WAIC 2026 Closes in Shanghai: China Launches Rival AI Governance Body
The World Artificial Intelligence Conference (WAIC) 2026 wraps up today in Shanghai after four days of exhibitions, product launches, and high-stakes geopolitical maneuvering. The headline event came on July 16, when Chinese President Xi Jinping presided over the launch of the World Artificial Intelligence Cooperation Organization (WAICO) — a new intergovernmental body headquartered in Shanghai with 29 founding member nations.
The signatories include major Global South players such as Indonesia, Brazil, South Africa, Malaysia, and Senegal, alongside Russia and Pakistan. UN Secretary-General António Guterres attended the signing ceremony, lending institutional weight to the initiative. Xi used his keynote to urge equitable global AI governance and warned against any single country dominating the technology — a pointed reference to the United States.
WAICO is widely seen as a direct counterweight to the US-led “Pax Silica” initiative, which has signed up 35 countries. The conference also featured over 300 new product debuts, including Moonshot's Kimi K3, which topped coding arena leaderboards within hours of its release. The dual-bloc dynamic makes it clear: AI governance is now a geopolitical chess match, and the Global South is the board both sides are playing on.
Google's Gemini 3.5 Pro Misses Another Launch Window
Google's highly anticipated Gemini 3.5 Pro has missed yet another launch deadline, and the delays are starting to hurt. The model was first promised for June at Google I/O 2026, then pushed to July 17. That date has now also passed with no release and no official update from Google.
According to 9to5Google, the core issue is coding performance. Google updated its training data in late June to address the gap, but internal benchmarks were reportedly disappointing. The company ultimately scrapped the original base model and restarted pretraining — a dramatic reset that explains the extended timeline.
The stakes are high. Gemini 3.5 Pro is supposed to introduce a 2 million token context window, a “Deep Think” reasoning layer, and autonomous workflow capabilities. But OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5 are already in production and setting the benchmark that Google now has to clear.
Making matters worse, four senior Gemini researchers departed for Anthropic in the last month, and Alphabet shares dropped roughly 4% on the delay reports. For a company that pioneered the Transformer architecture, the optics of falling behind on its own invention are particularly painful.
The Custom Chip Race: OpenAI and Anthropic Break Away from Nvidia
July 2026 may be remembered as the month the AI industry's Nvidia dependency began to crack. Two of the three leading labs are now building their own silicon — and the strategic implications are enormous.
OpenAI and Broadcom unveiled Jalapeño on June 24: a custom inference ASIC designed from scratch around OpenAI's model architecture. The chip was developed in just nine months — with OpenAI's own models assisting in the design process. Engineering samples are already running production workloads, including GPT-5.3-Codex-Spark, at target frequency and power. Early testing shows substantial performance-per-watt improvements over current GPUs, with deployment planned for late 2026.
Not to be outdone, Anthropic is in talks with Samsung Electronics to manufacture a custom chip using Samsung's cutting-edge 2nm SF2 process, which employs Gate-All-Around nanosheet transistors. The company hired Clive Chan — one of the earliest engineers on OpenAI's chip program — in June, signaling serious intent. While the talks are still exploratory, the direction is clear.
The logic is straightforward: custom inference chips eliminate the general-purpose overhead that GPUs carry, potentially cutting costs by 50% or more. With inference now accounting for the bulk of compute spending at frontier labs, even modest efficiency gains translate into billions in savings. AWS Trainium and Google TPUs showed the way — now the model makers themselves are following.
GPT-5.6: OpenAI's Three-Tier Strategy Reshapes the Market
OpenAI's GPT-5.6 family — Sol, Terra, and Luna — went generally available on July 9 and has rapidly become the benchmark against which everything else is measured. The three-tier naming is deliberate: Sol is the frontier flagship ($5/$30 per million tokens), Terra is the workhorse ($2.50/$15), and Luna is the lightweight option ($1/$6). All share a 1 million token context window.
Sol is launching on Cerebras infrastructure at up to 750 tokens per second, bringing frontier intelligence to production at unprecedented speed. The launch was accompanied by ChatGPT Work, an agent designed to carry out entire jobs rather than answer individual questions — a clear signal of where OpenAI sees the product evolving.
Notably, the US Commerce Department's Center for AI Standards and Innovation reviewed and cleared the models before wider access was granted — an early example of pre-deployment government review that could become the norm as regulation catches up to capability.
IPO Watch: Both OpenAI and Anthropic Heading for Public Markets
In a remarkable convergence, the two leading AI labs are both preparing to go public. OpenAI filed a confidential S-1 with the SEC on June 8 at a reported valuation of $852 billion, with Goldman Sachs and Morgan Stanley managing the offering. The company's annualized revenue has surged from $2 billion in late 2023 to roughly $25 billion by early 2026, though it continues to burn cash at a staggering rate — approximately $27 billion projected for 2026.
Anthropic is reportedly preparing its own S-1 for a potential IPO as early as October 2026, buoyed by locked-in compute deals that make its revenue more predictable. The company is valued at approximately $965 billion in private markets.
These dual IPOs would represent one of the most significant moments in tech market history — two companies built primarily on large language model technology, together valued at nearly $2 trillion, entering public markets within months of each other. For investors, the question isn't whether AI is big — it's whether these valuations can be sustained against cash burn rates that would make even the most ambitious startups blink.