AMD is paying about US$2.9 billion for Taalas, a three-year-old Toronto startup that builds silicon tuned for a single AI model, according to a filing analysis published this week. Up to US$1.4 billion of the price is contingent on milestones. Taalas raised more than US$225 million from investors including Fidelity, and its chief executive previously founded Tenstorrent. The deal, first announced in August, is one of the largest takeovers of a Canadian private tech company and shows where the chip race has moved: away from training and toward cheaper, faster inference.
Communities in ten states voted on 41 local measures dealing with data centers, and voters sided against the projects in all but one, according to Ballotpedia data analyzed by CNBC. The votes ranged from outright bans to caps on electricity use, with Ohio alone holding 21 measures. The issue moved races up and down the ballot: Sherrod Brown beat Senator Jon Husted in Ohio after calling him the face of data centers, and Gina Hinojosa unseated Texas Governor Greg Abbott. A CNBC voter poll found 56% of voters said data centers were a major factor in their decision.
Data centres built for AI workloads now consume an estimated 70% of the world's memory chip output, according to IDC, and the three companies that control more than 95% of DRAM production have shifted capacity toward high-bandwidth memory. HBM takes about 23% of DRAM wafers, leaving consumer-grade DRAM and NAND flash short. DRAM prices have roughly doubled since early 2025, and IDC expects PC, tablet and smartphone prices to rise 10% to 20% by the end of 2026. Smartphone shipments are projected to fall 12.9% and the PC market 11.3%. New wafer capacity takes years to build, so analysts expect tight supply into 2027.
OpenAI chief executive Sam Altman made an unannounced weekend visit to South Korea, meeting about 50 Samsung executives in Yongin and reportedly holding talks with SK Group chairman Chey Tae-won. The trips centre on memory chips, where Samsung and SK Hynix hold roughly 63% of global DRAM revenue and about 83% of high-bandwidth memory. Samsung's 2027 HBM output and SK Hynix's 2029 output are already sold out, and Micron has warned the shortage could run into 2028. No new agreements on pricing or investment were announced. OpenAI has also cut its latest fundraising target to as much as $20 billion.
GitHub's Octoverse 2026 report says AI agents started more than 40% of pull requests on the platform in September 2026, about 110 million, up from 17% in March. Developers pushed code 17.9 billion times over the year, more than 2024 and 2025 combined. Agent-started pull requests touched multiple files 87% of the time, against 79% for human-started ones. GitHub says the surge has strained its infrastructure and contributed to repeated outages. The report shifts the debate from whether agents are used to how much review work they create.
Nvidia has told the contract manufacturers that build its AI servers to plan for price increases of more than 15% on systems shipping in early 2027, according to reports from Bloomberg, Tom's Hardware and Fortune. The cause is memory rather than silicon: DRAM, NAND and high-bandwidth memory prices have climbed through 2026 fast enough that even a company with gross margins near 75% cannot absorb them. HBM contract prices could rise another 70% to 140% in 2027, and hyperscalers are accelerating in-house chip programmes to reduce exposure to both the shortage and Nvidia's pricing.
Spending on artificial intelligence data centres is approaching $1tn across the industry, and investors are starting to question the returns. Technology shares fell sharply, Intel continued to lose ground and the Nasdaq opened at its lowest level in more than two months. Nvidia is still pushing hardware out quickly, including a plan to invest $3bn in Meta's equity offering as the social media company deploys more of its chips for AI agents. The company's pre-built data-centre design is also cutting deployment time from months to about two weeks.
Meta's updated Commercial Terms take effect on 30 October and authorise automated scrapers and crawlers against websites tied to any business account on its platforms. The Register first reported the change on 16 October, when the wording named AI training among the permitted uses. After criticism, Meta removed the explicit training reference and said the terms apply to its ads and business products. Business accounts have no opt-out from the scraping provision, which has alarmed news publishers, who argue the terms let Meta avoid licensing fees for journalism used to train models. Meta has not said whether content gathered this way will be used for training.
Google unveiled Gemini 4 Carbon on Wednesday, fifteen days after widening access to Gemini 4 Argon. The new frontier model is restricted to a small group of cybersecurity defence firms through Google's Fairwind programme rather than a public API. Google says Carbon finds vulnerabilities, builds exploits and writes patches better than Argon, and produces secure code at more than twice Argon's success rate on internal tests. It holds first place on SEC-bench Pro and ExploitGym. Introductory pricing matches Argon at $2 per million input tokens and $10 per million output tokens.
The United States and China are building rival technology clubs, and middle powers are discovering that membership may soon be exclusive. Washington has told partners they cannot sit inside both the US-led Pax Silica grouping and China's World AI Cooperation Organization. Kazakhstan tried: its AI minister joined US supply chain talks in Washington in June, then signed in Shanghai three weeks later as a founding member of the Chinese initiative. China has funded AI projects in at least 77 low- and middle-income countries, according to AidData, while US export controls keep the most advanced chips on one side of the line.