AMD’s AI Powerhouse Unleashes Ryzen 400 at Computex 2025, as Cyberattacks Double on SMBs

When the virtual curtain dropped on the closing of the virtual Computex 2025 on September 21 in Taipei, AMD (with its bold announcement) snatched the limelight by purporting to be the AI hardware leader, making the neural processing power of 50 tera operations per second (TOPS) and its claimed ability to be embedded in low-end laptops a reality.

As a barnburner keynote, delivered by CEO Lisa Su in Austin, the chips were framed as a democratizing force to Nvidia leadership, offering 30 per cent faster AI workloads to creators and gamers without having to spend a fortune. At under $1,000 prices, the Ryzen AI 400 may carve out 80 per cent of its stranglehold on AI accelerators, prompting a price war that spreads through data centres to push devices all over the globe.

The flagship of the series, the Ryzen AI 9 HX 370, has a dedicated NPU to perform on-device generative workloads, such as real-time video editing and language translation, and is 25% faster on Adobe Premiere renders when compared to Intel Lunar Lake.

Su showed the chip that drives an example Asus Zenbook, on which Stable Diffusion created photorealistic art in under 10 seconds offline, without cloud access. We are not just creating chips, Su said, surrounded by partners such as Microsoft and Unity, but an open AI ecosystem, aimed at the masses of people. The lineup can utilise the open-source AMD ROCm software stack to port, eliminating the proprietary lock-in that afflicts competitors.

Non-silicon Computex buzz spread wider. AMD provided a glimpse of enterprise options on servers, suggesting a full rollout in Q4 2025 to enter the hyperscale training market as an alternative to Nvidia’s H200S.

IDC analysts project that the shift would capture 15 per cent of the $ 200 billion AI chip market in 2027, particularly with the U.S. imposing export restrictions on advanced semiconductors to China, opening opportunities for AMD to offer its so-called compliant designs. However, the supply chain buzzed with supply chain squeezes, associated with the TSMC Taiwan fabs, and the stock dropped 0.5% after the news, within wider semiconductor market volatility.

SMB Cyber Onslaught: Attacks Nearly Double in H1 2025, Demanding AI Defenses

In a sharp contrast to the success of AMD, a gloomy report by the cybersecurity company SentinelOne fell on September 21, stating that the number of cyberattacks on small and medium-sized businesses (SMBs) has almost doubled in the first half of 2025, soaring 98 per cent year-over-year to more than 1.2 million cases.

Ransomware was on the most prominent list with 42 per cent of the breaches, and average costs of downtime increased to 1.8 million per incident- up 22 per cent in 2024. The flood, which is fueled by AI-enhanced phishing and deepfakes scams, highlights the fragility of the sector as digital transformation is on the rise without security investment.

It was an anonymous dataset of 5,000 SMBs worldwide that found manufacturing and retail to be the most affected, with 65 per cent of the victims located in the Asia-Pacific. It is the soft underbelly of an economy; hackers understand that SMBs are the low-hanging fruit, as SentinelOne CTO Alex Stamos warned in a webinar.

There were such notable cases as a Michigan auto parts supplier that was crippled by LockBit 4.0, which took three weeks, and a UK e-tailer who lost customer data worth $2.5 million to a supply-chain attack. Attackers, who are more state-sponsored by North Korea and Russia, use generative AI to create hyper-personalised content. They have chased after them, avoiding conventional filters with success rates of 95%.

Reactions are building up: the U.S. Cybersecurity and Infrastructure Security Agency (CISA) has launched a free AI-enhanced threat scanner on SMBs, and the EU, with its NIS2 Directive, has set zero-trust architectures to be deployed by 2026.

Vendors such as Microsoft promised to allow endpoint protection grants worth 500 million dollars, and bundle Copilot with Azure. Nevertheless, the adoption is slow–28 percent of SMBs use multi-factor authentication everywhere–and the federal subsidies are being demanded. The accounting in boardrooms is inhumane: companies that do not practice cyber hygiene pay 3 times the cost of insurance policies, according to Lloyds of London.

Meta Llama 4 Tsunami: Open-source AI Model Floods Programming Tools

It was OpenAI’s shadow that was huge when on September 21, Meta open-sourced Llama 4, which is free to use commercially and can match GPT-5 in multimodal abilities. The model was downloaded 2 million times in the first hour through Hugging Face and is not only particularly effective at code generation, image captioning, and ethical reasoning but also has built-in precautions to prevent hallucinations. The Meta AI team of Zuckerberg billed it as the people frontier model with edge deployment fine-tuning kits on smartphones.

