Get tomorrow's brief in your inbox
Today: Anthropic filed for IPO after hitting $47B revenue run-rate. OpenAI frontier models are now available on AWS, removing major enterprise adoption barriers. Nvidia launched RTX Spark AI agent PCs with Microsoft, Dell, and HP.
Anthropic has confidentially filed for an initial public offering, the company announced Monday. Valued at close to $1 trillion after raising $65 billion in its Series H round last week, the AI lab behind Claude is heading toward what could be one of the biggest tech IPOs in history.
The filing comes as Anthropic's revenue exploded to a $47 billion annual run-rate, up from $9 billion at the end of 2025. That kind of growth puts the company on a collision course with OpenAI, which is also preparing for its own IPO after raising $122 billion in March at an $852 billion valuation. Both companies are testing whether the public markets can absorb AI labs with unprecedented revenue growth but equally unprecedented capital requirements.
For small businesses, this matters more than it seems. The IPO race signals that both companies are betting on sustained enterprise demand. That means more stable pricing, better SLAs, and pressure to ship features that actually solve business problems instead of demo well. If you've been waiting for AI tools to mature before committing budget, this is the market telling you they're ready for production workloads.
Source: TechCrunch AI, The Verge AI
OpenAI models and Codex now available on AWS
OpenAI's frontier models and Codex are now generally available on Amazon Bedrock. If you're an AWS shop, this removes the biggest barrier to adoption: you can now access GPT-4.5 and Codex through your existing security, compliance, and billing workflows. No separate vendor relationship, no new procurement process, no extra security review. Codex on Bedrock brings the software engineering agent used by 5 million people weekly directly into AWS environments, including GovCloud regions. If your IT team has been blocking OpenAI access because it doesn't fit your cloud governance model, that excuse just evaporated.
Nvidia RTX Spark AI agent PCs launching this fall
Nvidia unveiled RTX Spark, a new PC CPU designed to run AI agents like OpenClaw or Hermes securely on-device. Microsoft, Dell, HP, Lenovo, ASUS, and MSI will ship RTX Spark PCs this fall, with Acer and Gigabyte following. The chip delivers 1 petaflop of compute and includes secure sandboxes (developed with Microsoft) to run agents safely. Microsoft is positioning its own Surface Laptop Ultra as the most powerful Surface ever built. Pricing hasn't been announced, but these systems appear to be full-fledged Windows versions of the $4,800 DGX Spark mini-computer Nvidia already sells to developers. If Nvidia cracked the code on making AI agents easy, safe, and useful for mainstream users, it could reshape the PC market. If you're a creator or run compute-heavy workflows, this could be a meaningful upgrade cycle.
Google Gemini Spark AI agent
Google launched Gemini Spark, a "24/7" AI agent that runs tasks in the background across multiple steps. Early access reviews show it's shockingly good at multi-step tasks like pulling data from Drive, drafting emails in Gmail, and creating calendar events. The Verge tested it with real-world workflows and found it mostly delivered on Google's I/O demos, though some tasks failed or produced incomplete results. The catch: Spark requires broad access to your Google account, raising privacy questions. If you're all-in on Google Workspace and trust their data handling, Spark could eliminate hours of admin work weekly. If you're privacy-conscious, wait for clearer data policies.
Codex expanding beyond code
OpenAI released a report showing Codex is no longer just a coding tool. It now has 5 million weekly active users, up 6x since the desktop app launched in February. Knowledge workers now represent 20% of users and are growing 3x faster than developers. They're using Codex to create reports, spreadsheets, presentations, contracts, and lightweight tools that previously required engineering support. The fastest-growing tasks are data analysis, research, and knowledge artifact creation. If you're a solo founder or small team, Codex can now do the work of a junior analyst, researcher, or ops person. The shift suggests AI's long-term impact isn't replacing jobs but expanding the scope of what one person can accomplish.
