On June 24, the big players made their move. ChatGPT creator OpenAI, in a strategic partnership with networking giant Broadcom, officially unveiled its first-ever custom silicon: the Jalapeño. This isn’t just a side project; it marks OpenAI’s official entry into the high-stakes game of designing its own AI chips.
Here’s the scoop: The Jalapeño is a dedicated Application-Specific Integrated Circuit (ASIC) built from the ground up for Large Language Model (LLM) inference. It’s the flagship product of a new multi-generation computing platform they’re building together. The goal? To make AI run faster, more reliably, and—crucially—cheaper for everyone. Plus, it’s incredibly flexible, meaning it can handle just about any major LLM you throw at it.
The team-up was a masterclass in speed. While OpenAI handled the architecture design and Broadcom took care of the manufacturing (tape-out) and network hardware, Canadian electronics manufacturer Celestica stepped in to handle the board and rack integration. From concept to working prototype, they knocked this out in a blistering nine months. By optimizing how data flows through the system, they’ve boosted inference efficiency and slashed energy consumption. Early engineering samples are already running machine learning tasks at mass-production frequencies and power levels, successfully powering through OpenAI’s February code model, GPT‑5.3‑Codex‑Spark. The plan? Full-scale deployment by the end of 2026, backed by gigawatt-scale data center clusters.
Unlike standard GPUs, which are general-purpose accelerators often adapted for specific tasks later, the Jalapeño is native. It wasn’t cobbled together from old AI accelerator tech; it was designed from scratch specifically to solve the unique demands of LLM inference.
Let’s be clear: in the world of AI chips, training and inference are two totally different beasts. Training chips scream for raw, brute-force compute power to churn through massive datasets quickly. Inference chips, however, live in the real world—they need to be energy-efficient, ultra-low latency, and cost-effective because that’s where users actually interact with the AI. Right now, the global market is dominated by Nvidia, whose CUDA ecosystem controls over 80% of AI software development. AMD, Intel, and Google are fighting for the remaining slices of the pie.
Analysts say OpenAI’s secret weapon here is its unmatched expertise in building massive AI models, especially its “World Models” capable of understanding and simulating physical reality. Currently, OpenAI relies heavily on Nvidia. Earlier this month, Nvidia’s CEO Jensen Huang noted that OpenAI would be among the first to deploy Nvidia’s latest Vera series CPUs in their own data centers, with full production ramping up in Q3.
But OpenAI isn’t slowing down. On June 22, they released a long-term vision to bring advanced AI to billions of people globally. CFO Sarah Friar revealed some jaw-dropping numbers back in mid-May: ChatGPT boasts over 900 million weekly active users, while their engineering tool, Codex, has crossed 4 million users.
Looking ahead, the roadmap is intense. OpenAI is doubling down on creating personal General AI (AGI) assistants ready to tackle daily chores, work projects, study sessions, and exploration. They’re also gearing up to launch the next-gen flagship, GPT-5.6. Market watchers expect this model to deliver a massive leap in coding, agent workflows, and 3D generation—all while undercutting competitors on price.
It gets even bigger. Back on June 7, reports citing the UK’s Financial Times suggested OpenAI is preparing its biggest upgrade since 2022. They want to transform ChatGPT into a “super app,” integrating programming tools with autonomous AI agents and adding revenue-generating features. A huge chunk of resources is going toward Codex. The message is clear: OpenAI is pivoting from simple chatbots to intelligent agents that can actually get things done for you.
Of course, this ambition comes with a steep price tag. Keeping these AI systems upgraded requires massive spending on inference costs. As a result, NUPIAO observes that OpenAI is pouring unprecedented capital into compute and AI R&D. According to recent disclosures, the company burned through $37 billion in cash during the first quarter alone, with revenue hitting around $57 billion—both figures tripling compared to last year. In 2025 total, they spent $340 billion, with roughly $190 billion going to R&D and nearly $60 billion on sales and marketing.
The scale is staggering. Co-founder Greg Brockman stated on May 5 that they plan to spend $500 billion on compute this year alone. Insiders whisper that by 2030, OpenAI could have racked up a total compute spend of $600 billion.
Despite the heavy spending, their coffers remain surprisingly deep. At the end of Q1, OpenAI held over $730 billion in cash and securities, up from $400 billion at the end of last December. However, the burn rate is real. They anticipate a net loss of roughly $140 billion for all of 2026, with cash burn accelerating further. Breaking even might not happen until 2030.
Faced with this pressure, OpenAI is moving fast on financing. They are aggressively preparing for an IPO. On June 8, CEO Sam Altman sent a Slack message to employees stating that OpenAI expects to go public “within the next year.” He admitted the timeline could shift, but emphasized that filing now gives them maximum flexibility.
Snap! Same day, they announced they had secretly filed an S-1 draft with the SEC. OpenAI clarified: “We haven’t decided on the exact timing… going public involves complex trade-offs. Filing now ensures we can hit the market faster when it truly benefits us.” Sources say they are working closely with investment banks like Goldman Sachs and Morgan Stanley, aiming for an autumn listing.
This follows a historic funding round in late March, where OpenAI raised a mind-blowing $122 billion—the largest in Silicon Valley history. Their post-money valuation skyrocketed to $852 billion. The deal was led by Amazon, Nvidia, and SoftBank, with Microsoft sticking around. Heavy hitters like Cathie Wood’s Ark Invest, BlackRock, Sequoia, Temasek, and Thrive Capital were all in the room. Strategic investors had already committed $110 billion as the base, including $50 billion from Amazon, $30 billion from Nvidia, and $30 billion from SoftBank.
Wall Street is predicting the IPO could value OpenAI at over $1 trillion, potentially raising $60 billion and becoming one of the largest tech IPOs in US history.
And they aren’t alone in the race. Arch-rival Anthropic is also quietly prepping for an IPO, though no date is set. Just a month ago, Anthropic closed an H-round funding of $65 billion, pushing their valuation to $965 billion—surpassing OpenAI’s $852 billion to become the world’s most valuable AI startup.
