This week handed you three data points that tell the same story: the infrastructure layer is minting money at a pace that embarrasses every other business on earth, Google finally showed up to the agent fight with something real, and Cerebras proved Wall Street still has an insatiable appetite for any company that can credibly threaten Nvidia. Meanwhile, the Musk-Altman trial wrapped — and the real winner was everyone who ignored it and kept building.

Subscribe now

Nvidia Just Reported the Most Profitable Quarter in Semiconductor History

What happened: Nvidia reported Q1 revenue of $81.6 billion, up 85% from a year ago, with Data Center revenue of $75.2 billion — up 92% year-over-year — driven by the ramp of Blackwell 300 products and demand for InfiniBand, Spectrum-X Ethernet, and NVLink solutions. Jensen Huang called it “the largest infrastructure expansion in human history,” adding that agentic AI “has arrived, doing productive work, generating real value, and scaling rapidly across companies and industries.” Nvidia also announced an $80 billion share buyback and raised its quarterly dividend 25x — from $0.01 to $0.25 per share. LatestLY, Cybernews

Why it matters: No shipments of Data Center Hopper products to China occurred during the quarter, compared with $4.6 billion in the same period last year — and Nvidia still grew 85%. The China ban is priced in. The rest of the world is buying faster than China ever did. More importantly: hyperscalers represent about 50% of data center revenue, with the other 50% now coming from AI clouds, industrial, enterprise, and sovereign customers. This is the demand diversifying away from just five hyperscalers, which means the AI infrastructure build is no longer a Big Tech story — it is an economy-wide story. LatestLY

How to make money from this: The 50% of Nvidia’s revenue coming from non-hyperscalers is your market. Those are mid-size enterprises and sovereign AI programs that need help deploying Blackwell infrastructure — integration, optimization, cooling, security, and governance. None of that is glamorous. All of it will be paid. Also: Nvidia’s Vera Rubin platform ships H2 2026, and Huang said he expects to be supply-constrained through its entire lifespan. That means compute scarcity continues. Any SaaS product with efficient inference — smaller models, batched requests, optimized prompts — has a structural cost advantage over competitors burning tokens carelessly.

Share

Google I/O 2026: They Finally Built the Agent Layer

What happened: Google launched Gemini 3.5 Flash at I/O 2026 — the first model in its new series combining frontier intelligence with agentic action — claiming it outperforms Gemini 3.1 Pro on coding and agentic benchmarks at four times the output speed of competing frontier models. Google also announced WebMCP, a proposed open web standard that allows websites to expose structured tools like JavaScript functions and HTML forms directly to browser-based agents, and Chrome DevTools for agents — available now across more than 20 coding agents. Gemini Spark, an always-on background agent that runs across Google Workspace, connects to third-party apps, and will eventually interact with local files, was also unveiled alongside Project Astra moving from research demo to developer preview. Tech Startups + 2

Why it matters: WebMCP is the most underreported announcement of the week. If it becomes a standard — which Google has the Chrome market share to make happen — it means every website can natively expose its functionality to AI agents without scraping or brittle automation. That shifts the question from “how do I build an agent that can use websites” to “how do I make my website agent-native.” Every SaaS product that adds WebMCP support early becomes a first-class citizen in the agentic web. Every one that waits becomes a scraping target.

How to make money from this: Three concrete moves. First, implement WebMCP support on your product now — it is early enough that Google will actively promote compliant platforms. Second, Gemini 3.5 Flash is available in the API today at Flash-tier pricing with Pro-tier performance. Switch your inference stack and pocket the margin difference immediately. Third, Project Astra is in developer preview — build a workflow demo with it before the crowd arrives. Being in Google’s showcase materials for Astra integrations is worth more than any ad spend.

Leave a comment

Cerebras Pulled Off the Biggest Tech IPO Since Uber

What happened: Cerebras Systems priced its IPO at $185 per share on May 14, opened at $350 on Nasdaq — nearly double the offering price — and closed up 68% at $311.07, raising $5.55 billion in the largest U.S. tech IPO since Uber’s 2019 debut. The company has a $24.6 billion backlog, a multiyear agreement with OpenAI valued at more than $20 billion for 750 megawatts of AI compute capacity, and a binding term sheet with AWS to deploy its CS-3 systems inside Amazon data centers. bitcoinworld

Why it matters: Cerebras claims inference speeds up to 15 times faster than traditional GPU clusters. If that holds in production at scale, the economics of running AI workloads shift dramatically — lower latency, lower cost per token, new use cases that were previously too slow or expensive to be viable. The IPO also signals something bigger: the AI chip sector, measured by the VanEck Semiconductor ETF, has jumped 58% in 2026 alone. The public markets are now open for AI infrastructure companies. SpaceX, OpenAI, and Anthropic are all watching Cerebras’s post-IPO performance very carefully before setting their own price targets. TNW | Finance

How to make money from this: Cerebras’s speed advantage is most valuable in real-time, latency-sensitive AI applications — voice agents, live coding assistants, real-time translation, interactive AI tutors. If you are building in any of these verticals, Cerebras’s API access is worth testing now before everyone discovers it. Also: the IPO wave is starting. Three of the most anticipated AI listings — SpaceX, OpenAI, Anthropic — are still private. Build content, newsletters, communities, or analyst tools around the AI IPO cycle. Retail investor appetite for this sector is enormous, and the information gap is massive.