Small business tutor uses Notion AI as a virtual secretary
Sam Finnegan-Dehn, a London-based math and philosophy tutor, uses Notion AI to handle secretarial work across his tutoring business. He lets it record meetings with clients (with consent), generate automated summaries to refine his teaching strategy, and handle goal-setting, invoicing, and social media posts. For goal-setting, he writes down a North Star goal (like a target number of clients by year-end), then asks Notion AI to generate the concrete steps to get there. He reviews the results and picks which tasks to tackle first. The tool integrates with his existing Notion notebooks, where he already tracks client progress and readings. If you're a solopreneur juggling multiple hats, this is a practical template: use AI for rote admin work (invoicing, scheduling, note-taking) and keep the creative, high-value work for yourself.
HubSpot's reality check on AI outcomes
HubSpot published six contrarian takes on AI after three-and-a-half years of building and watching customers deploy it. Key insight: AI activity is not AI outcomes. Drafting emails and generating summaries are activities; they're not results. The companies winning with AI work backwards from a business problem, not forward from a model demo. HubSpot moved its Customer Agent and Prospecting Agent to outcome-based pricing in April because outcomes are what matter. Customers using Customer Agent respond to tickets 25% faster; those using Prospecting Agent generate 76% more leads. If you're evaluating AI tools, ask what outcome you're buying, not what activity the tool performs. HubSpot also warns that lowering the floor on generating code doesn't raise the ceiling on shipping value. You still need clean data, integration with existing apps, and a full customer view. The future belongs to companies that build AI into a coherent system, not those that pile up disconnected point tools.
VC firm interns use AI to review pitch decks and conduct market research
Tim Draper's VC firm is having interns develop better workflows with AI for reviewing pitch decks and conducting market research. The brief doesn't provide details on which tools or how, but the signal is clear: even investment analysis work is being augmented by AI. If your business involves research-heavy workflows (competitive analysis, market sizing, due diligence), your junior team members should be experimenting with AI tools to move faster.
Strava now charges $11.99/month for API access
Strava is restricting access to its API, requiring developers to pay $11.99/month to build apps using Strava's data. The company blames "zero-code AI tools" that allow users to quickly create apps that overwhelm APIs. Developer applications to Strava's program are up 448% year-to-date, and scraping attempts have degraded platform performance. If you're building on third-party APIs, expect more companies to follow Strava's lead and start charging for access or rate-limiting aggressively. Budget for API costs and build defensively.
Strava users can now link fitness data to Claude
On the flip side, Strava users can now use a new tool to link their fitness data (pace, heart rate, GPS data) directly to Claude. This integration won't be impacted by the new API restrictions. If you're building health or fitness tools, consider direct integrations with AI assistants rather than trying to build your own middleware.
Pre-ChatGPT startups getting crushed
Nearly half of the 857 U.S. unicorn startups haven't raised fresh funding in the last three years, making their billion-dollar valuations stale. Startups that last raised in 2021 are now worth 68% less on average; those from 2022 are down 52%. More than 220 companies that once held unicorn status are now "fallen unicorns," including Glossier, The Farmer's Dog, Brooklinen, Savage X Fenty, AG1, Betterment, and SeatGeek. Hit hardest are enterprise software companies, especially SaaS firms. The arrival of generative AI has redirected capital toward AI-native companies while making it impossible for older startups to justify their previous valuations. If you're evaluating software vendors, this creates negotiating leverage. Pre-AI companies are desperate for revenue and may offer steep discounts to keep you from switching to an AI-first alternative.
Alphabet raising $80B for AI infrastructure
Google parent company Alphabet plans to raise $80 billion by selling stock to fund its AI infrastructure buildout. Part of the plan includes selling $10 billion in stock to Berkshire Hathaway. The company says it's experiencing strong demand for AI solutions from enterprises and consumers at levels exceeding available supply. Alphabet expects to spend between $180 billion and $190 billion on capex this year. For small businesses, this signals that Google is betting big on sustained AI demand. Expect continued investment in tools like Gemini, Workspace AI features, and cloud AI services.
