Home Blog Page 3

DeepSeek V4 Runs on Huawei Chips. Jensen Huang Said This Would Happen.

By
Rohit Yadav
-
DeepSeek and Huawei logos on dark circuit board chips with Huawei Ascend AI processor in the background, representing DeepSeek V4 running on Chinese domestic hardware.

On April 24, 2026, Chinese AI lab DeepSeek released V4, its most capable model to date, and handed exclusive early optimization access to Huawei and other Chinese chipmakers. NVIDIA and AMD were shut out. For anyone tracking the US-China AI race, this is the moment the export control strategy began to crack in public.

What DeepSeek V4 Actually Is

DeepSeek V4 launches in two variants: V4-Pro, a 1.6 trillion-parameter Mixture-of-Experts model with 49 billion active parameters per token, and V4-Flash, a leaner 284 billion-parameter version built for speed and cost efficiency. Both support a one million token context window. V4-Pro is priced at $1.74 per million input tokens and $3.48 per million output tokens. OpenAI’s GPT-5.5 costs $5 input and $30 output. That is roughly a 10x pricing gap at the frontier.

Both models are open-source, available on Hugging Face and through the DeepSeek API. Developers can download the weights and run them locally.

The Hardware Signal Is the Real Story

DeepSeek V4 is the first frontier-class model built with deep optimization for Huawei’s Ascend 950 chips. When V4 was in development, DeepSeek gave Chinese chipmakers early access: the kind of pre-release collaboration that allows hardware teams to optimize drivers, compilers, and inference stacks ahead of launch. Nvidia and AMD did not receive that access.

On launch day, Huawei confirmed its Ascend 950 supernode infrastructure provides full support for DeepSeek V4. Shares of SMIC, the Chinese foundry that manufactures Huawei’s Ascend chips, jumped 10% in Hong Kong trading.

DeepSeek expects to lower V4-Pro API prices further once Huawei scales Ascend 950 production in the second half of 2026. Cheaper Chinese chips will mean cheaper Chinese AI inference. The trajectory is clear.

Jensen Huang’s Warning, Now Materializing

At NVIDIA’s GTC conference in March 2026, Jensen Huang made his position explicit: “There’s no question we need to have American tech stack in China.” His reasoning has been consistent for years. Pushing China outside the American hardware ecosystem does not eliminate Chinese AI capability. It accelerates the development of an alternative ecosystem.

In a mid-April podcast with Dwarkesh Patel, Huang debated the national security implications of chip exports directly. Critics argue his position is self-serving, given NVIDIA’s significant commercial interest in the Chinese market. That criticism is fair. But the underlying argument is also proving out in real time. China did not slow down. It built differently.

What Export Controls Actually Did

The US bet was that restricting access to advanced Nvidia GPUs would limit China’s AI compute ceiling. DeepSeek has now published three consecutive model generations, V3, R1, and V4, each competitive at the frontier, each developed under those restrictions. The constraint that was supposed to create a gap instead created pressure that drove efficiency innovation.

DeepSeek also faces a separate set of accusations. Anthropic and OpenAI have both accused the company of conducting industrial-scale distillation attacks, training their models on outputs from US frontier models to extract capabilities. China’s foreign ministry called the claims “groundless.” Those accusations remain unresolved and contested, adding a further layer to what is already a deeply adversarial technology relationship.

The Stakes

If Huawei’s Ascend roadmap delivers, with the 960 and 970 chips targeting roughly double the performance gains over each generation, China could have a fully sovereign AI infrastructure stack within two to three years. Frontier-class models, trained and deployed on domestic chips, priced at a fraction of US alternatives, distributed as open weights to the world.

That is not a hypothetical. With DeepSeek V4, it is already partially true.

Washington used chip controls as its primary tool in the AI race. That tool just got noticeably less sharp.

Advertisement

Meta Laid Off 8,000 Employees While Doubling Its AI Budget to $135 Billion

By
Rohit Yadav
-
A lone worker sits at a glowing desk in a vast, empty tech office at night, surrounded by rows of vacant workstations, with AI data visualizations in electric blue and neon green illuminating the dark.

On April 23, 2026, Meta sent a memo to employees informing them that 8,000 of them, roughly 10% of the company’s global workforce, would be let go effective May 20. The same memo made no apology for the timing. Meta’s capital expenditure on AI infrastructure is projected to hit at least $115 billion in 2026, up from $72 billion last year. Some estimates put the figure closer to $135 billion when talent and acquisition costs are included.

