AI Hiring Brief · 2026-08-25

32 frontier AI labs · 104 roles added in 24h · agihunt.info/en/jobs

Today at a glance

StepFun's hardware cluster: StepFun (阶跃星辰) posted 25 new roles in a single day — RTOS开发工程师 (RTOS development engineer), 芯片规划 (chip planning), GPU驱动与图形渲染 (GPU driver & graphics rendering), Android 设备稳定性/DFR 开发, and a run of camera/imaging titles (影像AE调试工程师, 影像AF调试工程师, 影像产品经理, 影像测评工程师, 影像SE) split between Shanghai's Xuhui district and Shenzhen's Nanshan district — this reads like a device hardware program standing up its own system software and imaging stack, not a model-only lab.
Physical compute build-out at OpenAI and Anthropic in parallel: OpenAI opened Systems Software Engineer, Silicon Bringup, Software Engineer, AI accelerator Runtime, and Manufacturing Quality Engineer – Datacenter Infrastructure, while Anthropic's Compute org added roles spanning Data Center Sourcing Manager, Manufacturing Engineer - Data Center Hardware, Global Supply Manager - Contract Manufacturing, and Environmental Health & Safety Manager, Data Center Construction and Operations — both labs are staffing custom silicon and datacenter manufacturing at the same time.
Scale AI leans into homeland defense: three of Scale AI's five new roles are Technical Program Manager (Homeland Layered Defense), Public Sector and Engagement Manager (Homeland Layered Defense), Public Sector, split between Colorado Springs, CO and Washington, DC.
Safety and policy roles surface at the research labs: Google DeepMind opened Research Scientist, World Models, DeepMind and Research Scientist, Provenance, DeepMind in London, and Anthropic added a Product Policy Manager, Product Risk role inside its Safeguards (Trust & Safety) org.
Microsoft AI is the volume outlier and the churniest: 22 additions and 27 closures today — the highest single-day closure count of any company — against a 7-day net of 225 additions, still the largest 7-day total tracked.

Company moves

OpenAI: 12 additions today (66 over 7 days), 7 closures. New roles pair Scaling-org silicon work (Systems Software Engineer, Silicon Bringup; Software Engineer, AI accelerator Runtime; Software Engineer, Model Runtime) with go-to-market hires (Partnerships Manager, Growth Programs Lead, Startups, Account Director, Startups | Tokyo) and an Enablement Lead, Government in Washington, DC — hardware and startup/government GTM expanding together.

Anthropic: 14 additions (97 over 7 days, the highest global figure after Microsoft AI), 6 closures. The Compute org dominates: Data Center Sourcing Manager, Transaction Principal, EU (London), Transaction Manager, Reporting and Controls Lead, Data Center Capacity Delivery, Regional Manager EMEA, Data Center Capacity Delivery, Repairs Program Lead, Product Engineer - Manufacturing Operations, Manufacturing Engineer - Data Center Hardware, Global Supply Manager - Contract Manufacturing, and Environmental Health & Safety Manager. Alongside that, an Account Executive, Startups and Partner Account Manager - Spain, Italy & MEA point to sales-channel expansion running in parallel with the infra build.

Google DeepMind: 5 additions (21 over 7 days), 5 closures. New roles split between research (Research Scientist, World Models; Research Scientist, Provenance) and infrastructure (Technical Program Manager, Compute Planning; Software Engineer, ML Accelerators), plus a Software Engineer, UX role — research and compute headcount moving together.

Microsoft AI: 22 additions (225 over 7 days, the largest 7-day total in the dataset), 27 closures. The spread runs from Principal Technical Program Manager and Datacenter Infrastructure Design Manager in Redmond to a cluster of Software Engineer / Senior Software Engineer roles across Beijing, Shanghai, and Suzhou tied to the M365 Storage Fabric and SWIQ Performance teams, plus AI Business Sales Manager - Corporate Accounts and a Polish-speaking Digital Enterprise Solution Sales Specialist - AI Workforce in the UK and Ireland — simultaneous infra, China engineering, and EMEA sales activity.

Meta AI: no additions today, 2 closures (12 added over 7 days). No new postings visible.

NVIDIA: 5 additions (34 over 7 days), 7 closures. New roles include ASIC Design Engineer - RTL and Circuits in Bengaluru, Senior SRAM Engineer, and Senior Technical Program Manager, DGX Cloud Software - Product and Services — chip design and cloud-infrastructure hiring opening together.

Apple AIML: 1 addition (18 over 7 days), 6 closures. The single new role, Machine Learning Engineer in Beijing, sits against a broader thinning of the list.

SpaceXAI: 1 addition (40 over 7 days), no closures. The new role, Mechanical Commissioning Technician (Data Center) across Memphis, Tennessee and Southaven, Mississippi, continues the data-center trades hiring seen in prior editions.

