This report is from this week’s The Tech Download newsletter. Like what you see? You can subscribe here.
Nvidia has cemented itself at the heart of the AI boom with a monopoly on the most powerful chips to train and run models, but a growing crop of startups are set on challenging the company’s supremacy.
And increasingly, investors are throwing huge sums behind them. In 2026, AI chip startups raised $8.3 billion in funding, globally, according to Dealroom. Barring a near total collapse of the market, the sector is expected to see record sums pumped into it this year.
So what’s causing the spike?
While Nvidia’s graphics processing units (GPUs) — which were originally designed for gaming — have been effectively repurposed for AI training, focus is now shifting to the most efficient ways to actually deploy the tech in applications, known as AI inference.
The argument of these chip upstarts is this: GPUs weren’t purpose-designed for AI, and therefore, novel system architecture will bring big savings in energy and cost.
“Inference is dominant now, and the existing GPU architecture wasn’t built for it in ways that matter most at scale,” Patrick Schneider-Sikorsky, director at the Nato Innovation Fund (NIF), which has invested in U.K. AI chip startup Fractile, told me.
Nvidia, which has huge advantages as the world’s most valuable company with an almost limitless supply of cash, is still racing to develop new chips to power AI.
In December, the company acquired assets from AI inference startup Groq for $20 billion and announced it had invested $4 billion into two companies developing photonics technology in March.
The chip giant also spent more than $18 billion on research and development in its most recent full financial year, ending January 2026.
Startup funding
But investors haven’t been deterred from throwing money behind new, and often untested at scale, AI chip technology.
In the U.S. — where many of the biggest rounds have been raised — Cerebras Systems picked up $1 billion in February, and there have been $500 million rounds in 2026 for MatX, Ayar Labs and Etched.
European companies have raised comparatively smaller sums, but Axelera and Olix have both raised rounds north of $200 million this year. Others, including Euclyd and Optalysys told me they’re planning rounds of at least $100 million in 2026, as are Fractile and Arago, according to reports.
“It’s no longer a niche bet,” said Carlos Espinal, managing partner at European VC Seedcamp, which backed chip startup Vaire Computing. “It’s becoming a core part of how people think about AI infrastructure.”
Latest updates
Anthropic and OpenAI both announced major U.K. expansion plans. Anthropic unveiled a new office space for 800 people, while OpenAI said it would open its first permanent London office with capacity for over 500 team members.
TSMC on Thursday reported a 58% increase in first-quarter profit, beating estimates and hitting a fresh record as demand for artificial intelligence chips stayed strong.
OpenAI abandoned plans to rent capacity directly from a Norwegian data center, with Microsoft taking up the extra compute, days after confirming it paused a similar project in the U.K. The ChatGPT maker would instead rent capacity from Microsoft, the company told CNBC.
Amazon said Tuesday it would acquire Globalstar in a deal worth about $11.57 billion, as it looks to give its nascent Leo satellite internet business a boost and compete with Elon Musk’s SpaceX.
Uber on Friday agreed to purchase an additional 4.5% of shares of German food delivery firm Delivery Hero from the company’s biggest shareholder Prosus.
Stock of the week
ASML stock has dropped after the company reported earnings on Wednesday.
ASML stock has been on the decline since it announced results on Wednesday, despite raising its sales forecast for 2026 and beating first-quarter revenue and profit expectations.
Astronomical expectations around the AI boom likely caused the negative reaction, alongside tightening restrictions on export controls, which caused a drop in the percentage of net sales to China.