That combination, mass Meta layoffs 2026 paired with record AI investment, has become the defining image of where the tech industry stands right now.

The Productivity Argument

Meta is not making an unusual argument. It is making the standard one. Chief People Officer Janelle Gale wrote in the internal memo that the cuts are part of “our continued effort to run the company more efficiently and to allow us to offset the other investments we are making.” Mark Zuckerberg said on Meta’s January earnings call that 2026 is “the year that AI starts to dramatically change the way that we work,” and that “projects that used to require big teams can now be accomplished by a single very talented person.”

This view has prominent supporters across Silicon Valley. Garry Tan, CEO of Y Combinator, has publicly documented shipping 37,000 lines of code per day using AI agents and noted that a quarter of current YC startups are writing 95% AI-generated code. His direct message to founders: you no longer need a team of 50 or 100 engineers. The capital goes further. The headcount goes down.

The productivity gains Tan describes are real and measurable. AI coding tools are documented to produce 40 to 55% more output per developer per sprint. Paul Graham has written about founders he has met who now write 10,000 lines of code per day with AI assistance, calling it a qualitative shift in what a small team can accomplish.

Also Read: Google Now Generates 75% of Its Code With AI

Meta Layoffs 2026 are Not an Isolated Event

Meta cutting 8,000 jobs is the largest single announcement, but it sits inside a broader pattern. Block cut close to 40% of its workforce this year, citing AI-enabled flat team structures. Atlassian reduced headcount by 10%, explicitly to redirect budget toward AI product development. Amazon announced 16,000 job reductions. Snap cut 16% of its staff, noting that AI now generates over 65% of its new code. Across the tech sector, more than 78,000 workers were laid off in Q1 2026 alone, with nearly half of those cuts attributed to AI and workflow automation.

The messaging is consistent: AI raises output per employee, so fewer employees are needed to hit the same targets. Block Atlassian layoffs AI 2026 follow the same template Meta is using today.

The Question No One is Answering

Here is what the productivity argument leaves out. If your team can now do twice the work, there are two ways to respond. You can cut half the team and return the savings to investors. Or you can keep the team, attack twice the market, and build twice the product.

The companies executing Meta layoffs 2026 are overwhelmingly choosing the first option. That is a financial decision. It is not a strategic one. A genuine AI spending $135 billion play would look like deploying that newly freed capacity into new revenue lines, new geographies, new products. Instead, the savings are being routed into infrastructure and investor returns, while the human capital that understood the business, the customers, and the edge cases walks out the door on May 20.

A Fortune 500 CHRO quoted this week put it plainly: “We didn’t have a lot of strategic intent when our layoffs were done.” That is the honest version of what most of these announcements are.

What this Moment Actually Reveals

Tech layoffs AI productivity 2026 are not evidence that AI has made workers redundant. They are evidence that companies have figured out how to use AI as a justification for decisions they were already inclined to make. Oxford Economics noted in January that if AI were genuinely replacing labor at scale, productivity growth across the economy should be accelerating. It is not. Goldman Sachs research published this year found no meaningful relationship between productivity and AI adoption at the economy-wide level.

The companies that will look smart in three years are not the ones that cut the most people. They are the ones that figured out what to do with the extra capacity.

Meta AI spending $135 billion is a bet that the infrastructure will eventually justify itself. Cutting 8,000 jobs simultaneously is a hedge that it might not.

Advertisement

Google Now Generates 75% of Its Code With AI — And the Rules of Software Engineering Just Changed

By
Rohit Yadav
-
Dark visualization of AI-generated code streams on black background representing Google's 75% AI coding milestone

At Google Cloud Next this week, Sundar Pichai disclosed a number that reframes the entire conversation about AI and software development. Seventy-five percent of all new code written at Google is now generated by AI and subsequently reviewed by human engineers. That figure was roughly 25% in October 2024. By last fall it had climbed to 50%. In twelve months, it has tripled.

This is not a startup’s claim. This is Google, a company that maintains production systems at a scale most engineers will never encounter, writing in its official blog post that the majority of its new code no longer originates from human keystrokes.

What Pichai Actually Said

In his Cloud Next keynote post, Pichai wrote that Google is shifting to “truly agentic workflows,” where engineers orchestrate AI agents rather than writing code directly. He cited one concrete example: a complex internal code migration completed by agents and engineers working together ran six times faster than a comparable project completed a year earlier with engineers alone.