Share Ekofi Pulse

The Musk-Altman Trial Ended. Nobody Won. Everybody Lost Time.

What happened: The Elon Musk vs. OpenAI trial wrapped in Oakland after weeks of testimony, with Musk accusing Altman of abandoning OpenAI’s nonprofit mission and seeking around $150 billion in damages directed back to the nonprofit. The proceedings featured accusations of lies and selective amnesia from both sides, with Musk and Altman exchanging public barbs throughout — including Altman publicly inviting Musk to the GPT-5.5 launch party, saying “the world needs more love.” No verdict yet at time of writing. Uncoveralpha

Why it matters: The actual verdict matters less than what the trial revealed. Every internal email, every founding document, every conversation about mission drift is now public record. The picture it paints is of a company that started as a nonprofit research lab and became a $850 billion commercial enterprise in under a decade — and the governance structures never caught up. This creates real downstream risk: any future regulatory challenge to OpenAI’s structure now has a documented paper trail to work from. For Anthropic, this is a competitive gift. Every enterprise legal team evaluating AI vendors just saw exactly why mission-aligned governance documentation matters.

How to make money from this: AI governance consulting is about to have a moment. Boards and legal teams at enterprises adopting AI are going to demand clearer documentation of vendor ethics commitments, mission alignment, and structural safeguards — especially after watching OpenAI’s founding promises get dissected in open court. If you can package that as a service — AI vendor due diligence, governance frameworks, ethics auditing — you are in front of a real procurement need that is arriving right now.

Share Ekofi Pulse

Quick Hits

• Google launched a $2 million Build with Gemini XPRIZE Hackathon — the largest prize pool ever for a developer hackathon. Enter it. Not for the money, but for the Google visibility. Tech Startups

• Nvidia CFO confirmed zero chip revenue from China this quarter after the Trump-era export controls held firm — yet Nvidia still grew 85%. The China decoupling is complete and Nvidia doesn’t need them. Cybernews

• Google also unveiled “Intelligent Eyewear” glasses at I/O in partnership with Warby Parker and Gentle Monster — no display, just always-on Gemini Live audio. The AR race is now a fashion race. The winner will be whoever signs the right frames deal. Uncoveralpha

• Google’s migration agent in Android Studio can turn a React Native or iOS app into a native Kotlin Android app — tasks that previously took weeks now take hours. Mobile dev agencies: your re-platforming service line just got a major tool. Tech Startups

The Takeaway

Three things happened this week that most people will underreact to. Nvidia proved demand is still accelerating. Google proved it has the distribution to make agents real. And Cerebras proved the public markets will handsomely reward anything that plausibly challenges the GPU monopoly. Together they signal one thing clearly: the infrastructure war is not slowing down, it is compounding. The founders who treat this as background noise and keep building will look back at May 2026 as the week the second wave started.

Ekofi Weekly drops every Friday. Forward this to one founder still waiting for the market to “settle down” before they start building.

Share

Three years ago, “AI will replace developers” was a hot take. This week it became a quarterly earnings disclosure. Airbnb said 60% of its new code is written by AI. An OpenAI model out-diagnosed Harvard doctors in the ER. Ramp is gunning for a $40B valuation on the back of autonomous finance agents. The pattern is identical across all three: AI isn’t assisting humans anymore — it’s handling the workflow while humans supervise. That distinction is everything right now for anyone building a product or running a team.

Subscribe now

Airbnb Just Made Every Startup’s Engineering Argument Obsolete

What happened: During its Q1 2026 earnings call, Airbnb revealed that 60% of the code its engineers produced in the quarter was written by AI — roughly twice the industry average by the company’s own estimate — while its AI-powered customer support bot now resolves 40% of support issues without escalating to a human agent, up from 33% earlier in 2026. CEO Brian Chesky said: “Where you might have needed a team of 20 engineers before, an engineer can now spin up agents to do a lot of work under supervision.” TechCrunch, Artificial Intelligence News

Why it matters: Airbnb is not a startup experiment. It’s a $90B public company with hundreds of engineers. When a company at that scale says 60% of its code is AI-generated and 40% of its support is automated — and posts 18% revenue growth on the back of it — the “AI productivity” narrative graduates from a slide deck into a verifiable business outcome. Google says 30%. Microsoft says 30%. Airbnb says 60%. The ceiling is moving fast. Also buried in the call: Chesky explicitly said there is no room for “pure people managers” at Airbnb anymore. Every manager must be hands-on with the work. That is a massive organizational signal.

How to make money from this: The 20-engineers-to-1 ratio Chesky described is your business model. Build a “fractional AI engineering team” service: one senior developer running AI agents, delivering what a small agency would have needed five people to ship. Price it at $15–$25K/month, target funded startups that can’t afford a full team. The margin is extraordinary. Alternatively: Airbnb’s 40% bot resolution rate means there’s a productized service helping companies get from 0% to 40% bot resolution in 90 days. That implementation gap is real, billable, and wide open.

