XRP faces $2 test: $1.3B outflows vs. weak network usage

CPU, Quietly Returning to the Center of the AI Computing Power Stage

Over the past three years, AI computing power narratives have been dominated by GPUs. However, starting in 2026, this story began to shift. While training large models remains GPU-intensive, the rapid growth of inference and AI agent workloads, which require high levels of task orchestration, concurrency, and data flow management, has highlighted a renewed critical role for CPUs. These are tasks GPUs are not designed to handle. Intel's recent launch of the Xeon 6+ processor, built on its Intel 18A process and featuring up to 288 efficiency cores (E-cores), exemplifies this strategic pivot. It is positioned not as a mere companion to GPUs but as the essential "control plane" for AI infrastructure, optimized for high-density, energy-efficient, and high-throughput workloads characteristic of AI agents and inference. This "CPU resurgence" is not about CPUs outperforming GPUs in raw computation. It reflects a systemic bottleneck: as AI scales from training single models to deploying countless intelligent agents, the demand for coordination and data handling surges. Major cloud providers are also developing their own high-density ARM-based server CPUs for similar workloads. However, Intel's success with this strategy faces significant challenges. Competition includes NVIDIA's integrated CPU-GPU solutions, the expanding adoption of cloud vendors' in-house ARM CPUs, and the crucial market test of Intel's 18A manufacturing process against rivals like TSMC's N2. In conclusion, CPUs are indeed reclaiming a central, though redefined, role in AI compute—managing the complex orchestration that enables massive-scale AI deployment. While the trend is clear, which company will ultimately lead this CPU resurgence remains an open question to be decided in the data centers of 2027 and beyond.

marsbitHá 7m

TON Rebrands Native Token As Gram, Reviving Original White Paper Name

The native token of the Toncoin (TON) network is being rebranded to 'Gram,' reviving its original name from the project's first white paper. This change, announced by Telegram co-founder and CEO Pavel Durov, is the fourth step in his "Make TON Great Again" (MTONGA) initiative. The transition period is expected to take about three weeks, and a new logo has been previewed. The rebrand follows Telegram's official return to the ecosystem in May as its largest validator, after a six-year absence that began with a 2020 legal dispute with the SEC. Durov stated that the move "returns to our roots" and paves the way for future developments, with three more steps remaining in the MTONGA roadmap. At the time of the announcement, Gram was trading around $2.02.

bitcoinistHá 18m

If We Make Musk's Wealth Tangible

Elon Musk stands on the threshold of becoming the world's first trillionaire, with a current net worth estimated at approximately $970 billion. This vast fortune, accumulated over 31 years of entrepreneurship, is predominantly held as equity in his companies. The largest share comes from SpaceX's pre-IPO stake, valued around $538 billion, followed by Tesla holdings worth about $167 billion. Additional assets include stock options, stakes in The Boring Company and Neuralink, and other investments. The scale of his wealth is difficult to grasp. It surpasses the annual GDP of over 125 countries and represents about 3% of the U.S. GDP, double the relative wealth of historical figures like John D. Rockefeller. On average, his wealth grows by $992 per second. For a typical U.S. household earning the median income, it would take over 11 million years to amass an equivalent sum. To illustrate its purchasing power, $970 billion could buy roughly 2.4 million average American homes, acquire all NFL and NBA teams with over $500 billion remaining, or purchase companies employing over 4 million people combined. Musk's wealth is largely illiquid, tied to company stock. His story reflects the immense value creation in modern technology sectors like electric vehicles, commercial spaceflight, and artificial intelligence, which have also generated significant returns for investors and employees.

marsbitHá 25m

After Collaborating with 35+ DeFi Projects, Pink Brains Discovers the New 2026 KOL Marketing Rules

After collaborating with over 35 leading DeFi projects on marketing over three years, Pink Brains identifies a key shift for effective marketing in 2026: prioritizing the user journey over traditional campaign tactics. The most effective marketing mirrors how users actually behave—starting with discovery on social platforms like X (formerly Twitter), followed by data-driven verification on sites like DefiLlama, and finally, participation with small test funds. Success hinges on genuine, verifiable mechanisms, not just marketing hype. Current user interest centers on several key themes: new DeFi trends (RWA, perps, crypto x AI), meaningful airdrops requiring real contributions, real yield from protocol revenue, and tokens with value capture mechanisms directly tied to product usage. Case studies like Hyperliquid's HYPE (with its aggressive buyback program) and Venice's VVV (linking demand to AI compute) exemplify how strong tokenomics foster user retention. New trading venues like prediction markets, collectibles platforms, and GambleFi are also gaining traction, driven by verifiable activity. The article outlines common mistakes in DeFi KOL marketing, such as using creators unfamiliar with the product, generic messaging, or relying on a few top-tier KOLs. Instead, effective strategies align with different KOL types—educators, content creators, airdrop hunters, and niche experts—for various stages of the user journey. Ultimately, long-term user retention depends on a combination of a genuinely useful product, responsive support, community-aligned tokenomics, and strategic community building. The core takeaway is that sustainable growth stems from products whose value is validated by data and real-world utility, not just promotional efforts.

marsbitHá 25m

a16z：视觉AI的未来不是图片，而是代码

marsbitHá 1h