He gave a second example: the team behind the Gemini app on MacOS built the initial release using Google’s internal agentic development platform, Antigravity, going from an idea to a working native Swift app prototype in a matter of days. Both examples point to the same shift — agents compressing the time between intent and working software.

The policy dimension is notable. Google is now factoring AI adoption into employee performance reviews. This means the 75% figure is not a passive outcome of engineers experimenting with useful tools. It is a managed operational target.

Also Read: SpaceX Secures $60 Billion Option to Acquire Cursor

The Friction Beneath the Headline

The story has a layer worth understanding. Some employees at Google DeepMind have reportedly been permitted to use Anthropic’s Claude Code for development work in recent months. That decision apparently created internal friction, which signals something real: even inside Google, engineers prefer whichever model works best for the task, not necessarily the one built in-house. It also tells you that Google’s internal AI coding infrastructure, however mature, is not yet unambiguous best-in-class in the eyes of people who use it daily.

What this Means for Software Engineers

The instinct is to read a number like 75% and ask whether software engineers are being replaced. That is the wrong question. Google has not reduced its engineering headcount in response to AI-generated code. What it has changed is the nature of the job.

Writing code is becoming an output of the pipeline, not the primary skill of the engineer. What the job increasingly demands is the ability to decompose complex systems cleanly, evaluate what an AI-generated function actually does versus what it appears to do, and catch subtle errors that look correct at the commit stage but create problems in production. These are architectural and judgment skills. They take years to build and do not come from learning syntax faster.

For engineers early in their careers who built their value around coding speed and recall, the trajectory of this number is a serious signal. For engineers with strong systems thinking, security awareness, and product context, the same trajectory represents an expansion of what one person can actually ship.

The Trajectory is the Story

25% to 50% to 75% in twelve months. If that rate of change continues, the practical question is not whether AI dominates software development at major technology companies. It already does, at Google’s scale. The question is how fast the same shift reaches mid-market engineering teams, and what the second-order effects look like when the majority of new code everywhere originates from a model.

Google’s disclosure is the clearest benchmark the industry has seen from a company of this complexity. Every CTO reading it is recalibrating their hiring plans. Every engineer reading it should be recalibrating their skill investment.

Advertisement

SpaceX Secures $60 Billion Option to Acquire Cursor as Musk Bets on AI Coding

By
Rohit Yadav
-
SpaceX Cursor $60 billion acquisition option

SpaceX has locked in the option to acquire Cursor for $60 billion, or pay $10 billion for their ongoing collaboration. On April 21, the company announced the deal in a post on X, just before the New York Times published a report citing sources who said a $50 billion acquisition had been agreed, forcing the Times to update its story within minutes. Either way, the move is deliberate, and the timing is not accidental.

The deal puts a number on something the market had been watching for weeks: Musk’s AI coding gap, and his plan to close it by acquisition rather than invention.

What Is Cursor, and Why Does SpaceX Want It

Cursor is an AI-powered coding environment built for professional software developers. Its flagship feature, Composer, is an AI agent that edits, creates, and understands code across multiple files simultaneously. The Cursor AI coding startup valuation story is one of the most extreme in recent tech history: $2.5 billion in early 2025, $9 billion by May, $29.3 billion at its Series D close in November, and now a $60 billion acquisition option price on the table before the year is out.

SpaceX described the partnership as combining Cursor’s reach among professional engineers with its Colossus supercomputer, which it claims carries the equivalent compute power of one million Nvidia H100 chips. The SpaceX Colossus supercomputer Cursor integration is central to the pitch: train Cursor’s next model, Composer, on infrastructure that OpenAI and Anthropic cannot easily replicate.

The xAI Problem This Deal Is Trying to Solve

The context is important. In March, Musk publicly admitted that xAI was behind its rivals in coding. Two of Cursor’s most senior product engineering leads, Andrew Milich and Jason Ginsberg, left the company to join xAI, where both report directly to Musk. Last week, reports surfaced that xAI was already renting compute from its data centers to Cursor for model training. The Cursor Composer AI model SpaceX deal formalizes what was already being assembled in pieces.

But here is the contradiction this $60 billion acquisition option does not resolve: Cursor still runs on Claude and GPT models. It licenses access from Anthropic and OpenAI and sells it to developers. SpaceX is paying for a product that is commercially dependent on the exact companies it is trying to compete with, namely xAI coding rivals Anthropic and OpenAI. That structural tension does not disappear with a partnership announcement.

