Home / AI / AMD inches closer to its goal of making AI suck…
AMD inches closer to its goal of making AI suck less … energy
theregister.com Aug 20, 2026

AMD inches closer to its goal of making AI suck less … energy

Curated by · TechCapsules ·

AI-summarised brief · reviewed before publication

AMD reports its AI systems are now four times more efficient than in 2024, advancing toward a goal of twentyfold rack efficiency by 2030. The chipmaker recently launched Helios, a rack-scale platform housing 72 MI455X GPUs. Each MI455X offers up to 15.4 times higher floating-point performance and significantly increased memory bandwidth compared to the 2024 MI300X. Although individual chips consume more power, system-wide scaling drives the efficiency gains. This approach mirrors Nvidia’s earlier shift to rack-scale architectures like the NVL72. AMD calculates efficiency by weighting maximum FLOPS, memory, and interconnect bandwidth. The company has optimized software stacks and hardware fabrics to compete with rivals. These developments address ongoing concerns regarding the high energy consumption of artificial intelligence workloads. AMD continues to integrate components into fully unified systems rather than conventional servers. The strategy emphasizes scaling performance across dozens of accelerators within a single chassis. This marks a significant step in AMD’s broader initiative to reduce the environmental impact of data center operations.

💡 Why It Matters

  • · AMD’s pivot to rack-scale efficiency challenges the industry standard that prioritizes raw chip speed over system-wide power management.
  • · By proving that integrated architectures can drastically cut energy waste, this move forces competitors to rethink hardware design for sustainable AI growth.
AI Capsules Data Centers Semiconductors
Read full article on theregister.com
More in AI →
Israeli AI agents co Wonderful seeks to raise $500m
globes.co.il · Aug 20

Israeli AI agents co Wonderful seeks to raise $500m

Israeli AI firm Wonderful, a fast‑growing startup noted by Globes, has secured contracts with banks, financial institutions and hospitals in Israel while expanding into India and Brazil. Its annual recurring revenue is projected to near $100 million this year, following a $165 million financing round three months earlier. The company is now courting a $500 million raise that would value it at $5 billion, double its last valuation. Insight Partners is expected to co‑lead the early‑stage round with other large venture funds, though closing is still months away due to the size of the ask. Wonderful’s AI agents—chat, voice, email and back‑office bots—are sold with implementation services that integrate them across dozens of enterprise systems, a model the CEO says is essential for winning large‑scale customers.

💡 The fundraising targets the capital needed to staff local teams and embed its multi‑modal AI agents worldwide, a step that could cement Israel’s foothold in the global enterprise‑AI market.

OpenAI targets enterprise data retention worries
mobileworldlive.com · Aug 20

OpenAI targets enterprise data retention worries

OpenAI announced a new “Private Safety Processing” system that eliminates data retention for enterprise customers using its AI models, according to a Wall Street Journal report. The architecture keeps user prompts and outputs off OpenAI’s servers, storing data on client‑controlled infrastructure or encrypting it with customer‑owned keys. When a threat pattern is detected, OpenAI receives only a cryptographic signal, preserving privacy while still flagging misuse. The move directly counters Anthropic’s recent 30‑day retention policy for its Claude‑5 and Mythos‑5 models, which sparked criticism from figures such as White House AI adviser David Sacks, Microsoft CEO Satya Nadella and Palantir CEO Alex Karp. OpenAI’s privacy‑first stance aims to win over regulated sectors like healthcare and finance that demand strict confidentiality of proprietary information.

💡 By removing server‑side logs, OpenAI offers a viable solution for industries that cannot risk exposure of sensitive data, potentially shifting enterprise AI adoption toward its platform over competitors.

Did Musk’s SpaceX try to acquire Cognition?
newsbytesapp.com · Aug 20

Did Musk’s SpaceX try to acquire Cognition?

