Five stories from the past 48 hours that every AI builder should know about. The revenue race just flipped, two companies are betting billions they can escape Nvidia’s gravity, one enterprise platform shipped a genuinely useful agent tool, and OpenAI quietly removed the usage ceiling that was throttling your work.
Anthropic Overtakes OpenAI in Revenue
The numbers are now public and they tell a clear story. Anthropic confirmed its annualized revenue crossed $47 billion, up from $9 billion at the end of 2025. OpenAI, by its own most recent disclosure, projects $25 billion to $33 billion in annualized revenue for 2026. That is not a close race anymore.
The engine behind the flip is Claude Code. What started as a developer tool hit $1 billion in annualized revenue by November 2025, then $2.5 billion by February 2026. It now sits at roughly $8 billion ARR, about 17% of Anthropic’s total. Enterprise customers including Netflix, Spotify, KPMG, and Salesforce are running it in production, not as a pilot.
Anthropic also filed confidentially for an IPO on June 1 at a $965 billion valuation after closing a $65 billion Series H. That puts it above OpenAI’s $852 billion valuation from March.
Builder verdict: matters. I run my content operation on Claude Code. The revenue numbers confirm what the builder community has been feeling for months: Anthropic’s product-led growth through developer tools is working. The API is the distribution channel, and the developers using it every day are the sales force. If you have been building on one provider and not testing the other, this is your signal to run a real comparison. The competitive pressure between these two is producing better tools faster. That benefits every builder, regardless of which one you pick.
Meta Puts Its Custom AI Chip Into Production
An internal Meta memo reported by Reuters revealed that Meta’s custom AI processor, code-named Iris, clears six weeks of testing and enters mass production in September. The chip is designed with Broadcom and manufactured by TSMC, and it is the centerpiece of Meta’s plan to double its total computing capacity to 14 gigawatts by 2027.
The capital commitment backing this is staggering. Meta’s projected capex for 2026 sits between $125 billion and $145 billion, most of it aimed at AI infrastructure. The company also committed $9.1 billion to build its first AI-focused data center in Alberta, Canada, its largest outside the United States.
Builder verdict: matters. Custom silicon is how you escape the Nvidia tax. Meta, Google, Amazon, and now Apple (on the device side) are all building their own chips because relying on one GPU supplier at current demand levels is a business risk. For builders, the practical impact arrives when Meta’s open-source models start getting optimized for Iris hardware and that optimization trickles into the inference APIs you actually call. More compute supply also means downward pressure on inference pricing over time. That is good for everyone running production workloads.
SambaNova Raises $1 Billion and Lands JPMorgan as an Inference Partner
AI chip startup SambaNova closed the first tranche of a $1 billion Series F at an $11 billion valuation, led by General Atlantic with participation from BlackRock, Intel Capital, Qatar Investment Authority, and Vista Equity Partners. The raise came just five months after its last mega round.
The bigger signal is the JPMorgan Chase deal. JPMorgan selected SambaNova’s SN40L and SN50 systems to power secure, on-premises AI inference at the bank.
Builder verdict: matters. JPMorgan is not a company that picks inference infrastructure for fun. Regulated industries (finance, healthcare, government) need on-prem AI that never sends data to a third-party cloud. SambaNova is positioning itself as the answer to that specific problem, and a $1 billion raise says the market agrees the problem is large. If you build for regulated verticals, on-prem inference options are expanding, and that changes your architecture decisions.
OpenAI Lifts the 5-Hour Usage Ceiling on Codex and ChatGPT
OpenAI’s Tibo Sottiaux confirmed on July 12 that the rolling 5-hour usage limit on Plus, Pro, and Business plans is temporarily removed. The change followed GPT-5.6 Sol demand doubling traffic in 48 hours after launch. OpenAI also pushed inference efficiency improvements to Sol that reduce how much usage each request consumes, effectively giving users roughly 10% more capacity.
Weekly limits still apply. This is a temporary measure, not a permanent policy change. But the combination of removed hourly caps and more efficient inference means you can run longer sessions without hitting the wall.
Builder verdict: breaks your stack (in a good way). If you have been rationing your Codex or ChatGPT Work usage around the 5-hour window, you can stop for now. The efficiency improvements to Sol are the more durable change. OpenAI is clearly motivated to keep Sol usage high during the post-launch window, so take advantage of it while the ceiling is gone. Test the workflows you have been putting off because of rate limits.
Alteryx Ships Agent Studio With MCP Integration
Alteryx unveiled Agent Studio at its Inspire 2026 conference, a tool that lets business analysts convert existing data workflows directly into autonomous agents. The feature includes an MCP Server that connects those agents to Slack, Salesforce, cloud data warehouses, and any other tool with an MCP connector, as well as to Claude and OpenAI models.
The pitch is that the bottleneck in enterprise AI is not model access anymore. It is business context: the logic, rules, and institutional knowledge that make model outputs actually useful inside a specific company.
Builder verdict: marketing, but with a real idea underneath. Most enterprise “agent” announcements are vapor. Alteryx has a real product angle here because they already sit on top of the data pipelines and business logic that agents need to be useful. Converting an existing, tested workflow into an agent is a much shorter path to production than building one from scratch. Agent Studio enters preview, not GA, so the proof is still ahead. But the MCP integration means it plugs into the same ecosystem Claude Code and other developer tools use. If you run Alteryx workflows today, this is worth watching.