🇮🇳 India's AI Summit: $350B in Commitments and One Very Awkward Photo

History was made in Delhi today. For the first time ever, every major AI CEO — Sam Altman, Dario Amodei, Sundar Pichai, Demis Hassabis — stood in the same room. Not in San Francisco. In India.

And when cameras came out for a photo with PM Modi... Altman and Amodei refused to hold hands. 🤝❌

Weeks after those Super Bowl ad wars, they couldn't manage basic professionalism. The people building superintelligence can't coordinate for 30 seconds.

But the real story is the money:

💰 Mukesh Ambani (Jio): ₹10 lakh crore (~$120B) in AI investment

⚡ Adani Group: $100B for "sovereign, green-energy-powered AI infrastructure" — part of a $250B ecosystem play

📊 Total summit commitments: Expected to exceed $100B (before counting the above)

Why India, why now?

🌏 First major AI summit in the Global South — this is strategic, not symbolic

👨‍💻 Every AI CEO needs India's developers and 1.4B potential users

📉 Sam Altman: AI costs dropped 1,000x in 14 months — "Global South benefits most"

The sovereignty angle:

India learned from its IT revolution — they captured services revenue while Microsoft & Google captured platform value. This time, they want the platform.

Jeet Adani: "This AI revolution gives India a once-in-a-century opportunity to change that equation."

The awkward truth:

The AI industry talks constantly about cooperation on safety and responsible development. Then they can't shake hands at a photo op.

While American companies squabble, India is building infrastructure to ensure it has options.

$350B in commitments. Every major CEO present. The hosts are thinking in decades, not quarters.

🔗 Full analysis: https://devdigestnow.com/blog/2026-02-19-india-ai-summit-trillion-dollar-handshake

⚖️ 20 Million Receipts: OpenAI Just Lost a Critical Legal Battle

A federal judge just ordered OpenAI to hand over 20 MILLION anonymized ChatGPT user logs to The New York Times and authors suing for copyright infringement. This is huge.

🔥 What Happened

• Judge rejected OpenAI's "privacy concerns" argument
• Full production ordered, not limited searches
• Plaintiffs want evidence that ChatGPT reproduces copyrighted content
• This is the biggest discovery ruling in AI copyright cases yet

💡 Why It Matters

The logs are a goldmine for proving the plaintiffs' core argument: that ChatGPT memorized and can reproduce copyrighted material. If they find systematic reproduction of NYT articles or book passages, OpenAI's "fair use" defense is toast.

🎯 What Developers Should Know

• Licensing costs incoming — If content creators win, AI companies pay up. API prices go up. Budget accordingly.

• Output liability unclear — If your product uses an AI that regurgitates copyrighted content, who's liable? This case will help define it.

• Open source matters — Clean training data provenance becomes a competitive advantage.

📊 The Bottom Line

The AI industry's "train first, litigate later" strategy is hitting reality. OpenAI built a $100B+ company assuming training on internet data is legal. Now 20 million pieces of evidence will test that assumption.

Settlement talks incoming? Probably. But the precedent being set here will reshape how every AI company thinks about training data.

The next year is going to be expensive for a lot of people in San Francisco.

👉 Full analysis: https://devdigestnow.com/blog/2026-02-20-openai-20m-user-logs-copyright/

#AI #OpenAI #Copyright #LegalTech #DevDigest

🎧 Spotify's Senior Devs Haven't Written Code in 2026

Spotify CEO Gustav Söderström dropped a bombshell this week: the company's most senior engineers—their best developers—haven't written a single line of code since December.

"They actually only generate code and supervise it," he told investors. Two months. Zero code. These are the people you'd hire to architect your systems.

The numbers are brutal:
• UC system CS enrollment dropped 6% (after -3% in 2024)
• Overall college enrollment is up 2% nationally
• Students are fleeing CS for dedicated AI programs
• UC San Diego's new AI major is the only UC program growing

Students aren't abandoning tech—they're abandoning coding. They're migrating to programs that teach them to work with AI.