Leave a comment

An OpenAI Model Just Out-Diagnosed Harvard Doctors in the ER

What happened: A study published in Science by researchers at Harvard Medical School and Beth Israel Deaconess Medical Center found that OpenAI’s o1 model outperformed internal medicine physicians in emergency room diagnostic accuracy, correctly identifying the exact or near diagnosis in 67% of triage cases compared to 55% and 50% for the two attending physicians tested. The researchers stressed the models received identical information to what physicians saw in electronic health records, with no data preprocessing — and the AI’s advantage was most pronounced at the first triage touchpoint, where urgency is highest and information is scarcest. TechCrunch

Why it matters: The lead author said: “We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines.” Critics noted the study used non-specialist physicians as the comparison baseline and stopped well short of advocating clinical deployment, calling instead for formal prospective trials. But the direction is now undeniable. Healthcare AI just moved from “useful tool” to “beats the doctor in a peer-reviewed journal.” The liability and regulatory structure hasn’t caught up. That gap between clinical capability and legal deployment authorization is a multi-billion dollar problem waiting to be solved — and it will be. TechCrunch

How to make money from this: The near-term play isn’t “replace the doctor” — that’s five regulatory years away. The play is the layer between AI diagnostic capability and clinical workflow: second-opinion tooling, documentation automation, differential diagnosis surfacing for overworked ER staff. Build a narrow tool that puts AI diagnostic suggestions in front of a physician rather than bypassing them. You get the capability, you sidestep the liability, and you sell to hospital procurement. Also: medical coding and billing is a $20B industry where misdiagnosis costs hospitals money. An AI that’s 67% accurate on triage is probably 90%+ accurate on billing code selection. That’s a product you can sell Monday.

Share Ekofi Capital - Pulse

Ramp Is Worth $40B and It’s Basically an AI Agent Company Now

What happened: Ramp is in talks to raise $750 million at a pre-money valuation of more than $40 billion — a 25% jump from its $32 billion valuation just six months ago, and its fifth valuation step-up in fourteen months, which more than tripled the company’s worth across that stretch. Ramp crossed $1 billion in annual revenue, with autonomous AI agents now handling real-time fraud prevention, policy enforcement, cash-flow optimization, and the full procurement lifecycle — researching vendors, generating RFx materials, running compliance checks, and surfacing renewal intelligence. OpenAI

Why it matters: Ramp’s AI procurement agents report customers saving 16% on annual vendor spend and 46 fewer hours per month of manual purchasing work. Its accounting agent automates bookkeeping end-to-end, letting finance teams close books 3x faster on average, saving 40+ hours per month. This is not a fintech story anymore. Ramp is an AI agent operating system for corporate finance, with a corporate card attached. The card is the distribution. The agents are the moat. Meanwhile, its closest rival Brex was acquired by Capital One for $5.15 billion — a humiliating outcome against Ramp’s $40B. The lesson: the company that builds AI agents deepest into the workflow wins the category. Gamebezz

How to make money from this: Ramp’s partner and integration ecosystem is expanding fast — international markets, stablecoin payouts, embedded finance. There is a services layer around Ramp implementation, onboarding, and customization for mid-market companies that doesn’t exist yet at scale. Build it. Also, the “Ramp for [vertical]” opportunity is open: construction finance, healthcare billing, law firm expense management — each niche needs the same autonomous agent stack but with domain-specific rules baked in. Ramp won’t build all of those. You can.

Share

CB Insights Just Told You Exactly Which AI Startups Win

What happened: CB Insights released its AI 100 for 2026, identifying the clearest patterns in which AI startups are building durable businesses. Three patterns stand out: companies working with non-textual data build their own models and own the domain; companies in regulated industries like fintech embed so deeply into compliance workflows that switching becomes genuinely painful; and companies with access to rare proprietary datasets build compounding advantages that no amount of capital can easily replicate. Cybernews

Why it matters: Chai Discovery grew from a $150M to a $1.3B valuation in 15 months by training its own antibody design model on molecular data that general-purpose AI cannot natively represent. Salient, building AI for lending workflows, has 0% customer churn and 100% pilot conversion rate — pure proof that deep workflow embedding kills competition. The companies on the AI 100 building observability and evaluation tooling have collectively raised $278M in three years — early capital for what is effectively the operational rulebook for autonomous systems in enterprise. Cybernews

How to make money from this: The CB Insights framework is your product strategy filter. Ask three questions before building anything: Is my data non-textual or domain-specific enough that general LLMs fail on it? Am I embedding deep enough into a regulated workflow that switching costs are structurally painful? Do I have access to data that compounds as more customers use the product? If the answer to all three is no, you’re building a feature, not a company. If even one is yes, you have a defensible thesis. Start there.

