New chips, a cybersecurity alarm, and the AI adoption gap

Today's brief spans the infrastructure layer — new chips from South Korea, California, and Texas — a collective alarm over AI-enabled cyberattacks, a privacy dispute over enterprise data, and fresh evidence that most organizations have yet to turn AI investment into measurable results.

SK Hynix breaks ground on a $4 billion AI memory plant in Indiana

South Korea's SK Hynix held a groundbreaking ceremony this week for a $4 billion semiconductor facility in West Lafayette, Indiana. The plant will produce HBM4E (high-bandwidth memory) chips using advanced packaging technology, with volume production targeted for 2029. SK Hynix projects that AI-driven demand for memory will outpace global supply through at least 2030 — a forecast that underpins its decision to accelerate US-based manufacturing capacity ahead of schedule.

Over 100 tech firms warn governments of an AI cyberattack wave

A coalition of more than 100 companies — including OpenAI, Microsoft, and Anthropic — has signed an open letter calling for coordinated defense against AI-powered cyberattacks. The letter cites recent incidents, including a breach at Hugging Face's platform, as early indicators of what the signatories describe as an imminent and escalating threat. The coalition urges governments to treat cyber defense as a public-safety priority and act before the window for preemptive action narrows further. The letter marks an unusual moment of cross-competitor solidarity on a shared risk.

OpenAI's Jalapeño chip outperforms Nvidia's Blackwell in benchmark tests

OpenAI's custom inference chip, called Jalapeño, has topped the InferenceX benchmark published by industry analysts SemiAnalysis, surpassing Nvidia's flagship GB300 (Blackwell) on two key metrics: tokens delivered per user and throughput per kilowatt of power consumed. Co-designed with Broadcom, the chip is planned for small-volume deployment by end-2026, with wider rollout in 2027. If the results hold in production, it would represent a meaningful shift in the economics of large-scale AI inference — the compute-intensive process of running a trained model to generate responses.

Apple launches M6 and M5 Ultra chips for on-device AI

Apple has unveiled two new chips: the M6, which powers the updated Mac mini, and the M5 Ultra, which drives the Mac Studio. Both chips are engineered to run AI workloads locally — without routing data through a cloud server — and deliver significantly higher AI compute than their predecessors. Apple is positioning the hardware as an alternative to cloud-dependent AI processing for professionals and organizations handling sensitive data, extending its strategy of keeping inference on the device itself.

Anthropic activates shared memory across Claude and Cowork without an opt-out

Anthropic has quietly enabled shared memory between its Claude AI assistant and its Cowork collaboration platform, allowing context from one product to carry into the other. Enterprise customers were given no option to disable the feature before it was activated, prompting criticism from data-protection advocates and corporate clients who handle confidential information. The episode has renewed debate about whether AI vendors are moving fast enough on consent mechanisms for enterprise users. Anthropic has not announced a timeline for introducing opt-out controls.

Nvidia forecasts 70% revenue growth; shares rise 6%

Nvidia reported a quarterly sales forecast that exceeded Wall Street's estimates by a wide margin and projected roughly 70% revenue growth for its next fiscal year. The figures signal that corporate and government spending on AI infrastructure remains strong, countering concerns that the buildout might be slowing. Nvidia's stock rose approximately 6% in a single trading session on the news. The company's results are closely watched as a leading indicator for the broader AI hardware market, given its dominant position in chips used to train and run large language models.

Most finance teams deploy AI but cannot measure what it returns

A survey of 902 finance executives by advisory firm Protiviti reveals a sharp gap between AI adoption and strategic readiness. Seventy-seven percent of finance organizations have deployed AI tools, yet only 35% say they can confidently measure the return on that investment, and just 14% report having a detailed AI strategy in place. The findings suggest that many organizations adopted AI rapidly — often in response to competitive pressure — without building the measurement frameworks needed to evaluate whether those tools generate value. Protiviti recommends that finance leaders prioritize governance structures and clear performance metrics before expanding deployments further.