Also Read: Apple CEO Steps Down

The IPO Angle

None of this can be read outside the context of SpaceX’s imminent public offering. SpaceX filed for a confidential IPO earlier this month at a valuation exceeding $1.75 trillion. Adding a $60 billion coding AI platform to the prospectus might not be a product strategy. It is likely an IPO narrative designed to reframe SpaceX as a technology conglomerate rather than an aerospace company. The trial in Musk v. Altman begins in days. OpenAI was an early investor in Cursor. The timing is not subtle.

For developers and the AI industry, the signal is clear: the coding tools market is now explicitly a battleground between Musk’s empire and OpenAI. Cursor is the piece Musk needed, and he moved to lock in the option before anyone else could.

Advertisement

Tim Cook Steps Down as Apple CEO: John Ternus Inherits a $4 Trillion Company and an AI Gap

By
Rohit Yadav
-
Tim Cook steps down as Apple CEO

Tim Cook steps down as Apple CEO effective September 1, 2026, ending a 15-year run that turned Apple into a $4 trillion company and one of the most operationally efficient businesses in history. John Ternus, Apple’s SVP of Hardware Engineering, will take over as CEO. Cook will remain with the company as executive chairman.

The numbers Cook leaves behind are staggering. Apple’s market cap grew from approximately $350 billion to $4 trillion under his tenure, a more than 1,000% increase. Annual revenue nearly quadrupled from $108 billion in fiscal 2011 to $416 billion in fiscal 2025. Apple’s stock delivered a 1,886% return over that same period, compared to 483% for the S&P 500. Services revenue alone grew from roughly $12.9 billion to $85.2 billion — a business that now operates at 75% margins.

But Tim Cook steps down as Apple CEO at a moment of real strategic uncertainty. Apple Intelligence, the company’s flagship AI initiative announced at WWDC 2024, has underdelivered. The promised AI-supercharged version of Siri, capable of deep app integration and personal context awareness, was delayed out of 2025 entirely and pushed to 2026. Apple disabled its AI notification summaries for news apps after the feature generated fabricated headlines. John Giannandrea, Apple’s AI and machine learning chief, announced his departure. The company is now reportedly preparing to power Siri using Google’s Gemini models, a striking admission from a company that has spent a decade building its own silicon precisely to avoid such dependencies.

Who Is John Ternus?

John Ternus Apple CEO succession has been widely anticipated. Ternus, 50, has spent 25 years at Apple. He joined in 2001 as a mechanical engineer on the product design team and rose through hardware leadership to become SVP of Hardware Engineering in 2021. His team is responsible for iPhone, Mac, iPad, AirPods, Apple Watch, and Apple Vision Pro. He led the transition to Apple Silicon, arguably the most significant platform shift Apple made under Cook, and oversaw the MacBook Neo and iPhone 17 lineup.

What Ternus is not is a software executive, an AI researcher, or a services strategist. He is a hardware engineer. The board’s decision to appoint him is a signal about where Apple believes the competitive battle will actually be won.

Also Read: AI Now Generates More Internet Content Than Humans

The Hardware Bet

Apple’s AI strategy 2026 under Ternus will almost certainly center on devices as the AI interface. Apple has 2.5 billion active devices globally — a distribution advantage no AI lab, not OpenAI, not Google, not Anthropic, can replicate. Forrester analyst Dipanjan Chatterjee framed the Ternus appointment directly: a hardware leader signals that Apple will seek differentiation in its physical products, reframing the device itself as the substrate for intelligent experiences.

If that bet is right, Ternus is exactly the right person. A foldable iPhone is expected to launch shortly after he takes over. Rumored smart glasses are in development. The next hardware cycle could become the AI interface cycle, where the device form factor matters as much as the model behind it.

The risk is that the software and model gap widens faster than the hardware cycle can close it. OpenAI is building its own AI-first device. Google Gemini is already deeply embedded across Android. Microsoft Copilot is in every enterprise workflow. Apple’s moat is distribution, not capability, and distribution advantages erode when users start reaching for a competitor’s app on their own iPhone.

What Cook Leaves Behind

Apple market cap $4 trillion is Cook’s most visible legacy, but the operational infrastructure underneath it is equally significant. He built Apple’s services business from near zero to a Fortune 40-equivalent revenue line. He oversaw the AirPods and Apple Watch categories, both of which became the dominant products in their segments globally. He navigated tariff wars, supply chain disruptions, and a global pandemic without a single year of revenue decline.