Quick Hits

• Nectar Social emerged from stealth running more than 10 million autonomous brand conversations per week and attributing $100 million in revenue across those interactions, with customers including e.l.f. Beauty and Liquid Death. AI-native marketing ops is now a real revenue category, not a demo. Deadline

• Andrej Karpathy renamed “vibe coding” to “agentic engineering” — a deliberate reframe signaling that AI-assisted development needs more rigor, not less. If Karpathy is pushing for discipline, the cowboy phase of AI coding is ending. Deadline

• Iceotope raised fresh capital for precision liquid cooling systems, citing next-generation GPU rack densities moving toward 1MW — a level where air cooling breaks down entirely. The AI power problem just became a cooling problem. Infrastructure picks-and-shovels keep multiplying. Artificial Intelligence News

• Stitch raised a round led by Andreessen Horowitz — a16z’s first investment in the GCC — to build a cloud-native operating system for financial institutions, citing that banks cannot meaningfully adopt AI on top of fragmented legacy cores. Gulf fintech infrastructure is now globally contested ground. Artificial Intelligence News

The Takeaway

Airbnb, Harvard, and Ramp just handed you the same memo in three different languages: the human-to-AI ratio in every knowledge workflow is collapsing faster than anyone publicly admitted six months ago. 20 engineers to 1. Doctors losing to a language model at triage. Finance teams cutting 40 hours a month. This is not a future trend. It’s Q1 2026 earnings data. The founders who internalize this now — and rebuild their team structure, pricing model, and product around it — will look prescient in 18 months. The ones who wait for more proof will be the proof.

Ekofi Weekly drops every Friday. If this is the clearest signal you read all week, forward it to one founder who needs to hear it.

Share

This was the week OpenAI stopped pretending it’s only an AI lab. Ads went global. GPT-5.5 Instant shipped with a personalization layer tied to your Gmail. A $950M round for an AI customer service startup proved the vertical agent market is real money, not hype. And SpaceX is now literally powering Anthropic’s models. The infrastructure layer and the application layer are colliding — and founders who aren’t paying attention are about to get squeezed from both sides.

Subscribe now

ChatGPT Ads Just Went Global — and OpenAI Opened the Self-Serve Manager

What happened: OpenAI expanded its ChatGPT ads pilot to the U.K., Brazil, Japan, South Korea, and Mexico this week, while simultaneously opening up a self-serve Ads Manager to U.S. advertisers of all sizes, adding cost-per-click bidding and third-party measurement tools. Since the ads pilot launched in February, average monthly ad spend sits around $109 million — but OpenAI has forecast a $2.5 billion ad revenue target by end of 2026, making the gap between current traction and the target the real story. TechBriefly, Tech Startups

Why it matters: People use ChatGPT to work through medical decisions, relationship problems, and financial anxieties — conversations far more intimate than anything Google or Meta built an ad business on. That intimacy is a double-edged sword. The intent signal is extraordinary. The trust risk is real. But the self-serve manager opening is the bigger story: any business can now buy ChatGPT ads. This is Google AdWords circa 2002. The founders who figure out what converts here before the playbook gets written will have an enormous head start. Tech Startups

How to make money from this: Get into ChatGPT Ads Manager this week, not next quarter. The CPM and CPC rates right now are pre-competitive — before every agency piles in. Test narrow, high-intent queries in your niche. Users asking ChatGPT about a specific problem they’re trying to solve are warmer than any Google search lead you’ve ever bought. Also: if you run a content or SEO business, your model just broke. Shift budget immediately to “Generative Engine Optimization” — building topical authority so AI systems cite you as a primary source.

Leave a comment

GPT-5.5 Instant Ships — With Your Gmail Baked In

What happened: OpenAI released GPT-5.5 Instant as the new default ChatGPT model for all users, claiming 52.5% fewer hallucinated claims than the previous version on high-stakes prompts covering medicine, law, and finance, and 37.3% fewer inaccurate claims on the hardest conversations users had flagged for factual errors. Enhanced personalization from past chats, files, and connected Gmail is rolling out to Plus and Pro users now, with plans to expand to all tiers. CNBC

Why it matters: Gmail integration into the default model is not a feature announcement — it’s a data moat move. Once ChatGPT reads your inbox, your calendar, your files, and your chat history, switching costs become enormous. Every personalized response reinforces lock-in. OpenAI is quietly becoming the operating system for knowledge work. The hallucination reduction in law, medicine, and finance is equally important: those are the three verticals where AI has been blocked by liability concerns. That blocker just got meaningfully smaller.

How to make money from this: The Gmail integration creates a prompt engineering gap — most users will never configure it well. Build a “ChatGPT personal setup” productized service: $99 one-time, you configure someone’s memory, connected sources, and system preferences for their specific workflow. It’s trivial to deliver and has high perceived value. Also: the hallucination reduction in legal and medical contexts opens enterprise sales doors that were previously shut. If you’re building vertical AI tools in those sectors, update your pitch deck this week with the new accuracy benchmarks.

Share Ekofi Capital - Pulse

Sierra Hits $15.8B — The AI Customer Service Land Grab Is Real

What happened: Sierra, the AI customer service startup co-founded by OpenAI chairman Bret Taylor, raised $950 million in a Series E at a $15.8 billion post-money valuation, led by Tiger Global and Google’s GV. The company hit $150 million in annual recurring revenue in just eight quarters — a growth timeline described as unprecedented in traditional software. Taylor himself forecasts a “culling effect” in AI within two years, where capital dries up for all but market leaders. Abhs

Why it matters: Taylor said industries that have historically been slow to adopt tech are realizing “a watchful, waiting approach in AI is a path to extinction.” $150M ARR in eight quarters on top of a “constellation of models” — not a proprietary LLM — means the moat is workflow design, enterprise relationships, and trust. Not the model itself. That’s the template for every vertical AI play right now. The valuation at $15.8B on $150M ARR is a 105x revenue multiple. The market is still paying for category leadership regardless of margin. Abhs

How to make money from this: Sierra’s success is a map. Pick one vertical — healthcare admin, legal intake, real estate, logistics — and build the Sierra for that niche. You don’t need $950M. You need one painful workflow, one integration with the systems that vertical already uses, and five reference customers who’ll talk to prospects. At $150M ARR in eight quarters, Sierra was doing $18.75M per quarter before their Series E. A $2M ARR version of that in a tighter niche is a fundable, acquirable business.

