Jensen Huang in 2026: Vera CPU and Nvidia's Agentic Pivot

In May 2026, Jensen Huang once again became the centre of gravity of the tech industry: he announced a quarter of $81.6 billion, opened a new $200 billion market with the Vera CPU, admitted that Nvidia had "largely ceded" the Chinese AI chip market to Huawei, and even flew to Beijing with Donald Trump for the summit with Xi Jinping. This guide breaks down each of these stories and shows what they mean for those building AI products in Brazil.

TL;DR — Jensen Huang in May 2026

• Nvidia Q1 FY27 revenue: $81.62B (+85% YoY); guidance of $91B for next quarter.
• Vera CPU launched in March 2026 as the world's first "agentic" CPU; already sold $20B and opens a TAM of $200B.
• Jensen Huang admitted Nvidia "largely ceded" the Chinese AI chip market to Huawei.
• On 13 May, he boarded Air Force One with Trump en route to the summit with Xi Jinping in Beijing.
• Nvidia surpassed $5 trillion market cap, approved an $80B buyback and raised the dividend.

Who is Jensen Huang (quick recap)

For those just joining: Jensen Huang founded Nvidia in 1993 at a Denny's in California, alongside Chris Malachowsky and Curtis Priem. For decades, the company was synonymous with gaming GPUs. In 2012, when a neural network called AlexNet won ImageNet running on two GeForce GTX 580s, it became clear that that chip for "rendering Crysis" could also do mass linear algebra — exactly what training deep learning requires. From there, the thesis built itself.

More than a decade later, Nvidia crossed $5 trillion in market value and Jensen Huang leads more than 42,000 people. By the company's numbers for fiscal year 2025, annual turnover stood at 2.5% — somewhere between five and ten times lower than the tech sector average — and about 40% of the workforce has been at the company for over five years. This is no coincidence: Huang's management style has become a study in itself, and I'll return to it below.

Q1 FY27 Results and the "Parabolic Demand"

On 20 May 2026, during the first-quarter fiscal earnings call, Jensen Huang opened with a phrase that went viral in seconds: "Demand has gone parabolic. Agentic AI has arrived." The numbers justify the tone:

Indicator	Q1 FY26	Q1 FY27	Change
Total revenue	$44.06B	$81.62B	+85%
Q2 guidance	—	$91B	—
Approved buyback	—	$80B	—
Dividend	—	Increased	—
Market cap	~$3T	>$5T	+65%

For context: Nvidia's entire annual revenue in FY24 was $60.9B. The current quarter already exceeds that. When Jensen Huang talks about the transition from "AI factory" to "agentic AI factory", it's not loose narrative — it's the sales chart torn in two.

And the market priced it in: after the call, the stock rose, the board simultaneously approved an $80B buyback and a new dividend increase. In capital allocation terms, this means "the thesis still holds, and there's still cash to spare."

Vera CPU: The $200 Billion Agentic Pivot

The missing piece in the puzzle is the Vera CPU, announced at GTC in March 2026 and now in the spotlight. Traditionally, Nvidia sells GPUs to train models. Vera is a CPU designed from scratch to run agents — the orchestration, planning, tool calling and decision-making work an agent does between one LLM "thought" and the next.

Why a dedicated CPU for agents?

• A typical agent spends 70-90% of its time on CPU-bound code: API calls, JSON parsing, flow control, vector search, database writes.
• GPU solves the "think for me" part. But using a GPU to do JSON.parse() is throwing money away.
• Hyperscalers (AWS, Azure, GCP, Oracle) want CPUs close to Rubin GPUs without relying solely on Intel x86 or AMD EPYC.

The numbers Jensen Huang put on the table

On the call, he mentioned that Vera has already done $20 billion in sales in the current year — sold both in a bundle with Rubin GPU and standalone — and that the combined TAM becomes $200 billion. By comparison: the total x86 server market in 2025 was around $110B. In other words, Nvidia is saying it will open, from scratch, a market larger than conventional servers.

For those developing AI agents for businesses, the message is direct: the topology "agent runs on commodity x86 CPU" may age quickly between 2027 and 2028.

The "Cession" of China to Huawei

In the same period, Jensen Huang was unusually direct in a press conference: Nvidia "largely ceded" the Chinese AI chip market to Huawei. The phrase caught everyone off guard.

To unpack: US export restrictions, escalating since 2022, have been cutting which AI chips can go to China. Nvidia adapted (H800, H20) and lost margin each round. Huawei, in parallel, launched the Ascend line (910B, 910C) and, without direct competition from top-tier Nvidia, ate the core of the Chinese domestic market — banks, telecoms, government and local foundation models like DeepSeek, Qwen and ERNIE now run mostly on Ascend silicon.

Huang is acknowledging the scale of the damage. By Nvidia's own estimates, this was a TAM of about $50 billion in 2026. That's not small. And there's a second-order effect: every Chinese hyperscaler that got used to Ascend won't easily switch back — there's an entire software layer (CANN, MindSpore) that creates lock-in equivalent to CUDA. To understand the other side of this fight, it's worth reading what Huawei has been doing with AI chips and Cloud in Brazil in 2026.

