Ex-Apple designer behind iPhone Air joins AI startup Hark

Abidur Chowdhury, a former Apple designer involved in the iPhone Air project, is joining AI startup Hark as head of design.

Hark, founded by Brett Adcock, who also leads robotics firm Figure AI, is hiring engineers from companies such as Google, Meta, and Amazon.

Adcock said he has invested US$100 million of his own funds into Hark, which is building AI models with its first release planned for this summer.

The company currently has 30 engineers and is aiming to reach 100 in the first half of 2026.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

With a $100 million budget plus a 30‑engineer team, a summer release puts Hark below labs like OpenAI or Anthropic, which run larger teams with bigger compute budgets.
On Hacker News (HN), a thread estimated GPT‑3‑scale training on older NVIDIA V100 GPUs took about 36 days 1. NVIDIA says H100 data‑center GPUs deliver up to 4× faster GPT‑3 (175B) training than A100s 2. H100 units list at roughly $25,000–$40,000 3. Complete 8‑GPU systems run about $297,000–$521,000 1. The $100 million budget likely funds a modest cluster compared with Tesla’s reported ~35,000‑H100 setup 4.
The summer target likely means a smaller first model or use of rented cloud GPUs. Building dedicated H100 stacks can hit supply constraints and long lead times 3.

Pre‑launch testing, red‑teaming (adversarial testing), and safety checks ramp up near release. Vendors that sell compute, tools, or evaluation services can watch Hark’s hiring and partnerships to time offers.
Data annotation and labeling firms can time outreach by tracking Hark’s engineering headcount toward its 100‑engineer target in the first half (H1) of 2026.
Cloud GPU platforms like Lambda Labs and RunPod, which rent GPU servers 4, could see more demand if Hark picks flexible rentals over owned gear. H100 cloud prices have dropped 64–75% to about $2.85–$3.50 per hour 4.