Share

SAP Just Bought the “Anti-LLM” Enterprise AI Lab for €1 Billion

What happened: SAP is acquiring Prior Labs, an 18-month-old German AI research lab spun out of the University of Freiburg, committing €1 billion over four years. Prior Labs focuses on AI models for tabular — structured — data, aiming to improve machine learning for complex business datasets rather than building general-purpose language models. TechCrunch

Why it matters: SAP has 280 million users across 440,000 companies globally. When SAP acquires an AI lab and commits €1 billion to it, they are routing the AI transformation of European enterprise through that lab’s technology. Prior Labs’ focus on tabular data — the messy spreadsheets, ERP exports, and database outputs that run actual businesses — is not glamorous, but it’s where AI fails most in production. Solving that is worth more than another chatbot wrapper.

How to make money from this: Two plays. First, the “SAP AI implementation” niche is about to explode. SAP’s existing partner ecosystem is massive and mostly unprepared for AI integrations. Get certified, build a practice, and charge for implementation. Second: Prior Labs’ open-source releases will be worth watching. Early contributions to a lab backed by €1B with SAP’s distribution behind it gives you positioning in the next wave of enterprise AI tooling before the crowd arrives.

Quick Hits

• SpaceX signed a compute agreement with Anthropic, addressing Anthropic’s infrastructure strain after 80x growth in Q1 2026 that caused reliability issues for Claude Pro and Max users — with Elon Musk publicly praising Anthropic’s team afterward. Former rivals. Now compute partners. AI is stranger than fiction. OpenAI

• Fazeshift — AI accounts receivable automation — raised a seed round backed by F-Prime, Gradient Ventures, and Y Combinator. Boring finance automation is still getting funded. It should be, because it actually works. TechCrunch

• Cerebras is on track for an IPO targeting a $26.6 billion valuation — one of the first major public tests of whether the AI chip market has room beyond Nvidia. If it prices well, the AI infrastructure IPO wave officially starts. TechCrunch

• A new benchmark found that AI agents using visual web interaction consume 45 times more tokens than agents using APIs. The “human-like browser agent” demo is a cost trap. API-first agent architecture is not a style preference — it’s a business model decision. OpenAI

The Takeaway

OpenAI running ads, connecting to your Gmail, shipping a new default model, and expanding hardware — all in one week — is not a coincidence. It’s a deliberate race to own every layer before the IPO. The companies that will survive next to OpenAI are the ones that own a specific workflow so deeply that switching to ChatGPT native feels like too much friction. Depth beats breadth. Niche beats general. And if you’re still building a “general AI assistant,” you’re building OpenAI’s next feature.

Ekofi Weekly drops every Friday. Forward this to one founder who’s still A/B testing their landing page instead of watching what’s actually happening.

Share

Q1 2026 earnings week just happened and the verdict is brutal in its clarity: Google won, Meta got punished for spending too much, Microsoft is quietly squeezed, and Anthropic just decided it wants to be worth more than OpenAI. Meanwhile the China-US AI divorce is finalized. This is the most consequential week for the AI industry since ChatGPT launched — and most people are still talking about model benchmarks.

Subscribe now

Anthropic Is Racing to Be Worth $900 Billion Before Its IPO

What happened: Anthropic is asking investors to submit allocations for its latest fundraise within 48 hours, with the round expected to close within two weeks. The company is targeting a valuation of about $900 billion — and given soaring investor demand, the final figure may well exceed that. Anthropic’s annual revenue run rate recently surpassed $30 billion, up from about $9 billion at the end of 2025, with current estimates suggesting the run rate may be closer to $40 billion, driven largely by Claude Code and Cowork. TNW | Finance, CNBC

Why it matters: A $900 billion valuation for Anthropic would reset expectations overnight — signaling that capital is still willing to back category leaders at extreme levels. But it also raises the bar. A number that large puts pressure on the business to prove it can generate sustained profits, not just growth supported by funding. This is Anthropic’s last private round before going public. Everything from here is about proving the revenue story holds without the hype. Abhs

How to make money from this: The gap between Anthropic’s $40B revenue run rate and a $900B valuation is a 22x multiple — which means the market is pricing in years of dominance Anthropic hasn’t proven yet. That creates an arbitrage: enterprise buyers are still early on Claude adoption. Get in now as an implementation partner, build the Anthropic-native workflow tools, and your installed base becomes extremely valuable if and when Anthropic’s IPO forces enterprise buyers to standardize. Also: Anthropic’s push into finance, life sciences, and healthcare is explicitly stated. Those verticals need trusted, regulated-environment AI wrappers. Build one.

Big Tech Just Spent $725 Billion. Only Google Has Anything to Show For It.