The AI chapter is the exception. Cook had the resources, the silicon, and the installed base to lead in AI. The window was open from 2022 onward. Apple chose caution, on-device processing, privacy-first architecture, deliberate rollouts, while the rest of the industry moved at a speed that made caution look like paralysis.

Ternus inherits a company with extraordinary fundamentals and a genuine capability gap in the most important technology cycle of the decade. Whether a hardware engineer can close a software gap is the question that will define his tenure.

Advertisement

Sam Altman’s World ID 4.0: AI Now Generates More Internet Content Than Humans

By
Rohit Yadav
-
World ID 4.0 Sam Altman

The internet has an identity problem, and it stopped being a future concern on April 17, 2026.

That is when Sam Altman’s digital identity project World unveiled World ID 4.0 at its Lift Off event in San Francisco, announcing what the company calls “full-stack proof of human” infrastructure. The launch carried a list of integration partners that signals mainstream arrival: Zoom, Tinder, DocuSign, Shopify, Okta, and Vercel. These are not crypto-native platforms. They are the apps where hundreds of millions of people work, date, sign contracts, and build software every day.

The timing was deliberate. Crypto investment firm Pantera Capital put the underlying reality plainly this week: we have reached an inflection point where AI generates more information than humans. Distinguishing agents from humans, they argued, is now a critical moat for trust online.

What World ID 4.0 Actually Does

At its core, World ID uses a proprietary device called the Orb, a spherical iris scanner, to generate a unique cryptographic identifier for each verified human. The iris images are deleted after processing. What remains is an anonymous proof of personhood that can be used across integrated platforms without exposing personal data, using zero-knowledge cryptography.

World ID 4.0 introduces a redesigned account-based architecture for portable credentials across apps, key rotation and recovery, multi-device sessions, single-use anonymity nullifiers, and an open-source SDK that lets any developer integrate proof of human into their platform. The World ID app launches in public beta alongside the protocol.

The most consequential new addition is AgentKit, first launched in March 2026 in partnership with Coinbase and Cloudflare. AgentKit allows AI agents to carry cryptographic proof they are backed by a verified human, so platforms can distinguish a legitimate agent from rogue automated traffic. Platforms can cap usage per verified human, regardless of how many agents are deployed on their behalf.

The Tinder integration brings human verification to US dating app users, rolling out globally after a successful Japan pilot. Zoom’s integration uses a three-way biometric match to confirm the person on a video call is the verified human expected, addressing deepfakes in meetings. DocuSign’s adoption targets identity fraud in digital document signing. World ID 4.0 now has 18 million verified users across 160 countries, with over 150 million credential uses recorded.

There are other approaches emerging. Early-stage tools focused on AI content detection are beginning to appear, taking a different route to the same problem by flagging synthetic content rather than certifying human identity. None carry the infrastructure depth or enterprise partnerships that World is now assembling.

The Conflict Nobody Is Talking About

The market’s response to the April 17 announcement was revealing. Worldcoin’s native token WLD fell approximately 10% on the day, even as the broader crypto market rose. That divergence is not a verdict on whether the problem is real. It is a verdict on whether the market trusts this particular solution from this particular founder.

Sam Altman is the CEO of OpenAI, the company whose AI tools are among the primary drivers of the content authenticity crisis that World ID 4.0 is designed to address. The same person whose products helped flood the internet with synthetic content is now building the passport system that verifies you are real enough to use it. That tension has not disappeared because the product is technically sophisticated.

Why This Matters for AI and Data Professionals

For AI and data science professionals, the emergence of proof of human verification as a serious infrastructure category has direct implications. If Zoom and Tinder normalize iris-based identity verification, the expectation will spread into enterprise software, financial services, healthcare, and government platforms. Developers building agentic systems will need to consider human-linkage from the start, not as an afterthought.

The deeper question World ID 4.0 forces is not technical. In a world where AI agents act, transact, and communicate indistinguishably from humans, who gets to define what a verified person means online, and who gets to be the authority that issues that credential?

Sam Altman has a clear answer. The internet is still deciding whether to trust it.

