Humankind is witnessing a technological revolution so extreme that its full magnitude might extend beyond the scope of our intellect. Generative AI (GenAI) is doubling its performance every six months [1], outpacing Moore's law in what the industry calls Hyper Moore's Law. Some cloud AI chipmakers expect to double or triple performance every year for the next ten years [2]. In this three-part blog series, we'll explore today's semiconductor landscape and innovative chipmaker strategies, dive deeper into the significant challenges ahead in Part II, and conclude in Part III by examining the emerging changes and technologies powering the future of AI.
At this explosive pace, experts project that Artificial General Intelligence (AGI) will materialize around 2030 [3][4], followed soon after by Artificial Superintelligence (ASI) [5]. AGI will possess human-like reasoning while ASI will surpass it, reprogramming itself beyond the comprehension of even the most expert minds. The rapid evolution of AGI into ASI through self-modification is commonly known as the intelligence explosion.
The chart below makes this exponential acceleration impossible to ignore. In just a few years, AI has blown past human performance in many complex tasks once thought to require deep expertise. It is also rapidly closing in on others, such as reasoning, math problem-solving, and code generation. Capabilities that stagnated for decades are now leaping forward in months. If this pace holds, AI will soon outperform humans across nearly every cognitive domain, turning AGI from theory to inevitability.
The impact of this rapid evolution on the semiconductor industry is profound, as GenAI is driving strong demand for advanced cloud SoCs powering training and inference. By 2030, analysts project this specialized sector will near $300 billion, with a CAGR close to 33% [7].
This surging demand is shattering old assumptions about how fast the semiconductor market can shift, with GenAI’s rapid advances proving they can disrupt it overnight. The chart shows that generative AI reached adoption levels in two years that took the PC nearly a decade to achieve, and did so even faster than smartphones, tablets [8], and the internet.
Geopolitics further amplifies these market tremors. The U.S.-China tech rivalry has turned semiconductors into a strategic asset in the arms race between the two superpowers. The U.S. has imposed broad export restrictions that block China’s access to American AI processors, aiming to slow its progress toward AGI [11]. China is fighting back with disruptive moves, such as open-sourcing DeepSeek-R1, which was built using earlier-generation chips due to U.S. chip curbs.
Diverse Chipmaker Strategies For Maximizing Throughput
The table below compiles the latest specifications for several leading AI chips. All figures are based on a single chip, not multi-chip systems, such as NVIDIA's GB200 NVL4. Only chips available at the time of publication are included.
This data offers a high-level overview rather than a strict apples-to-apples comparison, which would require testing all chips under identical workloads and conditions. Performance Per Watt (PPW) was deduced by calculating (PFLOPS⋅1000)/Watt(PFLOPS·1000)/Watt, yet some chipmakers do not publicly disclose wattage (see N/A below).
The construction of the table relied primarily on official vendor specifications and reputable third-party sources. In a few cases, values were extrapolated, such as estimating 16-bit Floating Point (FP16) performance based on 8-bit Floating Point (FP8) Peta Floating Point Operations Per Second (PFLOPS).
Comparing all vendors side by side like this reveals the diverse strategies in AI accelerator design:
The accelerating trajectory of generative AI isn't just transforming technology—it’s reshaping the semiconductor industry and intensifying geopolitical tensions. As chipmakers race to deliver unprecedented processing power and efficiency, the strategies deployed are diverse and innovative, yet the challenges are profound. This rapid progress comes with significant hurdles, especially for cloud-based AI deployments where scaling effectively and sustainably becomes increasingly complex.
Delving deeper into the painful realities of scaling cloud AI infrastructure. We'll examine practical obstacles chipmakers face—including hardware failures and reliability issues such as Silent Data Corruption (SDC), surging power demands, and workload growth that continues to outpace Moore's Law.
Discussing the critical optimization factors for GenAI chipmakers. We will explore how chipmakers differentiate their products using novel architectures, packaging strategies, and optimization techniques that target performance, power efficiency, and reliability.