What happened: Amazon, Microsoft, Meta, and Alphabet collectively posted earnings growth of about 60% compared to a year ago, and together they are planning to spend $725 billion this year on AI infrastructure — 77% more than the record $410 billion they spent last year. Google Cloud saw a meaningful acceleration in growth, with revenues increasing 63% to $20 billion — its strongest quarter ever. CEO Sundar Pichai said: “We are compute constrained in the near term. Our cloud revenue would have been higher if we were able to meet demand.” Meta beat on revenue but raised its full-year 2026 capex guidance to $125–$145 billion, sending its shares down 6% after hours. Fortune + 2

Why it matters: The market just told you something important. Google, which has the most vertically integrated AI stack — its own chips, its own models, its own cloud — is the only one being rewarded. Microsoft is capacity-constrained and burning cash. Meta is spending $145B and investors are spooked. The pattern is clear: vertical integration wins. Renting someone else’s GPU stack while building a general AI product is not a business model — it’s a bill.

How to make money from this: AWS Bedrock customer spend grew 170% quarter-over-quarter and processed more tokens in Q1 than in all prior years combined. That’s your signal. The enterprise AI workflow layer built on Bedrock is just beginning to be populated. Build a Bedrock-native tool for one specific enterprise workflow, price it on usage, and ride that 170% QoQ growth curve underneath you. Also: Google Cloud’s $462 billion backlog nearly doubled this quarter. A backlog that size means hundreds of enterprises just committed — and they all need implementation partners. Crescendo AI

Leave a comment

Microsoft and OpenAI Quietly Rewrote Their Relationship

What happened: On April 27, Microsoft and OpenAI restructured their partnership: Microsoft ends its Azure revenue share to OpenAI, while OpenAI keeps paying Microsoft a capped 20% through 2030. The “exclusive” that defined their relationship is effectively dead. TechCrunch

Why it matters: OpenAI is quietly decoupling from Microsoft before its IPO. It still owes Microsoft a 20% cut, but Microsoft no longer has the cloud lock-in it used to have. This signals OpenAI is actively courting other cloud providers — likely AWS and Google — to create competitive pressure ahead of going public. For the rest of the ecosystem, the era of any single cloud having an AI monopoly is over. OpenAI is now a genuine multi-cloud business.

How to make money from this: The decoupling creates a short window where OpenAI needs to establish new cloud relationships fast. If you’re building integrations, now is the time to build for Azure, AWS, and GCP simultaneously — not to pick a winner, but because your multi-cloud AI integration layer becomes indispensable when your customers’ AI vendor is actively switching compute providers. Charge for that neutrality.

OpenAI Is Building a Smartphone

What happened: Analyst Ming-Chi Kuo reported OpenAI may be building a smartphone, with MediaTek and Qualcomm developing the custom chip and Luxshare handling manufacturing. Rather than traditional apps, the device would use AI agents to complete tasks and maintain continuous context through a combination of on-device and cloud models. TechCrunch

Why it matters: OpenAI already bought Jony Ive’s devices studio for $6.4B. Now there’s a phone in the pipeline. If the app model dies on this device — replaced by persistent AI agents — then every app you’ve ever built is competing with a system prompt. This is not hypothetical. This is where the $122B raise is going. The question isn’t whether this device launches. It’s how fast developers need to think in agents rather than screens.

How to make money from this: Don’t build another app. Build agent-first experiences now — workflows that work via natural language with no UI, designed to run inside persistent context. The developers who figure out agent monetization in the next 12 months will own the equivalent of the “first App Store apps” moment when OpenAI’s device ships. Start with one workflow, one niche, zero screens.

Share Ekofi Capital - Pulse

Quick Hits

• The White House opposes Anthropic’s plan to expand access to Mythos, citing security concerns over a tool capable of carrying out cyberattacks. AI cyber capabilities are now a White House-level policy issue — regulation of offensive AI tools is coming fast. CNBC

• GPT-5.5 launches May 5 at 5:55 PM with a public event — Codex is picking attendees from replies. OpenAI now treats model launches like album drops. The distribution playbook is changing. CNBC

• Alphabet’s profit rose 81% as AI investments boosted revenue, moving its market value toward $4.5 trillion. Google is the only Big Tech company the market fully trusts right now. That gap matters for where enterprise budgets flow next quarter. CNBC

• Amazon launched “Join the Chat” — an AI-powered conversational audio Q&A directly on product pages, building on prior conversational commerce features. Amazon just made every e-commerce competitor’s product page feel three years old. TechCrunch

The Takeaway

The companies winning this week aren’t the ones with the smartest models. They’re the ones with the most vertically integrated stacks — chips, models, cloud, distribution, all in one loop. Google owns that loop. Anthropic is sprinting to build it. OpenAI is buying its way into hardware to complete it. If you’re a SaaS founder still renting every layer of the stack, the next 18 months will tell you very clearly whether that’s a business or a feature someone else will absorb.

Share

Ekofi Weekly drops every Friday. If this helped you think clearer, forward it to one founder who needs signal, not noise.

Everyone’s still arguing about which model is smarter. Meanwhile, the companies quietly building the pipes, the storage, and the workflow layer are being valued like they own the internet. This week made that abundantly clear. Three moves — one massive deal, one product power grab, and one funding round that rewrites what “infrastructure” is worth — tell you exactly where the next decade of money is going.

