A new article argues that AI application startups are being squeezed from above by model companies that absorb popular features, and from below by system integrators with direct model vendor backing. It highlights revenue data: OpenAI ARR reportedly hit $25B, Anthropic over $47B, while Cursor reached $500M and Glean $300M—only a few apps cross $100M. Early tools like Jasper were quickly replicated by ChatGPT, and AI search faces direct competition from model-native search. The author sees the biggest threat in the Anthropic-DXC alliance, which bypasses app developers for enterprise delivery. The future, it contends, lies in AI-powered physical services (robotaxi, AI hospitals) where value extends beyond software.
Leaked audited financials reveal OpenAI generated $13.07 billion in revenue but posted a net loss of $38.53 billion in 2025, driven by $41.55 billion in equity revaluation losses and $20.92 billion in operating losses. Payments to Microsoft for R&D and services totaled $17.2 billion. Concurrent Sensor Tower data shows ChatGPT's share of the AI assistant market slipped to 46.4% by May 2026, down from over 50% earlier in the year, as Google Gemini (27.7%) and Anthropic Claude (10.3%) gained ground. Anthropic now leads in enterprise API spending share and engineer poaching. Chinese open-source models from DeepSeek, Alibaba Qwen, and others are capturing developer usage via dramatically lower pricing and competitive performance, further eroding OpenAI's pricing power. While OpenAI has confidentially filed for an IPO, the financials force a critical narrative shift: whether its multi-hundred-billion-dollar losses represent infrastructure for a monopolistic AI entry point or an increasingly expensive arms race against diversified competitors.
On June 15, 2025, DeepSeek completed a ¥51 billion Series A round at a post-money valuation of ~¥400 billion, with CATL and its investment platform Puquan Capital contributing approximately ¥5 billion, the second-largest external investment after Tencent. This follows CATL’s recent acquisitions of a controlling stake in HVDC leader Zhongheng Electric and IDC operator Century Internet. The three deals form a ‘green electricity → energy storage → HVDC distribution → computing power’ vertical chain, enabling CATL to supply energy storage, peak-valley arbitrage for lower electricity costs, green direct power, and backup power to AI data centers. Experts view the investment as a strategic pivot: CATL is transitioning from a battery manufacturer to an AI computing energy infrastructure platform, leveraging its government relationships and energy network to secure electricity quotas and negotiate tariffs, thereby unlocking higher valuation multiples beyond manufacturing PE ratios.
Google DeepMind published a paper titled "The Topological Trouble With Transformers," arguing that the Transformer architecture has a structural flaw in state tracking: as sequences grow, internal state updates are pushed into deeper network layers and become inaccessible to later processing steps. The paper demonstrates this defect with failures in a number-guessing game and the "bank" ambiguity test, where models give contradictory answers despite having disambiguated the word earlier. Chain-of-thought prompting mitigates the problem by externalizing hidden states as visible text, but it is computationally expensive and does not fix the underlying architectural limitation. The authors advocate shifting focus toward recurrent architectures that explicitly pass state along the sequence dimension, such as MAMBA, RWKV-7, and DeltaNet, and suggest future directions like coarser-grained recurrence and staged training from feedforward pretraining to recurrent fine-tuning.
In an AP interview and at the Coherent groundbreaking ceremony in Sherman, Texas, Jensen Huang announced plans to quadruple the capacity of the world's most advanced 6-inch indium phosphide laser fab within a year, a critical component for optical interconnects that scale AI data centers. He outlined a five-layer AI stack (energy, chips, infrastructure, models, applications) and argued that energy production is the foundational bottleneck, lauding market-driven investment in solar and nuclear without subsidies. Huang predicted global investment of trillions of dollars over the next decade to replace sixty years of legacy computing infrastructure with AI-native systems that perform real-time reasoning on facts. He framed the AI boom as a historic opportunity for US reindustrialization, claiming it has already created 600,000 manufacturing jobs and will restore dignity to builders and makers alongside information workers. On China, he acknowledged Nvidia’s 95% market share before export controls and advocated for balancing national security with maximum exports to prevent competitors from becoming dominant.