The Trip with Trump to Beijing

On 13 May 2026, in a last-minute move, Trump personally called Jensen Huang asking him to join the official US delegation at the summit with Xi Jinping. Huang flew to Alaska to board Air Force One. In Beijing, he attended a state dinner with Xi, took selfies with Elon Musk, and even appeared in a "food escape" to eat noodles on the street — recorded in local media.

The logistics seemed informal; the subtext was not. The White House announced new export licences that would allow Nvidia to resume selling its most advanced second-generation chips to China. For Huang, it's the game of the possible: sell what you can, keep a seat at the table and — in his words to journalists — "represent the United States" in the dialogue.

Translation for business: if 2024-2025 were years of Nvidia "losing" China, 2026 could be the year of trying to re-enter with an intermediate catalogue while the Vera + Rubin duo opens markets in the US, Middle East and Europe.

"Tough Love" Management: What to Learn from Huang's Style

Jensen Huang gave an interview to Fortune on 26/05/2026 about internal culture. The summary is confrontational: "You can't go a day without some criticism". Everything any of the 42,000 employees shows him receives some criticism. He credits the style to his Taiwanese parents, describing his childhood as "a form of torture" where nothing is good enough.

The nuance usually left out of clippings: Huang says that after the feedback is given, the relationship resets to zero. No grudges, no retaliation, no forced stack ranking like Microsoft 2010. That's why turnover is tiny.

Points worth learning for those who lead — and here I speak as a CTO who has seen various internal cultures succeed and fail:

1. Direct criticism works when it's technical, frequent and without personal dimension.
2. Criticism without follow-up becomes toxic. Huang's model assumes the next meeting is already looking at the next delivery, not rehashing the previous one.
3. Those who deliver 7 days a week (as Huang does) have the social licence to demand 7 days. Those who don't, don't.
4. Criticism must come with a way out. If feedback doesn't come with a path to improvement, it's just an attack.

It's not a style replicable in every Brazilian corporate culture, but elements of it fit small, high-trust teams.

The Speech at Carnegie Mellon

In the same month, Jensen Huang was the commencement speaker at Carnegie Mellon University — home to one of the world's top robotics and AI schools. The speech was titled "Shape what comes next". The points he hammered home:

• "You are entering the job market with the most powerful tool any generation has had." A direct reference to generative AI.
• "AI won't take your job. A person who uses AI will." A pragmatic take on the debate.
• For founders: "Start with a problem painful enough to justify all the 'no's you're going to hear."

It's not a revolutionary speech — Huang has been repeating these lines for a year — but the symbolic stage (CMU) and timing (a week after the earnings call) carried extra weight.

Key Events of May in Chronological Order

To round out the picture:

1. 13/05 — Trump invites Jensen Huang at the last minute; boarding in Alaska, flight to Beijing.
2. 14-16/05 — Summit with Xi; state dinner; signalling of new export licences.
3. 19-20/05 — Q1 FY27 results; $80B buyback; "agentic AI has arrived".
4. 21/05 — Statements about ceding the Chinese market to Huawei.
5. 24-26/05 — Speech at Carnegie Mellon and Fortune interview on internal culture.
6. 27/05 — Market already pricing in Vera + buyback; analysts begin revising agentic TAM.

Looking at these five days, it's easy to see why Jensen Huang became the tech personality of the month: he operates on a register that mixes geopolitics, capital allocation, product and culture in rapid succession — and each move supports the next.

What This Changes for Your Product

If you develop software, sell SaaS in Brazil, or are building something on the agentic wave, three practical effects:

1. Inference cost will drop again. Vera + Rubin reduces x86 CPU per agent and improves GPU amortisation. In 12-18 months, this will pass through to supplier API prices — so architectures that are "too expensive" today become viable.
2. An "all-GPU" architecture will look naive. There will be a more explicit split between "thinking" (large model) and "doing" (agentic CPU + tools). Those who anticipate this gain latency and margin.
3. The chip sovereignty debate will heat up in Brazil. If China built an ecosystem with Huawei even under restrictions, BNDES, MCTI and local operators will look at that playbook. The case of the Nintendo Switch 2 and Nvidia's AI paradox already showed that even consumer products now depend on this chain.

When not to ride the wave now

In technical honesty: if your product is a classic B2B SaaS (CRM, ERP, payment gateway) without an embedded agent and your end customer isn't asking for AI, resist the pressure to embed an agent "just for show." The operational cost is still high, and the overhead of maintaining prompts/eval/observability eats into margins. Wait for the Vera + Rubin cycle to lower inference costs around 2027 and enter with a ready product, not a demo.

Conclusion

Jensen Huang's Nvidia in 2026 is not just selling chips: it's rewriting the topology of computing while doing industrial policy between Washington and Beijing. For Brazilian developers and managers, the work is to understand which pieces of this game fit your product now — and which fit in an 18-month plan. If your AI roadmap doesn't mention agents, Vera CPU or supplier strategy, it's probably outdated.

I'll keep following here on the blog: subscribe to the RSS, or follow Agathas Web on LinkedIn and we'll swap notes on what's happening on the front line of AI.