But here's what the execs don't mention: AI fatigue is real. Engineer Siddhant Khare's viral essay described the new dev experience: "Every time it feels like you are a judge at an assembly line and that assembly line is never-ending, you just keep stamping those PRs."

We automated code writing to make devs more productive. We accidentally created a new job—code reviewer—that some find more exhausting.

What matters now:
• Architectural thinking over syntax mastery
• Code review expertise is critical
• Domain knowledge AI can't learn from training data
• Understanding which problems to solve

The identity crisis is coming. Developers built careers around writing code as a craft. Now "senior engineer" means... reviews more AI output? Writes better prompts?

Söderström's warning: "The things you build now may be useless in a month."

The assembly line is running. The question is which side of the review queue you want to be on.

🔗 Full analysis: https://devdigestnow.com/blog/2026-02-21-spotify-devs-stop-coding-ai-supervision/

💸 OpenAI's $600B Reality Check: When AI Dreams Meet Math

Sam Altman was talking $1.4 trillion in AI infrastructure. Now it's $600 billion. That's not a tweak—that's cutting ambitions by more than half.

What happened:
• OpenAI told investors: ~$600B compute spend through 2030
• Revenue projection: $280B by 2030 (2025: $13.1B actual)
• Currently burning $8B/year in cash
• ChatGPT: 900M weekly active users
• Codex: 1.5M+ weekly active users

The funding situation:
Closing a $100B+ round with Nvidia ($30B), SoftBank, and Amazon. Pre-money valuation: $730 billion.

Why this matters:
The AI infrastructure arms race may be cooling. Investors started asking uncomfortable questions about when "spend a trillion" becomes "make a profit." OpenAI answered by getting more realistic.

$600B is still insane money—more than most countries' GDP. But tying it to actual revenue projections ($280B by 2030) at least attempts to answer "why."

My take:
This is healthy. The breathless trillion-dollar commitments were disconnected from reality. When numbers get so big nobody can evaluate them, you're not doing financial planning—you're doing marketing.

If other AI companies follow with similar recalibrations, we'll know the industry is collectively hitting reset.

📖 Full analysis: https://devdigestnow.com/blog/2026-02-22-openai-600-billion-reality-check/

🔥 The Great Refounding: Tech's New Survival Playbook

Something strange is happening. Companies aren't "pivoting to AI"—they're refounding themselves entirely. Like the last decade was just a warm-up.

Airtable's Bet

• Was valued at $11.7B in 2021. Now ~$4B.
• Instead of hunkering down, founder Howie Liu is launching Superagent—a multi-agent AI platform he says could "eclipse Airtable itself"
• Hired David Azose (ex-OpenAI, led ChatGPT business products) as CTO
• Acquired DeepSky, an AI agent startup with $40M raised
• Still has $700M in the bank, "throwing off cash"
Opendoor's Resurrection

• Nearly faced delisting. CEO announces they're "refounding as a software and AI company"
• AI reduced underwriting time from hours to under 10 minutes
• 46% sequential growth in home purchases
• Missed EPS by 1,045%. Stock surged 10%+ anyway. Because the transformation metrics are working.
Why "Refounding" Matters
It's not a pivot. It's an admission that the 2019-2021 playbook is dead. The new rules:

1️⃣ Be profitable or have a clear path
2️⃣ AI-native means AI at the core, not a feature
3️⃣ Hire from the frontier (ex-OpenAI people are the new ex-Google)
4️⃣ Ship fast or die

The Uncomfortable Truth
Most companies saying "we're adding AI" aren't refounding—they're bolting features onto dying architectures. That's not transformation. That's duct tape.

True refounding means admitting your core product might become obsolete. That the thing you built for a decade might be a feature of someone else's AI within five years.

We'll see a lot more refounding announcements in the next 18 months. The question for everyone else: are you sure your product isn't just an AI feature waiting to be subsumed?

📖 Full analysis: https://devdigestnow.com/blog/2026-02-23-corporate-refounding-ai/

💸 The AI Credit Crunch: Why Lenders Are Getting Cold Feet on Software

Something interesting is happening in financial markets that most tech people haven't noticed yet. Software companies are suddenly finding it harder to borrow money.

