AT&T cut its AI coding bill 56 per cent
AT&T now routes 40 per cent of employee AI queries to open models and has cut coding costs 56 per cent, and Alibaba shipped its Wan3.0 video model days after a record $10bn share sale.
AT&T cut AI coding costs 56 per cent using open models The Information reported last Thursday that AT&T routes 40 per cent of employee AI queries to open models — Nvidia’s Nemotron, Meta’s Llama, Google’s Gemma — using LiteLLM routers to judge which tasks need a frontier model, cutting coding costs 56 per cent for a 2 per cent drop in quality. It wants that share at 60 to 70 per cent and intends to hold spending with Anthropic and OpenAI flat, on an internal platform running 45 billion tokens a day. Source
Alibaba shipped Wan3.0 days after a record $10bn share sale Alibaba rolled out the video model on Monday, generating 30-second clips from documents, spreadsheets, slides and web pages; it has been in public beta since 6 August. The placement that preceded it was the largest primary follow-on ever by a Hong Kong-listed company, and the proceeds go to Alibaba’s AI build-out, after quarterly earnings fell 75 per cent on soaring AI capital spending. Source
RUM Group signed a $13.7bn GPU deal it cannot yet fund RUM Group — the renamed Rumble, which hosts Truth Social — announced a six-year contract on Monday to supply an unnamed US cloud customer with GPUs and infrastructure from a site in Maysville, Georgia, and granted that customer warrants over 50.8 million Class A shares at $0.01 each. It also said it still has to build the data centre and buy the chips, and does not currently have the financing to do either; the shares rose about 8 per cent in premarket trading. Source
XPeng’s robotics arm raised $900M at a $6.3bn valuation The unit behind the IRON humanoid closed its first external round on Monday, led by IDG Capital with Gaorong Ventures and strategic cheques from Tencent and Alibaba. XPeng calls it the largest single private round in Chinese embodied AI and says IRON — 76 degrees of freedom, 21 of them in each hand — reaches mass production before the end of the year, with deliveries from 2027. Source
Meta hired Luke Metz for its Superintelligence Labs Axios reported on Monday that Metz, an early ChatGPT researcher, starts at Meta this week reporting to Alexandr Wang. He left OpenAI in 2024 for Mira Murati’s Thinking Machines, went back to OpenAI earlier this year, and is now leaving again. Source