Florida sues OpenAI over safety concerns
Florida's attorney general sued OpenAI and CEO Sam Altman in a first-of-its-kind state-led lawsuit over ChatGPT's alleged links to violent incidents. The lawsuit accuses OpenAI of ignoring safety warnings to prioritize winning the AI arms race. This follows a criminal investigation into ChatGPT's potential role in a mass shooting at Florida State University and civil suits from families of victims. OpenAI denies responsibility, but the legal pressure is mounting. For businesses, this underscores the importance of human oversight when deploying AI tools in customer-facing or safety-critical contexts. Don't let AI make decisions without review.
Meta AI exploited to hijack Instagram accounts
Hackers exploited Meta's AI support chatbot to hijack Instagram accounts by simply asking it to switch the email associated with someone else's profile and reset the password. The issue, which Meta says has been patched, allowed hackers to take over high-value accounts (like single-letter usernames) just by asking the AI to link a new email address. Even security researcher Jane Manchun Wong had her account compromised. The attack worked because Meta wired its support system into an AI chatbot that had the ability to fast-forward through the entire account recovery process without proper verification. This is a stark reminder: don't wire support bots up to allow one-shot critical actions without multi-factor verification. If you're building AI-powered customer support, add human-in-the-loop controls for high-risk operations.
Source: The Verge AI, Simon Willison
AI weather startup WindBorne outperforms government agencies
WindBorne Systems released WeatherMesh-6, an AI weather forecasting model that's more accurate than the European Centre for Medium-Range Weather Forecasts (ECMWF), the gold standard for weather prediction. The model is "as accurate five days out as a traditional forecast is the day before," particularly on surface temperature. WeatherMesh-6 produces forecasts every hour (vs. every six hours for traditional models) at 3km resolution in Europe and the continental U.S. WindBorne benefits from collecting its own data via 400 balloons in flight at any given time, launched from 15 sites globally. The company sells weather data and forecasts to businesses. If your business is weather-sensitive (agriculture, logistics, construction, events), AI weather models now offer more frequent and accurate predictions than government agencies.
HPE stock surges 30% on AI server demand
Hewlett Packard Enterprise shares skyrocketed 30% after posting blockbuster Q2 results, with server revenue hitting $5.45 billion (vs. $4.66 billion expected). CEO Antonio Neri said traditional server bookings are up triple digits and it's the biggest backlog the company has ever seen. Industries focused on security are seeing rapid acceleration for AI on-premises instead of in the cloud. For SMBs considering where to run AI workloads, the trend suggests enterprises are prioritizing on-premises AI for security and control. If you're in a regulated industry (healthcare, finance, legal), expect vendors to pitch on-premises AI solutions aggressively.
Use prescriptive analytics to turn insights into action
Most businesses analyze what happened (descriptive analytics) or predict what might happen (predictive analytics) but stop short of prescriptive analytics, which tells you exactly what to do about it. Prescriptive analytics works backwards from a business problem and uses optimization techniques to recommend specific steps. If you're running predictive models but still making slow, inconsistent decisions, layer in prescriptive analytics. Start simple: if your model predicts which customers are likely to churn, build a decision tree that automatically triggers retention offers based on customer lifetime value and churn probability. Tools like Zapier Canvas can turn process maps directly into working automations.
Map your workflows before you automate them
If you're automating a messy process, you'll just get a faster mess. Use process mapping tools like Zapier Canvas, Lucidchart, or Miro to visualize workflows before automating. Zapier Canvas stands out because it converts visual maps directly into working automations. If your team spends hours in Slack arguing about handoffs, map the process first. You'll spot bottlenecks and redundant steps before you lock them into code.
The IPO race between Anthropic and OpenAI signals confidence that enterprise AI demand is sustainable, not a bubble. Expect both companies to prioritize stability and enterprise features over flashy demos. The AWS integration for OpenAI models removes friction for large organizations, which should accelerate adoption in regulated industries. Meanwhile, the collapse of pre-ChatGPT unicorns creates negotiating leverage for buyers and openings for AI-first startups. Watch for more API providers to start charging for access as zero-code AI tools create unsustainable demand on free tiers.