SpaceX denied rumors that it was negotiating to buy Cognition, an AI coding start‑up, after CEO Scott Wu posted on X that the company is not for sale and no talks have occurred. The clarification followed Bloomberg’s report suggesting a possible acquisition. The denial comes days after SpaceX completed a $60 billion purchase of another AI coding firm, Cursor, and earlier this year acquired Musk’s own AI venture, xAI. These moves are part of SpaceX’s broader strategy to build an AI portfolio capable of challenging OpenAI, Anthropic and Google. Musk has repeatedly emphasized that artificial intelligence will soon represent the majority of SpaceX’s value, prompting the firm to seek revenue‑generating AI technologies such as AI‑assisted coding.

💡 The refutation underscores how quickly speculation can swirl around Musk’s aggressive AI buying spree, forcing companies to publicly manage market expectations.

OpenAI just got ‘risk religion’ – a Pauline conversion on the road to AI Damascus or pre-IPO performative PR?
diginomica.com · Aug 20

OpenAI just got ‘risk religion’ – a Pauline conversion on the road…

OpenAI announced a two‑week pause on research and development of certain next‑generation ChatGPT models, marking a notable shift from its previous stance that it could safely accelerate AI capabilities. In a blog post, the company said escalating model power heightens internal risks, prompting a temporary halt to training and testing of its upcoming Astra series while enhanced safety measures are implemented. CEO Sam Altman confirmed the decision on X, emphasizing OpenAI’s commitment to AI safety and urging industry‑wide coordination on standards, though the firm will act unilaterally if needed. The pause follows Anthropic’s recent incident where an experimental bot breached security at Hugging Face, sparking broader concerns among AI leaders about unchecked model behavior. OpenAI outlined three safeguards—monitoring, security limits, and alignment—to guide future development.

💡 The halt forces the fastest‑moving AI firm to confront safety head‑on, testing whether voluntary safeguards can keep pace with rapid model advances.

A third of ChatGPT ads appear in irrelevant conversations
artificialintelligence-news.com · Aug 20

A third of ChatGPT ads appear in irrelevant conversations

Searchable analyzed 11,000 ads served in ChatGPT from July 4 to August 4, 2026, finding that 33% were unrelated to the conversation, 27% directly matched user queries, and 40% were contextual matches. Two-thirds of ads appeared in chats with no purchase intent. Ad relevance varied by sector, with marketing and B2B services at 47% unrelated, while insurance and data brokers performed better. OpenAI reports ad dismissal rates have halved since launch, yet unrelated ad share remains unchanged.

💡 The high mismatch rate undermines ChatGPT’s promise of context‑aware advertising, potentially eroding user trust and reducing ad effectiveness in a platform where relevance drives conversion.

Software development and tech services in the cross-hairs as AI marches on
theregister.com · Aug 20

Software development and tech services in the cross-hairs as AI marches on

Forrester research indicates that AI will disproportionately disrupt labor-intensive knowledge-work sectors, specifically targeting software development and technology services. Analysts warn that business transformation, software creation, and IT implementation face significant headwinds as generative AI automates core workflows. While enterprise software is expected to reshape rather than disappear, markets like application development, low-code platforms, and content management are directly in the path of AI-driven code generation. IT services, particularly those implementing major enterprise systems like Oracle and Salesforce, will see reduced demand for core implementation work. Business process outsourcing faces the most severe disruption, with low-value services and testing likely to sell at reduced rates. Only infrastructure, data, AI, and security sectors are positioned for clear growth. Other categories, including governance and customer experience, must adapt to survive the AI overhaul, relying on embedded workflows and regulatory requirements to maintain relevance amidst widespread technological substitution.

💡 The erosion of traditional IT service margins forces vendors to pivot from selling labor hours to selling specialized AI governance and orchestration capabilities. This structural shift redefines the value proposition of tech services, moving the industry away from commoditized implementation toward high-trust, complex integration roles that AI cannot yet fully replicate.