Lovable: 3 additions (6 over 7 days), 1 closure. Growth Creative Operations (London), Product Counsel (Stockholm), and Brand Designer, Web (New York; London; Stockholm) — marketing, legal, and design build-out at a small team.

Cohere: 2 additions (13 over 7 days), 4 closures. Both new roles are Forward Deployed Engineer postings for its Agentic Platform and Infrastructure, scoped to UK Public Sector — an agentic-deployment, public-sector focus.

Scale AI: 5 additions (10 over 7 days), no closures. Three Homeland Layered Defense roles plus a Mission Product Specialist in Washington, DC concentrate this round in defense-sector public work.

Cerebras: no additions today, 6 closures (5 added over 7 days). No new postings visible.

Moonshot AI (月之暗面 Kimi): 3 additions (7 over 7 days), no closures. 热点/事件营销负责人 Moment Maker (hot-topic/event marketing lead), AI Operations Builder, and GTM Engineer — marketing and go-to-market hiring rather than research.

StepFun (阶跃星辰): 25 additions (26 over 7 days), no closures — the largest single-day addition count of any company today. As above, the cluster is concentrated in RTOS, chip planning, GPU drivers, Android system stability, and camera/imaging tuning roles across Shanghai (Xuhui) and Shenzhen (Nanshan).

Baidu AI (百度 AI): 4 additions (19 over 7 days), 7 closures. 后训练算法专家 (post-training algorithm expert), 智能体算法专家 (agent algorithm expert), and AI产品经理 (AI product manager) in Beijing, plus 云南售前架构师 (Yunnan pre-sales architect) in Kunming — algorithm/agent research paired with regional pre-sales.

MiniMax (海螺): 2 additions (5 over 7 days), no closures. 服务端工程师（Maas方向）(backend engineer, MaaS direction) and 采购专员（AI Native）(procurement specialist), both in Shanghai.

ByteDance Doubao/Seed (字节跳动豆包/Seed): no additions today, 4 closures (20 added over 7 days), on the largest open-role count in the dataset at 968.

Meituan LongCat (美团·龙猫): no additions, 1 closure (1 added over 7 days).

Perplexity, Mistral AI, Together AI, Pika Labs, Aleph Alpha, Genmo, DeepSeek (深度求索), Zhipu AI (智谱 AI), ModelBest (面壁智能), Tencent Hunyuan (腾讯混元), 01.AI (零一万物), SenseTime SenseNova (商汤日日新), and Alibaba Tongyi Qwen (阿里通义千问) posted no additions today; Cursor saw a single closure with no new listings.

What it signals

Over the trailing 7 days, the global labs added roughly 554 roles combined versus about 90 across the China-based labs tracked here — Microsoft AI's 225 leads, with Anthropic at 97 and OpenAI at 66 close behind, while China's largest single mover is StepFun at 26, followed by ByteDance Doubao/Seed at 20 and Baidu AI at 19. Even accounting for ByteDance's enormous base of 968 open roles, the pace of new posting on the China side this week is markedly slower than the infrastructure-heavy build-out underway at the leading global labs.

That global build-out looks physical rather than purely research-driven. Anthropic's Compute org is assembling a near-complete manufacturing and data-center supply chain job family in one week — sourcing, transaction management, manufacturing engineering, repairs, and EHS — while OpenAI is filling out silicon bringup and accelerator runtime positions under Scaling. NVIDIA's ASIC Design Engineer and Senior SRAM Engineer postings show chip design remains an active hiring lane even at the leading silicon vendor, and SpaceXAI continues adding data-center trades roles. Read together, this is less about training bigger models and more about the labs bringing compute supply — chips, data centers, manufacturing — in-house.

StepFun's cluster stands out as the week's clearest hardware signal from a Chinese lab: 25 of its 26 seven-day additions landed in a single day, almost entirely in device-level infrastructure — RTOS, chip planning, GPU drivers, Android system stability, and camera/imaging tuning. That is a distinctly different hiring shape from Moonshot AI's marketing and GTM-oriented additions or MiniMax's smaller backend/procurement postings, suggesting StepFun may be standing up a hardware device product line alongside its model business rather than just scaling model-serving infrastructure.

Safety- and policy-adjacent hiring continued at low but steady volume: Google DeepMind's Research Scientist, World Models and Research Scientist, Provenance roles, Anthropic's Product Policy Manager, Product Risk, and Cohere's UK Public Sector Forward Deployed Engineer roles all point to labs keeping governance and public-sector deployment capacity staffed even as the bulk of this week's hiring volume sits in compute and manufacturing. Scale AI's Homeland Layered Defense cluster is the most concentrated single-purpose block of the week, adding five defense-sector roles with zero closures — a narrower, faster-moving signal than the broader infra hiring elsewhere.