切换到"分发与成本竞赛"。一方面,Altman 与 Musk 的"奇点同频"更多是叙事层面的同框,而非技术路线共识------但它客观上把市场对 AGI 时间表的预期再次前移;另一方面,Apple 内部关于"算力短缺导致产品延期"的预警,说明即便是手握千亿美元现金的巨头,也无法在 GPU/HBM 排队赛中稳坐头排。
国产侧的看点则更为具体:长鑫科技一周 +523% 的股价曲线,反映出资本市场对"国产存储 + 国产 AI"双轮驱动的强烈认可;DeepSeek-V4-Flash 用 0.2 元/百万 tokens 的缓存命中价,把百万级上下文 AI 的使用门槛直接打到中小开发者和个人也能承受的区间;小米在涨价 300-500 元的同日仍在持续加码 AI 投入,本质上是用硬件毛利补贴模型与生态建设------这是 2026 年中国 AI 公司最具代表性的"硬件 + 模型 + 渠道"打法。
对技术管理者而言,今天有三件事值得立刻排进下周例会:一 ,重新评估对单一云厂商算力的依赖,引入昇腾 950、海外 H100/H200 之外的国产算力备选;二 ,对调用 DeepSeek、Claude、GPT 三家模型的应用做一次 Token 成本压力测试,把缓存命中率与提示词长度纳入优化指标;三,评估 AI 终端涨价潮对自家 SaaS 定价、用户增长曲线的影响------硬件成本的传导远比想象中快。
Looking at today's headlines together, the signal is clear: AI has shifted from a "training race" to a "distribution and cost race." On one hand, the Altman--Musk "singularity convergence" is more a narrative alignment than a technical consensus --- but it objectively pushes market expectations for AGI timing forward again. On the other, Apple's internal warning about compute shortages delaying products shows that even trillion-dollar balance sheets can't sit at the front of the GPU/HBM queue.
The Chinese-side storyline is more concrete: ChangXin Memory's +523% first-week curve reflects capital-market conviction in a "domestic memory + domestic AI" dual-engine thesis. DeepSeek-V4-Flash's RMB 0.2 / 1M cache-hit token price compresses the entry barrier for million-token-context AI to a level individual developers and SMBs can actually afford. Xiaomi's RMB 300-500 price hike on the same day it keeps stepping up AI investment is essentially cross-subsidizing model and ecosystem build-out from hardware margins --- the most representative 2026 "hardware + model + channel" playbook for Chinese AI companies.
For technical managers, three things deserve next week's steering meeting: (1) reassess single-cloud compute dependency and add a domestic alternative (Ascend 950) alongside overseas H100/H200; (2) run a token-cost stress test across DeepSeek / Claude / GPT, with cache-hit rate and prompt length as optimization KPIs; (3) model the impact of the AI-terminal price hike wave on your own SaaS pricing and user-growth curve --- hardware cost pass-through arrives much faster than people expect.
参考来源
英文来源
- The News (overseas tech media, July 31, 2026): Musk and Altman converge on AI singularity
- Artificial Analysis benchmark (Aug 1, 2026): DeepSeek-V4-Flash scores
中文来源
- 证券时报:长鑫科技上市首日市值突破 3 万亿元,A 股迎"存储"新标杆
- 东方财富网公司资讯频道
- PChome 电脑之家:存储涨价顶不住了 REDMI K90 全系涨价 200 元
- 36 氪:WWDC 2026,苹果首届 AI 还账大会
- 36 氪:马斯克重启对 OpenAI 和奥特曼的诉讼
- TechWeb DeepSeek 标签页(DeepSeek-V4-Flash 报道)
- DeepSeek 官网:DeepSeek-V4-Flash 正式版 API 已上线公测
- 腾讯云开发者社区:华为昇腾 950 + DeepSeek V4 王炸组合
- 腾讯云开发者社区:DeepSeek V4 联手华为昇腾 950
本文内容综合整理自上述公开信息来源,媒体人解读仅代表个人观点,仅供参考。
作者署名:刘一说 🎙️
发布日期:2026 年 08 月 02 日