• Production forecasts for Alphabet's custom AI chips have been sharply increased, with Morgan Stanley now expecting approximately 5 million Tensor Processing Units (TPUs) in 2027 and 7 million in 2028, up from prior estimates of 3 million and 3.2 million respectively.
  • The financial impact is significant, with analysts estimating every 500,000 TPUs sold could generate roughly $13 billion in revenue and $0.40 per share in earnings for Alphabet.
  • The scale of the ramp-up signals a strategic shift from internal use to a broader commercial push, with Google projecting production of 12 million TPUs over the next two years, a figure that far outpaces the 7.9 million produced over the previous four.

Supply chain checks are pointing to a far more aggressive production schedule for Alphabet's Tensor Processing Units than previously anticipated, a move that could translate into tens of billions in new revenue, according to a recent analysis from Morgan Stanley. The bank's Asia semiconductor team has significantly raised its forecasts, citing strengthened signals that Google is preparing for a massive increase in deployments, both for its own cloud services and for external sales.

"The magnitude of the increase is what's telling," said one person familiar with the semiconductor supply chain, who noted the revised figures suggest a concerted effort to scale commercial availability. The projections imply Google is building capacity for a total of 12 million TPUs over the 2027-2028 period, a clear signal that management views the custom silicon as a major growth pillar beyond internal infrastructure.

Morgan Stanley's model suggests the financial leverage is substantial. Their analysis indicates that for every incremental 500,000 TPUs sold, Alphabet could see a revenue boost of around $13 billion and an earnings-per-share increase of approximately $0.40. While a significant portion of the new production will likely fuel Google Cloud's own AI services, the sheer scale of the build-out has analysts questioning whether a direct-to-market hardware strategy is gaining momentum.

This comes amid a shifting competitive landscape. Industry sources have recently indicated that Meta Platforms is in advanced discussions to purchase billions of dollars worth of Google's TPUs, a potential deal that would mark a significant diversification away from its heavy reliance on Nvidia. While neither company has commented publicly, such an agreement would validate the commercial appeal of Google's custom silicon and could generate revenue equivalent to a meaningful slice of Nvidia's current data center business.

Google has also committed to delivering approximately one million TPUs to AI lab Anthropic and is actively promoting its software stack to other AI startups. The company's latest Ironwood TPU, which became generally available in late 2025, offers nearly double the performance-per-watt of its predecessor, a key selling point for cost-conscious enterprises running large-scale AI workloads.

The improved outlook for Alphabet's silicon business also had a ripple effect, helping to justify Morgan Stanley's recent upgrade of chip designer MediaTek to Overweight. The connection underscores the broader supply chain implications of Google's ambitious plans.

Efforts to reach Alphabet for additional comment on its external TPU sales strategy were not immediately successful. For now, the production forecasts serve as the clearest indicator yet that the company is preparing to compete more directly in the high-stakes market for AI compute, a sector long dominated by a single player.

Correction: An earlier version of this article misstated the performance comparison for the Ironwood TPU. It delivers nearly 2x better performance-per-watt compared to prior TPU generations, not 2x better overall performance.