How should we view DeepSeek's newly released model, DeepSeek-Math-V2?

After circling back, PRM is not a dead end—it’s a difficult yet correct path, and one that can even be meta-PRM. What makes DeepSeek (DS) remarkable is its willingness to blaze that trail for you—and unlike certain companies whose names include the word 'open', DS doesn’t hide the broader direction after paving the way. The strongest model capabilities are currently being showcased in turn by OpenAI, Google, and Anthropic; meanwhile, DeepSeek and Qwen serve as spiritual leaders—one offering conceptual frameworks, the other providing high-quality foundational models from its lab. The era of large language models is indeed delightfully chaotic. Moreover, engineering optimization is undoubtedly the right path—whether it’s reducing training costs or inference costs, both are essential. There’s no need to dismiss NVIDIA (‘Old Huang’) based on this alone: algorithm engineers will inevitably redeploy every GPU-hour saved by engineering teams elsewhere—ultimately, results will always stem from collective effort.

Key directions in the broader CRM domain—PRM (Part II)

In real-world cases, PRM is typically deployed alongside CRM. This integration is shaped by commercial systems both in China and globally—and also depends on product characteristics. For example, consider exporting products to Thailand in Southeast Asia. From a global revenue perspective, Thailand may contribute only a small share; thus, manufacturers may choose not to establish their own local office there, instead partnering with a Thai distributor to manage the local market—including marketing, sales, and after-sales service. The manufacturer and its Thai partner would then interact across logistics, financial flows, and information flows. However, if the exported goods are highly complex—or if fulfillment timelines are long and delivery challenging—the manufacturer must strike a balance between cultivating local partners and setting up its own subsidiary in Thailand. Often, a parallel approach is adopted: the manufacturer directly manages and engages key accounts (KAs), while local partners handle mid-to-large, mid-sized, and small-to-medium enterprises (SMEs). Clear boundary definitions are critical throughout this process. Even in direct-sales models—which don’t require PRM—CRM management systems remain fundamentally indispensable. Therefore, the design of PRM governance strategies and methodologies must be tailored to country-specific contexts, market scale, product type, and delivery complexity. Another example: air conditioners targeting the Chinese domestic market clearly split into two marketing strategies: 1. Named Account (NA) customer system—typically served via direct sales (brand-led). In many cases, however, key partners still play vital supporting roles for KAs—providing premium services, participating in delivery and fulfillment, or even handling order fulfillment. 2. Non-Named Account (Non-NA) customer system—generally comprising SMEs or end consumers, covered through channel partners (partner-led). One major client we previously served—a leading Chinese air-conditioning and white-goods enterprise—operates two distinct sales management systems: a channel-sales management system and a direct-customer sales management system. By collaborating with real-estate developers on service and product sales, the company embeds its appliances directly into home-purchasing packages—enabling ‘turnkey delivery with built-in appliances’ in premium residential projects. This contributes significantly to performance metrics. Simultaneously, the company maintains city-level agents across China and provincial-level distributors, who cover small-B clients and mass consumers via dealer networks. Meanwhile, major e-commerce platforms—treated as KAs—are managed directly by the manufacturer, including service delivery and pricing strategy formulation. Therefore, selecting the most competitive market-facing approach hinges on four core factors: market, competition, product, and channel. Accurate market sizing—and the precision of such estimates—is a top priority for every player in the industry. In other words: How big is the cake? How big will it be next year? Will it still exist in the future—and how might it evolve? This involves interpreting current market feedback and forecasting future trends.

PRM's Q2 Earnings and Revenues Miss Estimates, Sales Up Y/Y

Aug 8, 2026

Perimeter Solutions reported a Q2 2026 net loss of $181.6 million and missed earnings estimates, despite revenues growing 31% year-over-year to $213.8 million. Growth was driven by acquisitions and strong performance in Fire Safety and Specialty Products segments.

The Globe and Mail

2026-07-31

Perimeter Solutions Q2 Results: Net Loss Widens to $181.6M on Fees

2026-07-31

Perimeter Solutions Reports Q2 Revenue $213.8M, Missing FactSet Estimates

2026-07-31

Perimeter Solutions Reports Second Quarter 2026 Financial Results

2026-07-31

Perimeter Solutions Q2 2026 Earnings: Specialty Products Drives 31% Sales Growth

2026-07-31

Perimeter Solutions posts Q2 2026 net loss $181.6M, adjusted EPS $0.35 on $213.8M revenue