OpenAI Speeds Toward IPO, Unveils Custom Chip, and Offers Americans a Stake
OpenAI is navigating a pivotal moment in July 2026, balancing breakneck expansion with mounting scrutiny. The company, valued at $852 billion after its March funding round, is reportedly delaying its IPO until it can reach a $1 trillion valuation. Meanwhile, its co-developed custom AI chip with Broadcom, named Jalapeño, has been unveiled, and CEO Sam Altman is in talks with President Trump to give the US government a 5% stake in OpenAI — a move that could hand American households roughly $320 in equity.
These developments come as rival Anthropic, valued at $965 billion in May, is slated to debut as early as October 2026. OpenAI is expected to follow in 2027, but questions about its token-payment business model are growing louder. Palantir CEO Alex Karp recently blasted the pay-per-use token consumption model, calling it indicative that 'something has gone completely wrong.' Enterprise customers are increasingly questioning the cost, and some are switching to cheaper open-weight models from Chinese startups like Z.ai.
Jalapeño Chip: A Custom AI Processor Built in Nine Months
Broadcom and OpenAI jointly announced the Jalapeño chip in late June 2026, an inference processor designed to run AI models for users rather than train them. The chip went from initial design to completion in about nine months, with OpenAI's models helping speed the engineering. The partners plan to deploy racks of OpenAI-designed chips starting late this year, building toward systems that would ultimately draw 10 gigawatts of power.
Jalapeño is an application-specific chip — less flexible than a graphics processing unit but cheaper to run for dedicated jobs. This design choice matters for the industry: every workload that moves to custom silicon is one that no longer needs a general-purpose chip, making custom accelerators arguably the fastest-growing niche in the industry. Broadcom now counts six core custom-chip customers, including OpenAI, Anthropic, Meta Platforms, and Google parent Alphabet.
Broadcom's Q2 semiconductor revenue from AI hit $10.8 billion, growing 143% year-over-year, and the company expects AI chip revenue to grow more than 200% year-over-year this quarter, to $16.0 billion. Despite this blistering growth, Broadcom shares trade about 24% below their 52-week high of $495, as of early July 2026.
Why It Matters: The Stakes for OpenAI and the AI Industry
OpenAI's simultaneous push toward a public listing, a custom chip partnership, and a government equity stake underscores the high stakes for the AI industry. The company is spending heavily on data centers and has not yet turned a profit. Its IPO delay until it can reach a $1 trillion valuation reflects investor caution amid the AI build-out's massive capital requirements.
At the same time, the broader AI market is facing headwinds. Enterprise customers like Uber, Microsoft, Salesforce, and Meta have taken steps to ration employees' use of advanced AI because the token payment structure preferred by Anthropic and OpenAI has proven more expensive than it's worth. The slang term 'tokenmaxxing' has emerged to describe engineers using AI models to excess under pressure to demonstrate integration, without clear guidelines.
The Token Payment Controversy
Palantir's Alex Karp told CNBC last week that the excessive use of AI without regard for whether it creates value is 'kind of like a porn addiction.' He urged the US AI industry not to dismiss the potential for cheaper open-weight models, especially those from China. Beijing startup Z.ai's GLM-5.2 model is now ranked among the top 10 large language models by Artificial Analysis and is ranked as the second-best model for web development by Code Arena, placing it alongside Anthropic, OpenAI, and Google. The open-weight model is also four to six times cheaper than frontier AI.
Z.ai has also launched ZCode, a coding assistant built around GLM-5.2 that works much like Claude Code or OpenAI's Codex. With a 1M-token context window, ZCode can handle multi-step programming tasks without losing context. Z.ai gives new customers a free five-day trial with up to 5 million tokens per day, and subscribers get about 1.5 times more quota through July 2026. Some US and international enterprise customers have already reported switching to cheaper Chinese models like DeepSeek and cutting back on payments to OpenAI and Anthropic.
OpenWeights and Cheaper Models: A Threat to OpenAI's Dominance?
The rise of cheaper open-weight models from China poses a direct challenge to OpenAI's pricing power. While OpenAI's frontier models remain among the best, the cost advantage of models like GLM-5.2 — which is four to six times cheaper — is driving adoption among cost-sensitive enterprises. A hands-on comparison by Snowflake across 103 tasks showed GLM-5.2 and Opus 4.7 nearly tied after three attempts, suggesting the performance gap is narrowing.
OpenAI and Anthropic must now convince investors that their premium pricing is justified by superior performance and reliability. The token-payment model, while lucrative, is increasingly seen as a double-edged sword: it drives high revenue per user but also encourages customers to seek alternatives.
Altman's Stake Plan: A $300 Dividend for Every American Household
Amid these commercial pressures, OpenAI CEO Sam Altman is pursuing a radical political strategy. The Financial Times reported last week that Altman is in talks with President Trump about giving the US government a 5% stake in OpenAI. The idea, which Altman first proposed in 2021, is that AI companies should share their wealth with the public as compensation for AI learning from human-generated work and to provide a safety net against potential labor market disruption.
If the government distributed this 5% stake directly to Americans, each of the roughly 133 million US households would receive about $320 in equity. Alternatively, the government could let the stake grow in a wealth fund and share a portion of the returns, potentially delivering a bigger payout if AI companies ever start turning a profit. Senator Bernie Sanders has proposed a more aggressive version: giving Americans a 50% stake in top AI companies.
For OpenAI, the stake plan could help swing public opinion. A majority of Americans don't trust companies to use AI responsibly, oppose construction of data centers in their area, and are more concerned than excited about AI's spread. Altman might hope the promise of payouts could ease these anxieties and build political support for AI expansion.
Broader Implications: What This Changes for Investors and Regulators
OpenAI's convergence of a custom chip deal, IPO preparations, and a government stake proposal signals that the company is thinking long-term — and that the AI industry's relationship with governments and the public is evolving. Broadcom's Jalapeño chip shows that custom silicon is becoming a key competitive advantage, potentially lowering costs for inference workloads and reshaping the hardware supply chain.
Meanwhile, the token payment controversy and the rise of cheaper open-weight models suggest that the high-margin business model of frontier AI labs may face pressure. If enterprise customers continue to migrate to cheaper alternatives, OpenAI and Anthropic may need to adjust their pricing or find new revenue streams.
The IPO Clock Is Ticking
Anthropic's expected October 2026 IPO will be a bellwether for investor appetite. If Anthropic's debut is well-received, it could pave the way for OpenAI's 2027 listing. But if the market judges the token model harshly, both companies may face valuation adjustments. The AI IPO wave is also attracting scrutiny from regulators, who are increasingly focused on market concentration, data privacy, and the societal impact of AI.
For a deeper look at how Meta is trying to compete with OpenAI's latest models, read: Meta's 'Watermelon' AI Matches OpenAI GPT-5.5, Claims Superintelligence Chief. Meanwhile, the geopolitical dimension — particularly the rivalry with Chinese AI companies — adds another layer of complexity, as seen in recent developments like Ukraine's long-range drone strike deep inside Russia.
Conclusion: A Fork in the Road
OpenAI stands at a crossroads. Its custom chip deal with Broadcom could lower inference costs and improve performance, helping it defend its premium pricing. Its government stake proposal could build political goodwill and provide a tangible benefit to Americans. But the rise of cheaper open-weight models, the token payment backlash, and the looming IPO scrutiny mean that 2026 and 2027 will be defining years.
Whether OpenAI can maintain its lead — and justify a $1 trillion valuation — will depend on its ability to innovate, manage costs, and navigate an increasingly complex regulatory and competitive landscape.
Comments