Intel Foundry: Google and Nvidia Choose a TSMC Alternative

The timeline of an unexpected reversal

For years, Intel Foundry has been the most controversial bet at the Santa Clara group. Technology delays, operating losses and a merciless comparison with TSMC had convinced many analysts that the division was destined for drastic downsizing. In 2026, however, two events sharply reversed that narrative.

The first involves Google. According to The Decoder, the Mountain View giant has placed an order for more than three million AI chips to be made by Intel, with deliveries scheduled for 2028. The second involves Nvidia. Jensen Huang's company is testing its upcoming architecture, called Feynman, on Intel's production lines. So within just a few months, the two most influential players in the global AI ecosystem chose to bet on a supplier that seemed out of the game.

The market reacted with interest to these announcements. It's worth looking at the structural reasons behind this reversal, beyond the individual company decisions.

TSMC: the bottleneck nobody wanted to admit

TSMC today produces the dominant share of the world's advanced chips. This concentration worked well for years. But the exponential growth in AI demand has turned a competitive advantage into a systemic risk.

According to recent analysis by Gartner, the global semiconductor market grew at a double-digit rate in 2025. That surge has saturated TSMC's production capacity, especially on the most advanced nodes that AI chips require. As a result, wait times have grown longer and costs have risen. Big tech companies have started looking for real alternatives, no longer as a theoretical diversification exercise but as an immediate operational need.

Despite its struggles in recent years, Intel has physical fabs in the United States and Europe. It also benefits from the incentives of the American CHIPS Act. For Google and Nvidia, relying on Intel therefore also means reducing the geopolitical exposure tied to production being concentrated in Taiwan.

Winners, losers and everyone in between

Reading this scenario means telling the different players in the supply chain apart.

Intel is the most obvious beneficiary. The Foundry division is landing top-tier orders, renewed technological credibility and positive momentum after years of criticism. But the road ahead is still demanding. Intel's process technology has yet to prove it is reliable at industrial scale for the most complex AI chips. The recovery is real, but not yet secured.

TSMC is not losing its dominant position in the short term. On the contrary, the demand it receives still exceeds its production capacity. The risk for TSMC is medium-term: if Intel and Samsung manage to close the technology gap, market concentration will shrink. In particular, Intel's 18A node is the decisive test for the next two years.

Samsung Foundry sits somewhere in the middle. Like Intel, it is trying to attract AI customers by broadening its offering. But Samsung has also struggled with wafer yields on the most advanced nodes. The head-to-head with Intel for Google and Nvidia orders will therefore be an important indicator to watch in 2027.

End customers, including cloud companies, SaaS platforms and AI infrastructure providers, benefit from greater production availability and a potential drop in prices over the medium term. That means lower costs for accessing AI computing power, with positive effects on the digital tools available to Italian SMEs as well.

SHM Studio's take: supply chain as a strategic variable

At SHM Studio, we follow these developments not out of academic interest, but for a practical reason. Italian companies adopting AI solutions, from automated digital marketing to predictive analytics, from advanced SEO platforms to content generation tools, depend indirectly on the availability of AI chips in the data centers of the big cloud providers.

When the semiconductor supply chain jams, inference costs go up, APIs slow down and SaaS vendors revise their development plans. That's why production diversification across Intel, TSMC and Samsung isn't news reserved for Wall Street analysts. It's a variable that shapes the technology roadmap of any company that has built AI into its processes.

For Italian SMEs weighing investments in AI solutions, this scenario also points to an important consideration. The choice of cloud provider, and its dependence on specific chip manufacturers, becomes an additional evaluation criterion beyond features and pricing. In short, semiconductor geopolitics is joining the checklist for technology purchasing decisions.

Feynman architecture: what’s at stake for Nvidia

Nvidia's test on Intel Foundry deserves a closer look of its own. The Feynman architecture is the generation after the Blackwell chips, which are currently at the center of global AI demand. According to industry sources cited by Wired, Nvidia is exploring a multi-foundry strategy to reduce its dependence on a single manufacturer.

This approach has significant technical implications. Designing a chip that can be made by multiple foundries requires a flexible architecture and a longer qualification process. Still, the benefits in supply chain resilience justify the investment. So if Feynman were qualified on Intel Foundry, it would be a definitive sign that the AI chip market is no longer a de facto manufacturing monopoly.

For companies that develop or integrate AI solutions, including those that rely on platforms like the ones we use for our Google Ads campaigns and LinkedIn campaigns, wider availability of Nvidia GPUs at more stable prices is a tangible benefit.

The work still in progress: what could go wrong

It would be unwise to read these developments as a victory Intel has already secured. There are real risks that deserve attention.

The first is production yield. Intel has to show that its 18A process can produce complex AI chips with acceptable defect rates at industrial scale. The ongoing tests with Nvidia are aimed precisely at that verification. Even so, Intel's track record on advanced nodes calls for caution.

The second risk is timing. Google's deliveries are expected in 2028. That leaves a horizon of nearly two years in which a lot can change. TSMC could expand its production capacity. Samsung could solve its yield problems. New players, such as ASML with its alternative customers, could shift the balance.

The third factor is cost. Intel Foundry has to be competitive not only on technology, but on price too. If the cost per wafer stays higher than TSMC's, diversification will come at a price, and customers will have to decide whether it's worth paying.

To learn more about how these dynamics shape AI-driven web development and content marketing strategies, our blog publishes regular updates.

Next moves: what to watch over the next 18 months

If you're following these developments with a practical interest, there are a few key indicators worth watching.

• Qualification of Intel's 18A node: the moment Nvidia officially announces the qualification, or the failure, of the Feynman test on Intel Foundry will be a significant market event.
• TSMC capacity expansion: the new fabs in Arizona and Japan will gradually come into production. Their actual capacity on advanced nodes will determine how much room is left for Intel.
• AI API pricing: an indirect indicator, but one any SME can track. If the cost of accessing AI models on the major cloud providers starts to fall, the pressure on the supply chain is likely easing.
• Announcements from other hyperscalers: if Microsoft or Amazon follow Google in relying on Intel Foundry, the signal would become systemic rather than isolated.

For Italian companies that want to build a solid digital strategy, one that doesn't depend on swings in the global tech supply chain, the starting point remains a clear assessment of your own needs. The SHM Studio contact page is open for an initial no-obligation conversation.