Advertisement

An AI Agent Will Pay You: Inside the Humwork YC Launch

By
Rohit Yadav
-
Humwork YC launch

An AI agent will pay you to chat with it. That is not my framing. That is the opening line of Humwork’s own Y Combinator launch post, published by co-founders Yash Goenka and Rohan Datta on April 15, 2026. Humwork is a YC Spring 2026 company, and the product went live this week. The pitch underneath it is bigger than the product itself.

What Humwork Actually Does

When an AI agent hits a wall, Claude Code loops on a bug it cannot fix, Cursor produces code that does not compile, Lovable generates a design that breaks the flow, Humwork’s MCP server intercepts the failure and routes the problem to a vetted human in under 30 seconds. The expert sees the agent’s full context, the code it wrote, the errors it caught, everything it already tried, all PII-redacted. The human diagnoses and fixes. The agent picks up where it left off.

Setup takes 60 seconds through one MCP integration. The product works with Claude Code, Cursor, Codex, Lovable, Cline, OpenClaw, and any other MCP-compatible agent. The expert network includes senior engineers, lawyers, marketers, designers, and domain specialists. The founders’ own analogy in the launch post is clean: Waymo has remote driver assistance for edge cases, Humwork is the equivalent for AI agents.

Why the Humwork YC Launch Matters

The Humwork YC launch is worth paying attention to not because the product is novel in isolation, but because YC chose to back this thesis in Spring 2026, a batch dominated by autonomous agent pitches.

For almost three years, the AI industry sold one story. Agents will replace humans. They will book your travel, draft your contracts, review your legal work, ship your code, and you will not need the worker who used to. That story drove US private AI investment to $285.9 billion in 2025, per Stanford’s 2026 AI Index Report.

Humwork is the honest version of that story. In the founders’ own launch post: “The agent gets 80% of the way there, then loops on the same bug, makes the same bad architectural guess five times, hallucinates an important legal nuance, misses the brand judgment call, or quietly produces something that looks right but is subtly wrong.”

That is an admission the rest of the AI industry has been careful not to make in writing. The frontier labs know it. The enterprise buyers know it. The gap between the clean demo and messy production is the entire reason Humwork has a business.

The Power Dynamic Quietly Flipped

Yash Goenka’s founder bio on YC’s site describes Humwork as the company “where AI agents hire human knowledge workers.” Read that phrasing again. For three years we worried AI would replace human workers. Humwork inverts the relationship: humans keep the jobs, but the AI is the one doing the hiring.

Your Claude Code instance becomes the manager. You become the on-demand specialist it pages when it cannot figure out a race condition. The founders frame this as the future of all knowledge work: “AI will do most of the execution, but humans will still sit at the edge for the hard decisions: architecture, compliance, judgment, taste, tradeoffs, and exceptions.”

What the Humwork YC Launch Means for Builders

If you are building agent workflows, the Humwork YC launch is a signal to stop pretending your agent is end-to-end. Budget for escalation. Route hard problems to humans early. Design for the failure case, because the failure case is most of production.

If you are a real domain expert, a new income stream just opened up. The agents are not replacing you. They are hiring you, and YC just backed the infrastructure that makes the hiring transaction work.

Advertisement

NVIDIA Ising Is Not a Quantum Bet. It’s a GPU Bet.

By
Rohit Yadav
-
NVIDIA Ising quantum AI models pic

NVIDIA launched something quietly significant yesterday. Not a new GPU. Not a faster chip. A family of open-source AI models called Ising, named after a foundational physics model, designed to solve the two problems that are actually preventing quantum computing from being useful: calibration and error correction.

The announcement landed on World Quantum Day, April 14, 2026. The timing was deliberate. The strategy underneath it is even more deliberate.

What Ising Actually Does

Quantum processors are unstable by nature. Qubits decohere. Noise creeps in. Before you can run any useful computation, the hardware has to be tuned, and that tuning process has historically taken days of manual effort. Then, during any computation, errors accumulate and have to be caught and corrected in real time or the output is garbage.

NVIDIA Ising attacks both problems. Ising Calibration is a vision language model that reads measurements from quantum processors and automates continuous tuning, cutting calibration time from days to hours. Ising Decoding is a 3D convolutional neural network, available in two variants, one optimized for speed, one for accuracy, that performs real-time quantum error correction.

Both models are open-source, available on GitHub, Hugging Face, and build.nvidia.com. They integrate with NVIDIA’s existing quantum software stack: CUDA-Q and NVQLink, NVIDIA’s QPU-GPU hardware interconnect.