Early users were glorifying: IBM was incorporating Llama 4 into Watsonx to do enterprise analytics, cutting down query times by 40 per cent, and Bangalore startups were spinning up custom chatbots during the night. LMSYS Arena benchmarks topped it in the open model category, with the Mistral Large 2 missing the top spot by 5 points in the human evaluation.

However, the release does not come free of thorns. European watchdogs initiated an investigation into training data sourcing, were scared of GDPR violations, and senators of the United States grilled Meta on export controls to hostile countries. Llama 4 is priced at zero and may speed up a model bazaar; however, IP lawyers point to a litigation minefield on scraped web corpora.

The decline concurs with the Meta $50 billion capex commitment on AI, which invests in data centre facilities in Ireland and Texas. Shares were up 2.1 per cent, better than the Nasdaq, as analysts anticipate revenue on higher tiers of hosting.

Meta’s Llama 4 Tsunami: Open-Source AI Model Floods Developer Tools

On September 21, Beijing digital lords imposed a fine of 150 million and 50 million on Kuaishou and Weibo, respectively, the biggest to date under the 2025 Cybersecurity Law, on grounds of ideological shortcomings in content regulation.

The Cyberspace Administration referred to 12,000 unfiltered posts about sensitive areas such as slowdowns in the economy and autonomy in Hong Kong, and required AI rewrites at 99 per cent accuracy rates. Apple conducts daily audits on the platforms, and failure to comply may result in the app stores being delisted.

Timed to coincide with the holiday of the National Day, the crackdown cooled a creator economy with an estimated worth of $300 billion: revenues related to live-streams at Kuaishou plummeted by a fifth in Q3 projections, and user activity at Weibo declined by 11%.

The numbers of influencers reached 50 million, and they all moved into the Telegram channels, VPNs sales skyrocketed by 45 percent on Taobao. The actions bolster TikTok negotiations in the United States, with the case of ByteDance mentioning harmonious compliance globally. However, it is a creativity cull amongst the young people in China- state media lauds about purity, yet there are sheets of discontent in underground forums.

China’s Kuaishou-Weibo Fines Escalate Content Wars, Sparking Creator Exodus

On September 21, amid macro jitters, ShopVision Technologies, a clinic in Vancouver, raised $4.1 million in a seed round that was led by Brightspark Ventures, to put AI in e-commerce personalisation. The computer vision tools of the platform forecast the intention of the shopper with gaze patterns and increase conversion rates by 35% among customers, such as merchants on Shopify. CEO Priya Singh has a vision of a frictionless retail where pilots in 200 stores will bring in $50 million in uplift.

The capital, which will be accompanied by BDC Capital, will value ShopVision at 20m after the money, with the funds being used in hires in the field of ML ethics. One of the differences between its patents and those of Amazon is its privacy-first advantage, federal learning, without data centralisation in a congested sector. Greater VC sentiment became warmer: AI retail deals reached 1.2 billion YTD, according to PitchBook.

Echoes and Edges: Zinwais Keyboard Revival to NFL Streaming War

Niche innovations were the order of the day. Zinwa flaunted its Q27 Android 16 smartphone as an ode to BlackBerry, with a slide-out QWERTY, costing 699, aimed at typists tired of using virtual keyboards. The 1TB RAM accelerator by Gigabyte that uses stealth technology in workstations claims to enable AI training without swaps but costs 5,000 making it the prerogative of professionals.

Heat waves: The free Eagles-Chiefs NFL simulcast on Peacock attracted 15 million viewers, which put a strain on the Caroline charges at YouTube TV. The resignation of the CTO of M&S after the cyber attack is a warning of the dangers of digitisation in retail.

Positive-trending markets saw AMD gain 3.2% to $185, while Meta rose 2.1%. However, cyber fears deflected CrowdStrike’s 1.4% gain. As the innovations of September 21st become solid, the mantra of tech remains: power the future, but patch the present. The salvo of AI used by AMD can even the playing field, but SMB shields are years behind–open models will enable or open? The race of computers is quickening, with weaknesses flowing.