OpenAI Heads Toward $70 Billion in Annual Revenue While Oracle Trucks Gas to Its Data Centers

OpenAI's annualized revenue is expected to reach $70 billion or more by the end of 2026. Ed Ludlow gave the figure on Bloomberg Tech. The same episode covered Oracle, which is keeping several data centers on schedule because natural gas for them is delivered by truck. Kleiner Perkins partner Ilya Fushman talked about AI mega-IPOs and physical AI.

The episode is short, but its three stories are connected. Revenue depends on compute, compute depends on electricity, and electricity is already being hauled in tankers.

What $70 Billion Actually Means

Run rate is the revenue of the most recent period scaled up to a full year. Annual financial statements will not show that sum for the year. It is a snapshot of the current pace: one strong month becomes a big annual figure. For a fast-growing company the metric is useful, but it overstates results if demand is seasonal or rests on one-off contracts.

For scale: at the end of 2025, Sam Altman put OpenAI's annual pace at just over $20 billion. Measured against that benchmark, this is growth of more than three times in a year. The Bloomberg piece gives no breakdown across ChatGPT subscriptions, the API and enterprise contracts, so it is too early to judge the structure of that growth.

Gas in Tankers Instead of a Substation

The second story shows what this growth costs. Oracle, one of OpenAI's key infrastructure partners, is keeping several data centers on schedule by getting natural gas delivered by truck. This is apparently a stopgap for the period before the sites are connected to the grid or to a pipeline. The announcement does not say which campuses are involved or what volumes are being delivered.

The logic is clear. A grid connection can take years, while a compute contract demands a launch on time. Trucked-in gas is more expensive and less reliable than a pipeline. But it removes the main risk: expensive hardware sitting idle while a substation is awaited. Electricity has become the industry's bottleneck, and it now shows up in logistics.

The IPO Queue and Physical AI

Ilya Fushman of Kleiner Perkins talks about AI mega-IPOs and the future of physical AI, meaning robots and systems that operate in the real world. The announcement gives none of his specific estimates, so I won't summarize his points. The link to the first two stories is direct, though. If large AI companies go public, public investors will check the stated run rate against actual revenue, margins and capital spending on data centers. The gas-truck story shows what that spending turns into in practice.

Key Facts From the Episode

• OpenAI: annualized revenue is expected to be $70 billion or more by the end of 2026
• Metric: run rate, not reported annual revenue
• Oracle: several data centers stay on schedule thanks to natural gas delivered by truck
• Guest: Ilya Fushman, partner at Kleiner Perkins
• Topics: the queue of AI mega-IPOs, physical AI
• Source: Bloomberg Tech, October 9, 2026

What to Watch Next

Investors should separate run rate from recognized revenue and ask companies how much of the annualized sum comes from recurring contracts. Teams building products on APIs should plan for capacity delays at providers: data center launch dates now depend on fuel and grids as much as on chips. Physical AI founders should watch how the market prices the IPOs of the big players, because valuations and access to capital for the next wave depend on it. The main practical question: how much does each dollar of this revenue cost in megawatts that have to be hauled in tankers?