The artificial intelligence landscape has accelerated wildly in mid-2025 - with new models shattering learning benchmarks while tariffs and megadeals reshape its supply chain.
Azzet's Robot's Rise column lays down the latest ructions of rapidly evolving artificial intelligence - the most important tectonic shift in innovation we've seen since the advent of the internet.
This wasn't how the year was scripted; DeepSeek, ChatGPT-4, Claude and now Grok 4 large language models (LLMs) were meant to show incremental gains, yet here we are, watching xAI's creation leap ahead of the pack in what feels like Humanity's Last Exam (HLE) - the final moments prior to the advent of superintelligence.
Yet the frenzy reveals more about raw market forces than any hype cycle ever could. Let's take a look:
Tariff clouds gather over hardware
U.S. tariff hikes are slamming AI hardware costs - threatening to inflate prices for chips and sensors by double digits.
President Trump's proposed 20-50% duties on imports from 23 countries, now delayed to August 1, could drive up AI wearable production expenses by 10-15%, according to Yale's Budget Lab.
A recent QYResearch report pegs the global AI wearables market at US$29 billion in 2023, projecting US$276 billion by 2032 - but warns of adoption slowdowns from these supply squeezes, SNS Insider reported.
"For AI wearables, tariffs on components like 5G chipsets and MEMS sensors could increase production costs by 10–15%, per Yale's Budget Lab", the analysis states.
Compounding this, restrictions on AI chip exports to Malaysia and Thailand risk eroding U.S. edge against China.
M&A frenzy and supercharged investments
AI mergers are exploding - clocking US$65 billion in the first half alone and representative of ~10% of total US M&A volume of US$750 billion.
CoreWeave's US$9 billion all-stock acquisition of Core Scientific stands out, fortifying data centres for heavy AI workloads.
Other blockbusters include HPE's US$14 billion Juniper deal - with Meta, Microsoft, Amazon, and Alphabet eyeing US$325 billion in combined capex this year.
Israel has committed US$13.2 million toward building AI data infrastructure across key sectors including agriculture, health, climate and security.
The funding initiative, led by the Innovation Authority alongside multiple government ministries, aims to create accessible databases for training AI models.
Meanwhile, the debate around AI's actual productivity impact continues to deliver inconclusive findings.
Recent research has produced mixed results on whether AI tools actually make workers more productive, with some studies showing gains while others suggest workers may take longer to complete tasks.
xAI and Grok 4 surge ahead
xAI unveiled Grok 4 on 9 July - promptly dominating benchmarks and the HLE.
“With respect to academic questions, Grok 4 is better than PhD level in every subject, no exceptions,” Elon Musk proclaimed in a livestream.
Its Grok 4 Heavy variant, leveraging multi-agents and tools for a 44.4% HLE score, eclipsed rivals such as Gemini 2.5 Pro and OpenAI's o3.
However, it's only available for those willing to fork out a whopping US$300/month.
xAI found itself in damage control mode recently too, after its Grok chatbot began posting antisemitic content on X.
The bot praised Hitler and made offensive comments about Jewish people, prompting swift backlash from users and advocacy groups.
xAI issued an apology and blamed the incidents on a software update meant to make Grok less "politically correct."
Despite the Grok controversy, SpaceX committed $2 billion to invest in xAI as part of a broader $5 billion funding round.
The investment represents nearly half of xAI's equity raise and values the combined xAI-X company at $113 billion following their earlier merger.
Rivals scramble with delays and hires
OpenAI postponed its open model release indefinitely on 11 July - citing safety checks as an "unexpected and quite amazing" breakthrough, CEO Sam Altman told TechCrunch.
“We need time to run additional safety tests and review high-risk areas. We are not yet sure how long it will take us,” Altman said.
Meanwhile, Meta's Superintelligence Labs have snagged talents like Alexandr Wang from Scale AI, Reuters reported.
And Perplexity launched its AI browser Comet on 9 July, while Anthropic and Google continue readying LLM upgrades for release.
LLM leaderboard
Here's how the top LLMs stack up at the moment, as per TypingMind Blog:
Model: GPT - Best for: Content creation - Pros: Fluent, high context - Cons: Hallucinations in tech
Model: Claude - Best for: Coding, reasoning - Pros: Logical, grounded - Cons: Higher cost
Model: Gemini - Best for: Video analysis - Pros: Long input handling - Cons: Inconsistent output
Model: Perplexity - Best for: Factual search - Pros: Real-time citations - Cons: Limited creativity
Model: Llama - Best for: Self-hosting - Pros: Open-source, cost-effective - Cons: Variable quality
Model: DeepSeek - Best for: Math, logic - Pros: Strong benchmarks - Cons: Less polished
Model: Grok - Best for: Creative coding - Pros: Witty, coherent - Cons: Occasional inaccuracies
