Daily Update - July 6th, 2026
Nvidia Kyber delayed, Nvidia's revenue-share financing, Intel price hikes, $510B in AI funding, Hon Hai's record AI server quarter, and Korea's chip-tax growth fund.
Semis were all down last week; let’s see what this week has to offer. SemiAnalysis dropped the news that NVIDIA’s Kyber rack is delayed. A new Intel memory patent. And tons of news bits to cover. Slightly different format today with some cool tweets.
Let’s get into it. — Austin & Vik
Be sure to check out the Semi Doped podcast on YouTube or your favorite podcast player!
Nvidia Kyber NVL144 Delayed Over 12 Months to 2028
Nvidia’s Kyber NVL144, demoed at GTC three months prior, has been delayed by over 12 months, pushing its release to 2028. This delay also follows the cancellation of Nvidia’s NVL72x2 back-to-back rack architecture, which limits Rubin Ultra’s scale-up domain. (SemiAnalysis)
Vik: I had a wild conspiracy theory on X a month ago that this will happen! Lots of people seem to not believe that this is true though, and some say to wait for official announcements. The discussion on X is quite wide; lots of opinions and takes to process.
Austin: Favorable news for AMD, with Helios rack (MI455X) shipping 2H26 and MI500 series in 2027. AMD’s Advancing AI event is in two weeks, maybe we’ll hear more about the 2027+ roadmap and MI500 there?
Nvidia Introduces Revenue-Sharing Financing Model for AI Cloud Providers
Nvidia is deploying a revenue-sharing financing model to underwrite AI chip purchases by emerging cloud providers. In exchange for a cut of their rental revenues and a backstop for unused inventory, the model allows customers to acquire Hopper and Blackwell GPUs that their balance sheets might otherwise constrain. This strategy expands Nvidia’s addressable market by de-risking CapEx for customers. (The Information)
Vik: The recurring revenue model will smooth out financials for Nvidia instead of waiting for big purchase orders. Qualcomm’s % cut from handset provider sales for CDMA was a good source of money for decades for them. I wonder if this is the start of something similar.
Austin: Financing is an under appreciated competitive advantage Nvidia has against AI accelerator startups.
Global AI Startup Investment Reached $510B in H1 2026
Global startup investment hit $510 billion in the first half of 2026, with OpenAI and Anthropic capturing 43% of total venture capital. In parallel, AI CapEx from US hyperscalers is projected to reach 3.2% of GDP by 2027. The massive capital allocation confirms a structural industry shift from capability races to capacity deployment, with power constraints emerging as a tangible bottleneck. (Crunchbase)
Vik: Lion’s share of the startup investment is going to the top two, but still plenty of spending in AI overall. Looking forward to another half-tril in H2 2026.
Austin: It’s a good time to be raising money, even if you’re not OpenAI or Anthropic; the article says 16 companies raised billion-dollar rounds in the second quarter of 2026!
Intel Hikes Select CPU Prices, Citing Supply Costs and Demand
Intel confirmed price increases on some consumer and server CPUs, attributing the changes to market dynamics, rising costs, and high demand. Select enthusiast processors increased by $30 to $50, while data center-grade products saw increases of hundreds or thousands of dollars. For example, some Xeon 8000-series ‘Emerald Rapids’ processors now have higher Recommended Customer Prices than at their late 2023 release. (tomshardware.com)
Vik: If Intel can charge more for CPUs in an era of high demand, which AMD forecasts will continue till 2030/31, then CPU revenues 🚀… good for Intel’s financials for sure. Exciting to see how things will roll out at the next earnings call.
Austin: Inflation hits consumers again. First it was memory. Now CPUs. Folks are gonna have a love-hate relationship with AI. Love the usefulness, but hate the inflation.
Hon Hai Q2 Sales Beat Estimates on Record AI Server Revenue
Hon Hai Precision Industry reported second-quarter revenue that exceeded analyst estimates, with record sales in its AI server division as demand from hyperscale and enterprise customers stayed elevated. The Taiwan-based contract manufacturer, a key assembler in Nvidia’s supply chain, attributed the outperformance to sustained AI infrastructure spending; Oracle’s expanding capital commitments were cited among the demand drivers. (Bloomberg.com, Reuters)
Vik: Hon Hai, aka, Foxconn, assembles servers for Nvidia hardware. The article reports that AI servers have overtaken consumer electronics in terms of revenue. This is a proxy compute capacity really, and so far there is no slowdown even as there are fears of overcapacity.
South Korea Plans Growth Fund Backed by Chip Tax Revenue
South Korea is considering establishing a growth fund seeded by tax revenue from the semiconductor industry, according to Yonhap News. The move comes as domestic chipmakers Samsung Electronics and SK Hynix are committed to spending at least 1,350 trillion won ($880 billion) combined on chips and data centers, investments the government aims to leverage into a broader industrial financing vehicle. (Bloomberg Tech)
Vik: Feed the windfall from memory companies into a growth fund for other sectors. Good diversification strategy.
Austin: Heck yeah. Agree. Korea shouldn’t overinvest all its dollars in memory right now.
Tweet #1: Intel XBM
Intel XBM is good to see. Just don’t make it have the same fate as Optane. Too soon? Read the patent for yourself.
Tweet #2: Interconnects >> Memory
Amazing post by Big Boss on X, which explains from first principles why we should be designing AI accelerators with interconnects in mind.