The reason? Banks are scared of AI eating their lunch.

🔍 What's happening:
• Lenders are postponing debt deals with software companies
• Borrowing costs are rising as scrutiny tightens
• The question on everyone's mind: "What happens when AI replicates your features in weeks?"

📉 Why SaaS lost its halo:
For two decades, SaaS was bulletproof to lenders—recurring revenue, high margins, sticky customers. Now they're asking:

• What if switching costs drop to zero?
• What if an AI agent can migrate data effortlessly?
• What if "product breadth" gets collapsed by a foundation model?
💡 The new defensibility test:
Lenders now want to see:
• Proprietary data advantages
• Workflow lock-in AI can't break
• Regulated vertical positioning
• Distribution moats

⚡ The paradox: While traditional software struggles for capital, AI infrastructure companies are swimming in it. Meta just locked in millions of Nvidia chips. SK Hynix is ramping memory production.

The market is bifurcating: build AI infrastructure = capital flows freely. Build software AI might disrupt = the spigot is tightening.

For founders: AI defensibility isn't a nice-to-have slide anymore. It's a prerequisite for accessing capital markets.

📖 Full analysis: https://devdigestnow.com/blog/2026-02-24-ai-credit-crunch/

🔦 Taara Beam: Google's Moonshot Internet Uses Invisible Light

Forget fiber. Forget satellites. Alphabet spinoff Taara just unveiled something wild — a shoebox-sized device that beams 25Gbps internet through invisible light.

The Taara Beam specs:
• 25Gbps speeds (fiber-tier)
• 10km range, rooftop to rooftop
• Sub-100 microsecond latency (vs Starlink's 20-40ms)
• Deploys in hours, not months
• No trenching, no spectrum licensing

Why this matters:

The middle-mile problem has plagued internet infrastructure forever. Trenching fiber costs thousands per meter. Radio spectrum requires bureaucratic nightmares. Taara's pitch: mount, point, done.

T-Mobile and Airtel already deployed the older Lightbridge in 20+ countries. The new Beam is 50% smaller with faster speeds.

The smart positioning:

Taara is explicitly targeting autonomous vehicles — robotaxis and delivery vans offloading terabytes of lidar data at 25Gbps while charging. Plus V2X mesh networks for smart cities where sub-millisecond latency actually matters.

They're not competing with Starlink for rural broadband. They're creating a new category of urban infrastructure.

The catch: Weather sensitivity (fog, rain) is real, though their new Lightbridge Pro claims 99.999% uptime. We'll see.

Showcasing at MWC Barcelona next week. If they nail the partnerships, 2026 could be the year internet started literally beaming through the air.

🔗 Full analysis: https://devdigestnow.com/blog/2026-02-25-taara-beam-light-internet/

#infrastructure #google #connectivity #startups

🔥 DeepSeek Cuts Off Nvidia: The AI Cold War Enters a New Phase

Something unprecedented just happened in the AI industry. DeepSeek, the Chinese AI lab that rattled global markets with their efficient models, is withholding their upcoming V4 model from US chipmakers.

What's happening:
• DeepSeek gave Huawei and domestic Chinese chipmakers a multi-week head start on V4
• Nvidia and AMD were shut out—breaking standard industry practice
• This isn't just business; it's a strategic decoupling signal

The spicy part:
A US official claims DeepSeek actually trained on Nvidia's Blackwell chips (potentially violating export controls) while planning to publicly claim they used Huawei hardware.

Why it matters:
DeepSeek's models have been downloaded 75M+ times on Hugging Face. Downloads of Chinese models have now surpassed all other countries. The center of gravity in open-source AI is shifting—fast.

The bottom line:
DeepSeek isn't catching up. They're already here. And they're building a world where they don't need to ask American chipmakers for permission.

The AI cold war just got colder.