Subscribe now

Amazon Just Locked In Its AI Future for $33 Billion

What happened: Amazon committed $5 billion immediately to Anthropic, with up to $20 billion more tied to performance milestones, bringing its total commitment to over $33 billion. In return, Anthropic pledged to spend over $100 billion over ten years on AWS infrastructure, chips, and tools to train and deploy Claude models. Tech Startups

Why it matters: This is not an investment — it’s a decade-long revenue guarantee dressed up as a partnership. Amazon secures compute demand that rivals any hyperscaler contract in history. Anthropic gets capital and infrastructure without giving up control. But the deeper signal is this: the frontier model race is now inseparable from the cloud wars. If you’re building on Claude or AWS, you’re inside a fortress. If you’re outside it, you’re competing against a $33B moat.

How to make money from this: The Anthropic-AWS lock-in creates a services gap — enterprises need help migrating, optimizing, and governing Claude deployments on AWS. That’s a consulting and tooling play with zero competition right now. Build Claude-native internal tools that run exclusively on AWS Bedrock and charge for the integration work. Alternatively, watch for Anthropic’s partner ecosystem to open up — being early in a partner directory of a $33B relationship has compounding value.

Share Ekofi Capital - Pulse

OpenAI Wants to Own Your Entire Workday

What happened: OpenAI shipped a major Codex update aimed at its three million weekly developer users, adding the ability to operate your computer alongside you, work across tools and apps, generate images, remember preferences, learn from previous actions, and take on recurring scheduled work. AI Marketers At launch, over ninety plugins shipped, including Jira, the full Microsoft 365 suite, Notion, Slack, HubSpot, Salesforce, GitHub, and Zendesk. AI Business Magazine

Why it matters: OpenAI is not competing on code quality alone — they are competing on workflow ownership. TechCrunch A tool that controls your Mac, remembers your preferences, and handles recurring tasks is not a coding assistant. It’s an operating layer. The plugin count matters less than the intent: if Codex becomes the thing that connects your Slack to your Jira to your GitHub automatically, switching costs become enormous fast.

How to make money from this: Build vertical-specific Codex workflow packages — “Codex for agency project management” or “Codex for SaaS onboarding ops.” Sell them as productized setups with preconfigured plugins, prompts, and routines for $500–$2,000 one-time. Also: every enterprise that deployed ChatGPT Enterprise in 2024 now needs a Codex upgrade assessment. That’s a paid audit service you can offer immediately with zero technical overhead.

A $30 Billion Storage Company You’ve Never Heard Of

What happened: Vast Data raised approximately $1 billion in a Series F round at a $30 billion valuation, driven by growing demand for infrastructure to support AI workloads. IMFOUNDER Revenue has been roughly tripling year over year, the company generates over $100 million in cash per quarter, and it is free cash flow positive — unusual for a company at this growth rate. opentools

Why it matters: The AI revolution of 2026 is no longer defined by the size of LLMs or raw GPU quantities — the industry has hit a “data wall,” where the ability to store, retrieve, and process petabytes of information at the speed of light has become the ultimate competitive advantage. Spicy Advisory Vast Data is profitable, cash-generative, and just got valued at $30B. That’s the data infrastructure layer pricing itself like a platform. Every boring layer of the AI stack is repricing upward and nobody is talking about it.

How to make money from this: The “data wall” problem exists at every scale — not just hyperscalers. SMBs and mid-market companies running AI workflows have messy, slow, expensive data pipelines. Build a productized data audit and remediation service targeting companies spending over $10K/month on AI API calls. Their bottleneck is almost never the model. It’s the data plumbing. Charge to fix it.

Share

OpenAI Builds a Drug Discovery Model

What happened: OpenAI released GPT-Rosalind, a model built specifically for pharmaceutical drug discovery, currently available exclusively to US-based pharma companies. It’s designed to accelerate the identification and development of drug candidates — a process that traditionally takes years and costs billions. TechCrunch

Why it matters: Vertical-specific frontier models are now real products, not demos. When OpenAI names a model after Rosalind Franklin and ships it exclusively to one industry, it signals a deliberate strategy to own entire verticals, not just provide general-purpose intelligence. Every major industry is now a potential target.

How to make money from this: You don’t need to be in pharma to benefit. The pattern — specialized model plus exclusive early access plus premium pricing — is the playbook. Find a vertical where general-purpose LLMs underperform (legal discovery, construction estimation, agricultural compliance), build a fine-tuned wrapper, and charge 10x the API cost for the domain expertise baked in.

Leave a comment

Quick Hits

• NVIDIA gave 10,000+ employees early access to GPT-5.5 via Codex rss — when the GPU company validates your software product internally, it’s not a testimonial. It’s a distribution channel.

• AcuityMD raised $80M at a $955M valuation for AI-powered medtech sales Tech Startups — vertical AI CRMs are quietly becoming unicorns. Pick your niche.

• Vast Data’s valuation tripled from $9.1B to $30B in under three years OpenAI — infrastructure multiples used to be boring. Not anymore.

