AMD Buys Taalas As Anthropic Signs $10B Compute Deal
AMD acquired Taalas to hardwire AI models into silicon, Anthropic signed a six-year $10 billion Volta compute deal in Norway, Black Forest Labs shipped FLUX 3, Palantir revenue rose 93%, and the EU AI Act's transparency rules began to apply on August 2.

A single week captured the AI sector's new phase: record industrial acquisitions, a live European rulebook, and infrastructure commitments measured in gigawatts and billions rather than demos.
At a glance: The dominant thread across the past 24 hours was capital consolidating around industrial and applied AI rather than chatbots. As Forbes detailed on August 7, Schneider Electric has moved aggressively into the space, agreeing to buy Norwegian industrial-AI and data firm Cognite for roughly $3.1 billion through its AVEVA subsidiary, and separately acquiring utility risk-intelligence startup AiDASH in an all-cash deal valued near $350 million. The pattern extends across the automation majors, with ABB reserving some $14 billion for future strategic acquisitions after selling its robotics division to SoftBank for $5.375 billion, and Emerson having completed its $7.2 billion AspenTech takeover. Analysts read the buying spree as a signal that operational AI embedded in energy grids, logistics and manufacturing has become the sector's most contested territory.
Technology advance: On the model front, Freiburg-based Black Forest Labs completed the public rollout of FLUX 3, its most ambitious release to date, which MarkTechPost described as a multimodal flow model spanning image, video, audio and even robot action prediction. The FLUX 3 Video tier generates clips up to 20 seconds long with native synchronized audio and dialogue, rendering at 720p natively with upscaling to 1080p and offering text-to-video, image-to-video, video continuation and lip-sync across roughly 14 languages. The company positioned the system as a single generative backbone capable of driving both creative production and embodied robotics, a notable departure from the still-image models that first made the German lab prominent. The release moved from limited early access to broader availability in early August, priced at about $0.17 per second of generated video.
Partnerships: Anthropic deepened its scramble for compute by signing a six-year, $10 billion agreement with Volta, an AI cloud startup founded only earlier in 2026, as first reported by TechCrunch on August 4. Under the arrangement, Volta will supply cloud capacity from a 133-megawatt data center in Norway built around Nvidia's next-generation Vera Rubin systems, with crypto-mining specialist Bitdeer participating as an infrastructure partner and Volta operating inside Nvidia's Cloud Partner program. The deal extends a run of capacity alliances for the Claude maker, which has recently struck compute arrangements with both Amazon and SpaceX. The scale underscores how frontier labs are locking in multi-year, multi-billion-dollar power and silicon commitments well ahead of demand, treating guaranteed access to advanced accelerators as a strategic moat as much as a procurement line item.
Acquisitions/expansions: AMD announced on August 6 that it had acquired Taalas, a startup that hardwires trained AI models directly into custom silicon, in a move CNBC framed as a bet on radically cheaper inference. Taalas, founded by chief executive Ljubisa Bajic alongside Lejla Bajic and Drago Ignjatovic, all former AMD and Tenstorrent engineers, builds chips that encode model weights into a mask-ROM fabric; its HC1 part embeds Meta's Llama 3.1 8B model across 53 billion transistors and reportedly delivers around 17,000 tokens per second per user at roughly 200 watts. Financial terms were not disclosed and the deal remains subject to customary regulatory approvals. For AMD, the acquisition adds architects fluent in its instruction sets and a compressed design cycle that can turn trained weights into a manufacturable chip in about two months.
Regulatory/policy: Europe's AI rulebook shifted from statute to enforcement on August 2, when transparency obligations under Article 50 of the EU AI Act began to apply, as Al Jazeera reported. Chatbots and interactive systems must now disclose that users are dealing with a machine, generators of synthetic image, audio, video and text must apply machine-readable marking, and emotion-recognition or biometric-categorization systems must notify affected individuals. Breaches can draw administrative fines of up to 15 million euros, about $17.3 million, or 3 percent of global annual turnover, whichever is higher. Executive Vice-President Henna Virkkunen framed the phase-in as an effort to make it easier to innovate without lowering the safety bar, even as the bloc's Digital Omnibus package pushed the tougher high-risk obligations covering employment, education and border management back to December 2027.
Finance/business: Palantir reset expectations for enterprise AI monetization with an August 3 report showing second-quarter revenue of $1.94 billion, up 93 percent year over year and comfortably ahead of the $1.8 billion consensus, according to Fortune. US commercial revenue surged 149 percent while government revenue climbed 90 percent, and the company posted net income of $1.1 billion, or 41 cents per share, above the 35-cent estimate. Palantir lifted its full-year 2026 revenue outlook to roughly $8.15 billion from an earlier $7.18 billion forecast and guided third-quarter revenue near $2.16 billion. Chief executive Alex Karp said the business was compounding at a rate and scale he had never witnessed, noting quarterly net income now exceeds the prior year's total revenue. Shares jumped about 13 percent to $142.01 in after-hours trading.
Sources: forbes, marktechpost, techcrunch, cnbc, aljazeera, fortune