🔗 Full analysis: https://devdigestnow.com/blog/2026-02-26-deepseek-cuts-off-nvidia-china-ai-decoupling/

⚔️ Anthropic vs Pentagon: How Dario Amodei Turned an Ultimatum Into a PR Masterclass

The deadline is today at 5:01 PM Eastern. Defense Secretary Pete Hegseth has given Anthropic an ultimatum: remove Claude's safety guardrails—or face being blacklisted as a "supply chain risk."

The backstory:
Claude was the FIRST commercial AI in Pentagon's classified systems. $200M contract. Everything was fine until the Maduro capture operation in January, when reports emerged Claude was used during the mission.

Hegseth's three threats:

1. Cancel the $200M contract
2. Mark Anthropic as "supply chain risk" (the Huawei treatment)
3. Invoke the 1950 Defense Production Act to force compliance
The PR trap:

An anonymous Pentagon official told Axios: "The only reason we're still talking to them is that we need them. The problem for them is that they're that good."

That quote—meant to intimidate—became the headline. The Pentagon accidentally admitted it's dependent on Claude and can't easily switch.

Anthropic's countermove:

Dario Amodei's response is surgical. He opens with patriotic credentials, lists everything Anthropic already does for national security, then states the red lines:

"Mass domestic surveillance is incompatible with democratic values."


"Autonomous weapons are outside the bounds of what today's technology can safely do."


Who wants to defend "surveillance of Americans" or "robots that kill without human approval"?

The clock is ticking.

🔗 Full analysis: https://devdigestnow.com/blog/2026-02-27-anthropic-pentagon-standoff/

🚨 UPDATE: Trump Goes Nuclear on Anthropic

The deadline just passed. Trump's response? Full escalation.

What happened:

• ❌ ALL federal agencies ordered to immediately stop using Anthropic
• ⏳ Military gets 6-month transition period
• ⚠️ Threats of "serious civil and criminal consequences" if Anthropic doesn't cooperate
• 🚀 "Elon, your turn" — xAI just inherited the entire federal AI market
Trump on Truth Social:

"Leftist fanatics at Anthropic made a CATASTROPHIC MISTAKE trying to force the Department of Defense to comply with their terms of service instead of our Constitution."


What's still unclear:

• Supply chain risk designation (the real nuclear option for enterprise sales)
• Whether Defense Production Act will be invoked
Anthropic called Trump's bluff. Trump didn't blink—he escalated.

The AI safety movement just got its first real test. This is bigger than Apple's encryption fight. This is about whether AI companies can maintain any ethical boundaries when the government demands otherwise.

🔗 Full updated article: https://devdigestnow.com/blog/2026-02-27-anthropic-pentagon-standoff/

🔥 Trump Bans Anthropic — Then OpenAI Signs Deal With Same Restrictions

The irony: OpenAI got approved with identical AI weapons restrictions that got Anthropic blacklisted. The difference? They negotiated quietly.

🔗 https://devdigestnow.com/blog/2026-02-28-trump-bans-anthropic/

🧠⚡ Neuromorphic Computers Just Solved Physics Problems

Sandia National Labs just proved something that wasn't supposed to be possible: brain-inspired chips can solve partial differential equations (PDEs) — the mathematical backbone of every serious physics simulation.

Why this is huge:

• PDEs power everything from weather forecasting to nuclear simulations
• Traditional supercomputers need megawatts of power and entire rooms
• Neuromorphic hardware does it at a fraction of the energy cost

The key insight:

Researchers Brad Theilman and Brad Aimone developed an algorithm that lets neuromorphic systems handle rigorous mathematics — not just pattern recognition. The architecture mimics how neurons fire in the brain, processing information through massively parallel, event-driven circuits.

What it means for AI:

The current AI boom is hitting an energy wall. Training large models requires staggering amounts of power. Neuromorphic computing could be the escape hatch — handling not just pattern recognition but actual mathematical computation at dramatically lower energy costs.

The brain connection:

Humans perform "exascale-level" computations constantly (catching a ball, navigating crowds) on about 20 watts. Understanding how neuromorphic systems handle math could eventually inform treatments for neurological diseases like Alzheimer's and Parkinson's.