• 10x Science raised $4.8M from YC and Initialized to solve the AI drug candidate bottleneck TechCrunch — the picks-and-shovels play inside AI drug discovery is already being funded at seed.

The Takeaway

The model layer is commoditizing faster than anyone wants to admit. The real value is accumulating in three places: the data infrastructure underneath, the workflow layer on top, and the vertical-specific expertise baked into both. If you’re building a “general AI tool,” you’re already late. If you’re building the thing that makes AI actually work inside one specific industry — you’re exactly on time.

Ekofi Weekly drops every Friday. Forward this to someone still debating which LLM to use instead of building something.

The week’s story isn’t any single product launch. It’s the architecture of what’s being built and who’s quietly buying the picks-and-shovels while everyone else argues about chatbots. VC money is moving at a pace that makes 2021 look cautious. That should excite you — and make you nervous in equal measure.

OpenAI Just Showed You the Acquisition Playbook

What happened: OpenAI acquired personal finance startup Hiro Finance in what amounts to an acqui-hire — the product is shutting down April 20, all user data deleted by May 13, and the founding team walks straight into OpenAI. TechCrunch No deal terms disclosed. Backed by Ribbit, General Catalyst, and Restive, and it still ended as a talent grab.

Why it matters: This is OpenAI systematically pulling fintech domain expertise in-house ahead of a bigger personal finance play. They are not building a chatbot. They are building a financial operating system for consumers. Every fintech niche is now a potential acqui-hire target.

How to make money from this: Build a narrow, high-quality AI fintech tool with a real user base — budget tracking, tax prep, subscription auditing — and position for acquisition, not scale. Alternatively, build the integration layer that connects OpenAI’s coming finance product to existing banking APIs before they do it themselves. Or launch a “migration service” for Hiro’s now-stranded users before May 13. They need somewhere to go and they need it fast.

Perplexity Ships a Mac Agent That Never Logs Off

What happened: Perplexity dropped a Mac agent that lives on your machine full-time, working across local files, native apps, connectors, and the web in one orchestrated system. On a Mac mini, it stays on around the clock. AI Marketers Waitlist rollout is expanding now.

Why it matters: This is the first real challenge to the “chat when you need it” model. A persistent agent that knows your local files, your calendar, your open tabs — that is a fundamentally different product category than a chatbot. The shift from on-demand to always-on AI changes what users expect from every tool they pay for.

How to make money from this: Build niche Mac agent workflows as productized services — “your AI ops manager for solo founders” or “always-on client monitoring for agency owners.” The barrier right now is prompt engineering and workflow design, not code. First movers who package this well will charge $200–$500/month before the market catches up. Also worth watching: any SaaS that doesn’t offer a native agent integration in the next 12 months will feel legacy.

Subscribe now

$300 Billion in Q1 — The Bubble Nobody Wants to Name

What happened: Investors poured $300 billion into roughly 6,000 startups globally in Q1 2026, up over 150% quarter over quarter and year over year — an all-time high for venture investment by a wide margin. Crunchbase News

Why it matters: Most of that capital is chasing AI compute and frontier labs. That means the mid-market — your $5K–$50K/month SaaS buyers — is relatively underserved right now. Big capital goes to big bets. Boring profitable AI tools are available for the taking.

How to make money from this: Position as the “anti-hype” AI vendor. While everyone raises $50M to build another AI platform, sell a $299/month tool that actually works on one specific workflow. Buyers are fatigued by promises. Also: this funding wave creates M&A fodder. Well-funded companies acquire to ship faster. If you’re building in a niche they care about, you don’t need to scale — you need to be findable.

Share

A Shoe Brand Just Pivoted to AI Compute

What happened: Allbirds announced it is abandoning footwear and pivoting to AI compute infrastructure, rebranding as NewBird AI with a $50 million agreement to fund the new venture. NBC News

Why it matters: When a shoe company raises $50M to become an AI infrastructure business, the signal isn’t that compute is a good idea — it’s that public market investors still reward the word “AI” on a press release. This is the 2000 playbook with better branding.

How to make money from this: Watch for more zombie brands doing the same pivot. They’ll need AI strategy consulting, rebranding, and technical due diligence. There’s a short-term services play here. Longer term: if you’re building real AI infra tooling, this noise actually helps you — it lowers the bar for what buyers think they’re getting, and you can over-deliver cheaply.

Share Ekofi Capital - Pulse

Quick Hits

• Anthropic shipped Claude Opus 4.7 with the ability to run complex, multi-hour projects without supervision AI Marketers — autonomous agents just got a real upgrade, not a demo.

• South Korea’s DeepX is moving toward an IPO, building low-power AI chips for on-device inference Tech Startups — the edge AI hardware race outside Nvidia is real and it is fundable.

• K1x raised $175M to automate tax compliance for private markets Tech Startups — unsexy, deeply vertical, and exactly the kind of company that quietly becomes infrastructure.

• 108 malicious Chrome extensions were exposed this week Tech Startups — browser security for AI-augmented workflows is an underfunded problem with obvious SaaS upside.

The Takeaway

Everyone is chasing the frontier. The money is in the mile-wide, inch-deep problems that frontier labs will never bother solving. Build boring. Build specific. Charge monthly.

Ekofi Weekly goes out every Friday. Forward this to a founder who needs fewer hot takes and more angles.