Who’s Already Using It

This isn’t vaporware with a press release. Ising Calibration has been picked up by Atom Computing, IonQ, Infleqtion, and Lawrence Berkeley National Laboratory’s Advanced Quantum Testbed. Ising Decoding is running at the University of Chicago, Sandia National Laboratories, SEEQC, and IQM Quantum Computers. That adoption list reads like a who’s-who of serious quantum hardware research — not startups chasing hype.

Also Read: The Stanford AI Index 2026 Is Out

The Real Play Here

Jensen Huang’s quote from the announcement is the clearest signal of what NVIDIA is actually doing: “With Ising, AI becomes the control plane — the operating system of quantum machines.”

That’s not a technical description. That’s a positioning statement.

NVIDIA is not making a bet that quantum computing will replace classical computing or displace GPUs. It is making a very different bet: that quantum computers, when they eventually become useful, will require AI to function — and that AI will run on NVIDIA hardware.

It’s the same logic NVIDIA used to lock in the AI training market before most companies knew they needed GPUs. Get into the infrastructure layer early, make it open source so adoption has no friction, and become the default. Dynamo did this for inference. Ising is doing it for quantum.

Ising joins a growing portfolio of NVIDIA open model families: Nemotron for agentic AI, Cosmos for physical AI, Isaac for robotics, Clara for biomedical, Apollo for physics, Alpamayo for autonomous vehicles. Each one extends NVIDIA’s surface area into a vertical that will eventually need serious compute. Quantum is just the latest frontier.

What This Means for the Industry

The quantum computing market is projected to surpass $11 billion by 2030. Right now, the dominant narrative in that space is about hardware — who builds the best qubits, which modality wins (superconducting, trapped ion, photonic). NVIDIA is reframing that narrative. Hardware without a reliable control layer is a science experiment. Ising is the argument that AI is that control layer, and NVIDIA owns it.

For quantum hardware companies, this is a double-edged development. NVIDIA is solving real problems they’ve been stuck on for years. But the solution comes with a dependency: the better Ising gets, the more deeply quantum processors are tied to NVIDIA’s software and hardware stack.

That’s not a conspiracy. It’s a business model. And it has worked every time NVIDIA has run it.

Advertisement

The Stanford AI Index 2026 Is Out. The Capability Numbers Are Historic. The Trust Numbers Are a Crisis.

By
Rohit Yadav
-
Stanford AI Index 2026 report

Stanford HAI dropped its 2026 AI Index this morning — 423 pages, nine years of independent data, no lab PR budget behind it. If you only read one document this year to understand where AI actually stands, this is it.

Here’s what the report actually says, past the headlines.

Capabilities are Accelerating, Not Plateauing

Every pundit who called peak AI in 2025 was wrong. The AI industry produced over 90% of notable frontier models in 2025 alone. On SWE-bench Verified, coding performance jumped from 60% to near 100% of the human baseline in a single year. On Humanity’s Last Exam — questions designed by subject-matter experts to represent the hardest problems in their fields — the top score was 8.8% in 2025. It’s now 38.3%, with the best models as of April 2026 crossing 50%.

Organizational adoption reflects this. AI adoption has reached 88% in the tech industry, and 4 in 5 university students now use generative AI.

The Transparency Collapse Nobody is Talking About

Here’s the number that should be the story: the Foundation Model Transparency Index dropped from 58 to 40 this year, with the most capable models disclosing the least. Google, Anthropic, and OpenAI have all abandoned the practice of disclosing their latest model’s dataset sizes and training duration. Eighty of the 95 most notable models launched last year were released without their training code.

The labs have made a deliberate choice: as the models get more powerful, they get less legible. This isn’t a side effect. It’s a competitive strategy.

Also Read: AI Didn’t Kill Writing. It Killed Coding.

The US-China Gap is Nearly Gone

In early 2023, OpenAI had a clear lead with ChatGPT. As of March 2026, Anthropic leads, trailed closely by xAI, Google, and OpenAI. Chinese models like DeepSeek and Alibaba lag only modestly. The US still outputs more top-tier models and higher-impact patents, but China leads in total patent output, model publication volume, and industrial robot installations.

US private AI investment reached $285.9 billion in 2025 — more than 23 times China’s $12.4 billion. And yet the performance gap is measured in single-digit percentage points. That should alarm every American policymaker.

