Alphabet’s Two-Front War: New AI Silicon Meets Internal Coding Anxiety

Alphabet has unveiled its eighth-generation Tensor Processing Units at Google Cloud Next in Las Vegas, splitting the lineup into two specialized variants for the first time. But even as the company showcases hardware capable of 121 exaflops of compute power, internal concerns are mounting that Google is losing ground in the AI coding tools race — a vulnerability that could undermine its broader platform ambitions.

The Chip Strategy: Training and Inference, Separated

The TPU 8t and TPU 8i represent a deliberate architectural shift. The 8t is built for large-scale model training, with each Superpod aggregating up to 9,600 chips to deliver 121 exaflops — triple the performance of its predecessor. The 8i, by contrast, is optimized for inference, packing 288 GB of High Bandwidth Memory and promising 80 percent more performance per dollar than earlier generations. Both chips leverage liquid-cooled Axion processors and double performance per watt. Alphabet expects them to become available through the Google Cloud AI Hypercomputer later in 2026.

The hardware rollout dovetails with a platform reorganization. Vertex AI is being absorbed into the new Gemini Enterprise Agent Platform, which offers access to more than 200 models, including Gemini 3.1 Pro and Anthropic’s Claude Opus 4.7. CEO Sundar Pichai noted that paid Gemini Enterprise users grew 40 percent quarter-over-quarter in Q1 2026, and roughly 75 percent of Google Cloud customers now use the company’s AI products.

The Internal Coding Problem

Behind the polished Las Vegas presentations, however, a more uncomfortable narrative is unfolding. According to Bloomberg, senior Google managers are alarmed by the company’s position in the AI-assisted developer tools market. Anthropic’s Claude model is gaining traction with enterprises — and reportedly even within Google’s own DeepMind team, where employees have been using the external tool despite internal policies prohibiting it.

The structural issue is fragmentation. Google’s coding capabilities are scattered across more than six products under different brand names, diluting their collective impact. In response, DeepMind has formed a specialist team under engineer Sebastian Borgeaud focused on complex, long-horizon programming tasks — such as writing new software from scratch. Co-founder Sergey Brin and DeepMind CTO Koray Kavukcuoglu are directly involved. Brin’s internal memo was blunt: “The final battlefield in the AI competition is agents.”

Should investors sell immediately? Or is it worth buying Alphabet?

The irony is that Google’s own internal AI usage is surging. AI now generates 75 percent of all new code at Google, up from 50 percent last autumn. Over the direct API, the company’s models process more than 16 billion tokens per minute, compared to 10 billion in the prior quarter.

The $185 Billion Question

All of this sets the stage for Alphabet’s Q1 2026 earnings release on April 29 after U.S. markets close. The analyst consensus calls for earnings per share of $2.68 on revenue of roughly $107 billion. But the real focus will be on whether the company can justify its breakneck investment pace.

Alphabet has guided for capital expenditures of $175 billion to $185 billion for the full year 2026 — nearly double the $91.4 billion spent in 2025. Rising depreciation from new data centers is squeezing margins, and some analysts project negative free cash flow for 2026. Morningstar rates the stock as fairly valued with three out of five stars and a price target of $340.

Wall Street’s reaction to the Cloud Next announcements was measured. UBS raised its price target to $375 on April 23 but maintained a “Neutral” rating, with analyst Stephen Ju praising the conference signals while warning of historically elevated valuations. JPMorgan holds firm at “Buy” with a $370 target.

Alphabet shares trade at roughly €288, just below their 52-week high and more than double their level a year ago. The market has already priced in the infrastructure narrative. The earnings report will need to show whether that spending is translating into measurable cloud revenue momentum — and whether the internal coding anxiety is a manageable distraction or a deeper strategic vulnerability.