Nvidia and OpenAI abandon unfinished $100B deal in favour of $30B investment
https://www.ft.com/content/dea24046-0a73-40b2-8246-5ac7b7a54323
The much-hyped $100 billion Nvidia-OpenAI deal collapsing into a slimmer $30 billion investment reveals cracks in the AI hype bubble. This downsizing signals deep strategic misalignments or valuation inflation that mainstream coverage ignores. Investors should beware the overblown promises around AI’s immediate monetization; the real costs and integration challenges are forcing a reality check.
Toronto-based chip startup Taalas, which hardwires AI models into custom silicon to achieve faster inference, raised $169M, bringing its total funding to $219M
https://www.reuters.com/world/asia-pacific/chip-startup-taalas-raises-169-million-help-build-ai-chips-take-nvidia-2026-02-19/
Taalas’ fresh $169 million infusion spotlights the intensifying race to bypass Nvidia’s GPU dominance with niche AI silicon. But custom silicon hardwiring AI models risks obsolescence as models evolve rapidly. Betting big on fixed-function chips could backfire if flexibility and software adaptability remain undervalued.
Trump administration reaches a trade deal to lower Taiwan’s tariff barriers
https://apnews.com/article/trump-taiwan-china-trade-deal-2b1743397ba33010463d41132b75ce53
This trade deal, framed as strengthening US-Taiwan ties, actually escalates geopolitical risks by sidelining China without addressing semiconductor supply chain fragilities. Lower tariffs might boost Taiwanese exports but also deepen Washington’s reliance on a fragile island amid intensifying cross-strait tensions. The deal risks entangling tech supply with geopolitical flashpoints.
Don’t Trust the Salt: AI Summarization, Multilingual Safety, and LLM Guardrails
https://royapakzad.substack.com/p/multilingual-llm-evaluation-to-guardrails
The assumption that AI guardrails can reliably mitigate LLM risks is dangerously naive. Multilingual and cultural nuances expose current safety frameworks as patchy at best, leaving room for serious misinformation or biases. Overreliance on these “guardrails” feeds complacency, ignoring that AI governance requires constant, context-aware vigilance beyond technical fixes.
A US grand jury indicted two former Google engineers and one of their husbands for allegedly stealing trade secrets relating to the Tensor chip for Pixel phones
https://www.bloomberg.com/news/articles/2026-02-20/ex-google-engineers-charged-with-stealing-phone-processor-tech
This indictment exposes the thin line between talent mobility and industrial espionage in the hypercompetitive chip race. It also signals growing government willingness to criminalize internal leaks, which could stifle legitimate knowledge flow and innovation in the US tech sector. The narrative of “trade secret theft” conveniently obscures systemic pressures driving insider threats.
Sources: SoftBank plans to form a consortium to build a $33B power plant in Ohio, set to produce 9.2 GW for AI data centers, as part of the US-Japan trade deal
https://asia.nikkei.com/economy/trade-war/trump-tariffs/softbank-to-form-consortium-for-33bn-trump-deal-power-plant
SoftBank’s $33 billion power plant project underscores the overlooked carbon and infrastructure costs of AI’s explosive growth. Locking in vast energy demand for AI data centers under the guise of trade cooperation masks a looming sustainability crisis. This move also cements AI’s dependence on geopolitically sensitive energy partnerships, far from the “clean tech” narrative.
Sources: Hacker News, Techmeme, AP News, Ars Technica | Compiled 2026-02-20