The Talent Cliff

The number of AI researchers and developers relocating to the US has dropped 89% since 2017, with an 80% decline in the last year alone. The US is spending more on AI than any country in history while making itself less attractive to the people who build it. That’s a structural problem no amount of compute spending fixes.

The Public is Not Coming Along for the Ride

Only 10% of Americans say they’re more excited than concerned about AI in daily life. Meanwhile, 56% of AI experts believe it will have a positive impact on the US over the next 20 years. The US also reported the lowest trust in its government to regulate AI among surveyed countries, at 31%.

Employment for software developers aged 22 to 25 has fallen nearly 20% since 2022, and a third of organizations expect AI to shrink their workforce. The industry keeps pointing to benchmark scores. The public is looking at their job offers.

The Bottom Line

The 2026 AI Index is not a victory lap. It’s a stress test. The capabilities are real, the investment is real, the adoption is real. But the transparency is gone, the talent is leaving, and the public trust that makes any of this socially sustainable is at a low. Stanford’s data doesn’t editorialize. It doesn’t have to.

Advertisement

AI Didn’t Kill Writing. It Killed Coding. Here’s Why That Was Always Inevitable.

By
Rohit Yadav
-
Karpathy AI coding writing capability gap

Everyone had the same prediction. Writing would be the first casualty of the AI revolution. It was the obvious call: language models generate text, writers generate text, therefore writers go first. Clean logic. Wrong conclusion.

Andrej Karpathy, co-founder of OpenAI and one of the most credible voices on AI capability, published a thread on April 10 that explains precisely why this prediction failed. And his explanation is worth sitting with, because it reframes how we should think about where AI is actually winning.

The Perception Gap Nobody is Talking About

Karpathy’s first observation is about how unevenly AI capability is understood. A large group of people tried the free tier of ChatGPT at some point, saw it fumble basic questions, laughed at the hallucinations, and moved on. That experience formed their worldview on what AI can do.

A much smaller group uses frontier agentic tools like OpenAI Codex and Claude Code professionally, in technical domains, every day. They are watching these models restructure entire codebases, solve problems that would have taken weeks, and do it in an hour. For this second group, the experience borders on disorienting.

These two groups are not disagreeing about AI. They are talking about completely different products and calling them by the same name.

Why Coding Went First, Not Writing

The deeper point in Karpathy’s thread is structural. Coding has verifiable reward functions. When AI writes code, you can run a unit test. The test passes or it fails. That binary signal is exactly what reinforcement learning needs to improve. You can train a model on billions of correct and incorrect outcomes, and the gradient knows which direction to move.

Writing has no equivalent. There is no unit test for a good sentence. There is no pass/fail for whether a paragraph is compelling. Human judgment is the only signal, and human judgment is expensive, inconsistent, and impossible to scale the way automated tests are. RL cannot hillclimb on “this story resonated emotionally.” So it doesn’t.

Add to this the economic reality Karpathy points out: coding is where the B2B value is. The largest fraction of every AI lab’s engineering team is focused on improving coding performance because that is where enterprise revenue comes from. Writing improvements are slower, harder to measure, and lower on the priority list. Not because writing doesn’t matter, but because the incentive structure points elsewhere.

Also Read: How AI Is Replacing Personal Note-Taking

What This Means for Writers

The conclusion some will draw is that writing is safe. That’s the wrong takeaway. AI writing tools have gotten genuinely good at surface-level tasks: summarizing, drafting, formatting, rephrasing. Those tasks are being automated, and that pressure is real.

What is not being automated, and what is structurally difficult to automate, is the thing that makes writing worth reading: voice, narrative arc, the particular way a writer sees the world and chooses what to leave out. That is not surviving because it is artistically special. It is surviving because it cannot be scored. You cannot write a loss function for a great essay the way you can write a test suite for a working API.

This distinction matters. Writers who understand it will focus on developing exactly the capabilities that resist automation: original perspective, specific observation, structural judgment. Writers who miss it will compete against models on tasks the models are already winning.

The Real Lesson from Karpathy’s Thread

The gap in understanding AI capability is not just about technology. It’s about access. The people who have genuinely seen what AI can do are paying $200 a month and using it in highly technical environments. Most people are not. Their mental model of AI is outdated by roughly two to three product generations.

That gap is closing as agentic tools become more visible. But until it does, we will keep having the wrong conversation: debating whether AI is overhyped based on a version of the technology that no longer exists.

The capability is real. It is just unevenly distributed, and unevenly understood.

Advertisement
1234...358Page 3 of 358