Google’s ‘firebombs’ have sparked off battles over both hardware and software in the arena of AI

These are the Two Races of 2026, officially declared not by proclamation, but by two ‘firebombs’ thrown by Google. They target two tech giants that appeared almost unassailable till recently: the king of frontier AI models and the monarch of AI hardware. The ensuing battles for intelligence supremacy and silicon sovereignty are now the true market drivers, overshadowing the usual concerns about an AI market bubble or cyclical demand.

Google vs the frontier-model king: The target of Google’s first firebomb is the ecosystem of closed proprietary AI models led by OpenAI. Just a few quarters ago, the latter was celebrated for its breathtaking growth and ability to take on the reigning champ of search. Yet, with a single act of execution, Google has changed the competitive dynamic.

Google’s ‘firebomb’ was its seamless integration of Gemini with its existing services to sustain its dominance over a sprawling, billion-user ecosystem. It’s not about competing with a standalone chatbot, but about making its model ambient. Its token processing—the true measure of AI operational scale—grew 140 times over 18 months, a figure that underscores the sheer volume of AI woven into its existing products like Search, Android, and Gmail.

Rather than compete for chatbot subscribers, Google is leveraging its entire user base, making it nearly impossible for a newcomer to reach users with rival AI Agents. Competitors are left trying to build new browsers and ‘canvases’ from scratch, while Google already owns the basic interfaces of digital life.

But this is not merely about Google’s dominance. The greater peril facing closed-source giants like OpenAI is the tsunami of open-source AI from China.

While Google is a fearsome direct rival, the long-range threat comes from the efficiency and cost advantage of the open source movement, particularly in the East. Chinese research has extensively adopted a ‘mixture of experts’ architectural method, enabling multiple teams to explore diverse paths to model efficiency. Multiple open-source models have been achieving parity with proprietary giants across various domains, delivering solutions at a fraction of the cost. Closed-model businesses now face an accelerating global wave of free-AI alternatives.

Google vs the monarch of silicon: The second ‘firebomb’ was dropped directly on hardware major Nvidia. For years, its GPUs drove the AI revolution, but this dominance is being challenged by the concept of vertical integration championed by Google.

The latter’s firebomb is its commitment to proprietary, purpose-built silicon. The specifications of its latest Ironwood Tensor Processing Unit are not a threat to be ignored; they are a direct challenge. The chip boasts 4,614 T-flops of peak compute power and 192GB of high-bandwidth memory, showing Google’s ability to engineer hardware optimized exclusively for its models. This move will drive down costs and secure a strategic proprietary advantage over its rivals.

However, this war goes beyond the skirmish between Google and Nvidia. Google’s success is a shot in the arm for the inhouse chip design teams at every global giant, including Microsoft, Amazon, Meta, Tesla, Huawei and Alibaba. They recognize that the pursuit of custom silicon is no longer an optional investment, but an existential necessity. The promise is not only material cost savings, which can be staggering over time, but the competitive edge of a whole-stack solution.

By tuning proprietary models for custom chips and vice versa, these firms could gain significant product and service advantages that general-purpose hardware can’t offer. This second race has become a global sprint for hardware self-sufficiency, initiated by the undeniable success of Google’s early lead.

The AI future has been thrown open: Sam Altman has already issued a ‘Code Red’ alert within OpenAI, while Nvidia’s concerns were evident in its unusual backhanded compliment to Google in a tweet. The two races set in motion by Google will define how 2026 will play out. These are battles of hyper-innovation that may or may not displace the current leaders, but put them under pressure. And there will be many more product and model announcements in 2026 than we had this year.

Given the rapid evolution of technology and highly focused nature of the players involved, which company will ultimately prevail is unknown. However, the one certainty is that the massive spending required to fight these two wars—for intelligence and silicon—promises a stable boom for adjacent sectors that supply the computational infrastructure for AI model development and chip building.

AI doomerism is a thriving industry as we near the end of 2025, and with some reason too. But we should note that model improvements are not stalling, despite proclamations by sceptics. And the world’s largest cash-rich companies are in a chip war—or an unparalleled competitive race with nobody expected to admit defeat for many years.

The author is a Singapore-based innovation investor for GenInnov Pte Ltd.