After years of promising potential, neuromorphic computing just delivered proof of concept.

🔗 Read the full analysis: https://devdigestnow.com/blog/2026-03-01-neuromorphic-computing-physics/

💰 OpenAI's $110B Mega-Round: What It Really Means

Last Friday, OpenAI closed the largest private funding round in history. The numbers are staggering:

• $110 billion total raised
• Amazon: $50B (their largest single investment ever)
• Nvidia: $30B (locking in their most important customer)
• SoftBank: $30B (Masa Son's redemption bet)
• Valuation: $730B pre-money

Why this matters:

This isn't about money—OpenAI was already cash-rich. This is about consolidation of power. The AI industry is splitting into two camps:

Camp 1: The OpenAI Alliance (Amazon + Nvidia + Microsoft + SoftBank)
Camp 2: Everyone scrambling to catch up

The gap is widening at alarming speed.

The skeptic's case:

• $730B at ~$4B revenue = 180x sales (that's not a valuation, it's a prayer)
• Three companies controlling AI's future is... uncomfortable
• Despite massive revenue, OpenAI still burns cash at alarming rates

What it means for builders:

1. Platform wars are over. Build on their platforms or build something they can't replicate
2. Pick your niche wisely—vertical-specific apps where domain expertise beats raw model capability
3. Watch open-source (Meta's Llama, Mistral) as your escape valve
4. Pricing pressure is coming. Plan accordingly.
Bottom line: The AI Wild West era is ending. Welcome to consolidation.

🔗 Full analysis: https://devdigestnow.com/blog/2026-03-02-openai-110b-funding-round

🔥 OpenAI Took the Pentagon Deal Anthropic Refused. Here's Why That Matters.

Last Friday, something unprecedented happened in AI. Anthropic refused to sign a Pentagon deal without explicit bans on mass domestic surveillance and autonomous weapons. The response? They got blacklisted as a "supply chain risk."

Hours later, OpenAI signed that very deal.

The key difference: Three words — "any lawful use."

OpenAI claims the same red lines Anthropic was defending. But their contract ties everything to existing legal frameworks — the same laws that enabled post-9/11 mass surveillance programs like PRISM.

Key points:
• Anthropic's CEO Dario Amodei wanted explicit contractual prohibitions
• Trump called them "radical leftists," administration labeled them supply chain risk
• OpenAI's deal relies on Executive Order 12333 — the Reagan-era directive that historically enabled bulk surveillance
• Former OpenAI policy head Miles Brundage: "OpenAI caved + framed it as not caving"

The darkest twist: The military reportedly used Claude in the Iran strikes HOURS after blacklisting Anthropic.

Market response: Claude overtook ChatGPT in the App Store over the weekend. Turns out standing on principle resonates with users.

This isn't just an AI ethics story — it's a preview of every future negotiation between AI companies and governments. The era of claiming neutrality while quietly enabling government overreach is over.

📖 Full analysis: https://devdigestnow.com/blog/2026-03-03-openai-pentagon-anthropic-ethics/

📱💻 MacBook Neo: Apple Just Put an iPhone Chip in a Laptop

Apple's been drip-feeding announcements all week, but today's leak is the real story. A regulatory filing accidentally revealed the MacBook Neo — a budget MacBook running the A18 Pro iPhone chip.

Why this matters:

• $599-799 price point — Chromebook territory
• A18 Pro chip — same silicon as iPhone 16 Pro, transplanted into a laptop
• 8GB RAM — half of what other MacBooks get
• 12.9-inch display — new size for the lineup

The trade-offs for that price:
❌ No True Tone display
❌ No keyboard backlighting
❌ No fast charging
❌ Storage maxes at 512GB

The bigger picture:

Apple is creating an entirely new product category. The MacBook Neo isn't a worse laptop — it's a better iPhone with a keyboard. For students, first-time buyers, and the "email and video calls" crowd, an iPhone chip is actually overkill.

This is Apple's Chromebook killer. Get people into the ecosystem cheap, upgrade them to Air and Pro later.

The A18 Pro's single-core performance already beats the original M1. It runs Apple Intelligence. For 90% of tasks that 90% of people do? Plenty.

The wall between iPhone and Mac silicon now has a crack in it. And cracks have a tendency to spread.

🔗 Full analysis: https://devdigestnow.com/blog/2026-03-04-macbook-neo-iphone-chip-laptop/

🍎 Apple's $599 MacBook Neo: The Chromebook Killer

Yesterday Apple did something unprecedented: released a laptop regular people can afford.

The MacBook Neo:
• Starts at $599 — same as iPhone 17e
• 5-core GPU, 16-core Neural Engine (iPhone-tier)
• Fanless, silent operation
• Available in 7 iMac colors
• Runs full macOS

Why this matters:

📚 Education market assault — Google's Chromebook dominance is under direct attack. At $599, schools now have a real alternative with full macOS ecosystem access.

🧠 AI democratization — That Neural Engine means Apple Intelligence comes to budget users. Writing assistance, smart Siri, on-device AI — all at entry-level pricing.

💰 Strategic shift — Apple is prioritizing market share over margins for the first time in decades. Get students hooked at 18, sell them MacBook Pros at 28.

The catch:
• 128GB base storage (cloud-first design)
• Not for development work
• This is a consumption device with creation capabilities

My take: The Chromebook killer isn't revolutionary hardware. It's Apple finally deciding to compete on price. And when Apple competes, they don't half-ass it.

Google had a nice decade in budget laptops. That decade is over.

📖 Full analysis: https://devdigestnow.com/blog/2026-03-05-apple-macbook-neo-599-chromebook-killer/

⚔️ Anthropic vs Pentagon: The $200M AI Ethics Showdown

Last week, Dario Amodei walked away from $200 million. The Anthropic CEO refused to sign a Pentagon contract that would give the military access to Claude for "any lawful use."

His concern? That phrase could cover domestic surveillance and autonomous weapons.

What happened next:

• Defense Secretary Hegseth threatened to blacklist Anthropic as a "supply chain risk"—a designation usually reserved for foreign adversaries like Huawei

• OpenAI swooped in within hours with its own Pentagon deal, with suspiciously convenient timing

• Emil Michael (DoD official) called Amodei a "liar" with a "God complex"

• Amodei fired back, calling the OpenAI deal "safety theater" and "straight up lies"

The plot twist: The public sided with Anthropic. Claude saw massive download surges while ChatGPT reportedly saw a 295% spike in uninstalls. Altman had to backtrack: "We shouldn't have rushed."

Now: Negotiations have resumed. The Pentagon actually needs Anthropic—Claude is already being used in classified operations in Iran. An abrupt switch would be "disruptive at best, dangerous at worst."

The real question: What are AI companies willing to accept for government contracts? Anthropic bet $200M that principles matter. Whether that bet pays off remains to be seen.

🔗 Full analysis: https://devdigestnow.com/blog/2026-03-06-anthropic-pentagon-ai-ethics-showdown/

🏦 SoftBank's $40B OpenAI Gamble Is Either Genius or Madness

Masayoshi Son is doing it again — and this time the stakes have never been higher.

SoftBank is seeking $40 billion loan (the largest dollar-denominated borrowing in company history) to double down on OpenAI, just days after the $110B funding round pushed OpenAI's valuation to $730 billion.

📊 The Numbers:
• SoftBank already holds ~11% of OpenAI
• 12-month bridge loan from JPMorgan and 3 other banks
• This follows Son's $20M → $60B Alibaba bet... and his $4.7B WeWork disaster

🐂 The Bull Case:
OpenAI isn't just another startup — it's becoming infrastructure. $10B+ annual revenue. Thousands of apps built on their API. If AGI is coming this decade, owning a chunk of the leading company could be worth any price.

🐻 The Bear Case:
Anthropic, Google, and Meta are catching up fast. OpenAI's governance has been messy. $730B valuation for a company that could face commoditization risk? We've seen this movie before (hello, WeWork).

💡 What This Means:
The AI infrastructure race is now a trillion-dollar contest. When conglomerates take out $40B loans for single bets, we've moved beyond "promising tech" to "strategic imperative."

Whether Son is a visionary or overleveraged, one thing's clear: he's not hedging. History will judge.

👉 Full analysis: https://devdigestnow.com/blog/2026-03-07-softbank-40b-openai-bet

🏛️ Anthropic Said No to the Pentagon. Now Claude is #1

Two weeks ago, Anthropic walked away from a Pentagon contract. They wanted safeguards against mass surveillance and autonomous weapons. The DoD said no. Now Claude is the most downloaded app in America.

What happened:

• Pentagon designated Anthropic a "supply-chain risk" after talks collapsed
• OpenAI swooped in with a $200M deal within hours
• ChatGPT uninstalls jumped 295% over one weekend
• Claude hit 1 million daily signups and became #1 on both app stores

The bigger picture:

Sam Altman admitted the OpenAI deal was "definitely rushed" and "looked opportunistic." They had to amend it within days after backlash. Meanwhile, Anthropic's principled stand turned into the biggest user acquisition event in AI history.

New government guidelines now require AI companies to grant "irrevocable licenses" for "any lawful use." Anthropic is contesting their blacklisting in court.

Why it matters:

This might be the first time in tech history that ethics directly translated into market dominance. Users voted with their downloads. If standing up to overreach gets you users, more companies will stand up.

The AI industry just learned that users are watching—and users have choices.

📖 Full analysis: https://devdigestnow.com/blog/2026-03-08-anthropics-pentagon-gamble/

🏦 Three Companies Just Ate 83% of All VC Funding

February 2026 broke records—$189 billion in global VC funding. But here's the kicker: three companies took home 83% of it.

The Big Three:
• OpenAI — $110B at $730B valuation
• Anthropic — $30B at $380B valuation
• Waymo — $16B at $126B valuation

Combined: $156 billion. That's one-third of ALL global VC spending in 2025. In one month. To three companies.

What everyone else got: $33 billion. Split across every other startup in every sector, globally.

Why this matters:

AI isn't just hot—it's become the only game in town. When 90% of venture capital flows to a single sector, and 83% of that goes to three players, we're watching capital concentration on a scale we've never seen.

The optimistic read: we're in a genuine paradigm shift, and smart money is betting big on the obvious winners.

The pessimistic read: this is FOMO at institutional scale, and valuations have completely detached from fundamentals.

Either way, the message is clear: VCs have made their choice. They're not diversifying anymore—they're going all-in on AI, betting that the winners will be worth trillions.

If they're right, $730B for OpenAI will look cheap.

If they're wrong... pension funds are going to have a very bad decade.

Read the full analysis 👉 https://devdigestnow.com/blog/2026-03-09-three-companies-ate-vc/

💰 Paid in Tokens: AI Compute Is the New Equity

Silicon Valley compensation is getting a fourth pillar: AI inference budgets.

What's happening:
• Engineers at OpenAI are already asking about dedicated inference compute in interviews
• Tomasz Tunguz (Theory Ventures): AI tokens are becoming compensation like salary, bonus, equity
• A $375k engineer + $100k inference budget = 21% of comp coming from AI access

The uncomfortable math:
An engineer with unlimited Codex access vs one without isn't 10% more productive — they're potentially 3-8x more productive. Same salary, wildly different output.

Why CFOs are sweating:
New metric emerging: productive work per dollar of inference. Tunguz automates 31 tasks/day for ~$12k/year. "The engineer still burning $100k? They'd better be 8x more productive!"

My take:
This creates a flywheel favoring incumbents. Big tech can offer massive inference budgets. Their engineers become more productive. Gap widens.

But it's also an opportunity for startups — can't compete on salary? Compete on AI compute. A $50k inference budget might be more attractive than a $20k raise for the right hire.

2026 might be the year we recognize: those tokens aren't API calls. They're your new equity.

🔗 Full analysis: https://devdigestnow.com/blog/2026-03-10-ai-compute-compensation