Apple Gains Full Access to Google's Gemini, Accelerates On-Device AI Model Development with Distillation Technology

marsbitPublished on 2026-03-27Last updated on 2026-03-27

Abstract

Apple has secured full access to Google's Gemini model, aiming to accelerate the development of its on-device lightweight AI systems using advanced data distillation techniques. The company will utilize Gemini’s high-quality answers and chain-of-thought reasoning as training data to “feed” its own smaller, proprietary models. This approach, known as model distillation, enables compact models to achieve reasoning capabilities comparable to top-tier large models while maintaining computational efficiency. Although Gemini was originally designed for chatbots and enterprise applications—differing from Apple’s system-level integration vision for Siri—this collaboration significantly addresses Apple's need for high-quality synthetic data. In parallel, Apple continues its in-house development efforts through its Apple Foundation Models team. New AI features leveraging this distilled technology are expected to debut at Apple’s Worldwide Developers Conference (WWDC) in June. This partnership highlights a shift in the AI industry from pure computing power competition toward more efficient training strategies. By investing in access to leading model capabilities to enhance its edge computing advantages, Apple illustrates the ongoing balance between general-purpose large models and private on-device AI. This move also signals a future where edge devices will possess stronger local inference and complex task-handling abilities, further advancing the democratization of AI.

Recently, Apple has obtained extensive access to Google's Gemini model, aiming to accelerate the development of its lightweight on-device artificial intelligence through advanced data distillation technology.

According to related reports, Apple currently has full access to the Gemini model within its data centers. The core of this strategic move is to use the high-quality answers and logical reasoning chain records generated by Gemini as training data to "feed" Apple's self-developed small models. This "model distillation" approach, where large models guide the training of small models, enables the lightweight versions to maintain efficient computation while possessing logical processing capabilities similar to those of top-tier large models.

Although Gemini was initially designed for chatbots and enterprise-level applications, differing from Apple's deep system-level planning for Siri in terms of product logic, this collaboration significantly fills the gap in Apple's access to high-quality synthetic data. At the same time, Apple has not abandoned its self-development path; its Apple Foundation Models team is simultaneously advancing the in-house development of underlying models. It is expected that these new-generation AI features, incorporating distillation technology, will be showcased at the upcoming Apple Worldwide Developers Conference (WWDC) in June.

This collaboration marks a shift in the AI industry from pure computing power competition to more efficient training strategy competition. Apple's choice to "pay for data," by absorbing the capabilities of top-tier models to strengthen its edge computing advantage, not only reflects the game and balance between tech giants in general-purpose large models and private on-device AI but also预示着 that on-device equipment in the future will possess stronger local reasoning and complex task processing capabilities, further advancing the process of AI democratization.

Related Questions

QWhat is the core purpose of Apple gaining full access to Google's Gemini model?

AThe core purpose is to use Gemini's high-quality answers and chain-of-thought reasoning data to train Apple's own smaller, on-device AI models through a process called model distillation.

QHow does the 'model distillation' technique mentioned in the article work?

AModel distillation works by using a large, powerful model (like Gemini) to generate high-quality training data and logical reasoning traces, which are then used to 'teach' and train a smaller, more efficient model to achieve similar capabilities.

QWhat gap does this collaboration with Google help Apple fill?

AThis collaboration significantly fills Apple's gap in obtaining high-quality synthetic data for training its AI models.

QWhat is the name of Apple's team that is continuing its own foundational model research?

AThe team is called the Apple Foundation Models team.

QWhat industry shift does this Apple-Google collaboration signify according to the article?

AIt signifies a shift in the AI industry from pure computing power competition towards competition over more efficient training strategies.

Related Reads

Can DeepSeek Save China One Trillion Dollars?

"DeepSeek and the $1 Trillion Infrastructure Question" The article examines whether DeepSeek's AI optimization breakthroughs could potentially save China $1 trillion in future AI infrastructure costs. The analysis begins with Nvidia's upcoming Vera Rubin AI platform, costing ~$7.8 million, where memory (HBM4/LPDDR5X) constitutes $2 million—a 435% cost increase in one year, highlighting how AI hardware spending is shifting toward expensive memory components. DeepSeek's approach works in the opposite direction. Through three key technical innovations showcased in DeepSeek V4, the company dramatically improves hardware efficiency: 1. **Memory Compression (MLA)**: Re-engineers the attention mechanism to compress long-context memory (KV Cache) by over 90%, drastically reducing expensive HBM usage. 2. **Selective Activation (MoE)**: Employs Mixture-of-Experts architecture where only a small fraction of parameters (e.g., 49B out of 1.6T in V4-Pro) are activated per token, allowing most parameters to reside in cheaper memory/SSD. 3. **Computation Caching**: Reuses previously computed results via cache hits, replacing expensive GPU computations with cheap memory reads. Combined, these optimizations allow the same hardware to produce approximately 4x more tokens, effectively reducing required hardware investment by 75%. DeepSeek's pricing reflects this: a 10-billion token workload costs ~$522 monthly versus ~$9,000-$10,000 for competitors. The $1 trillion savings projection stems from McKinsey's estimate that global AI infrastructure will require ~$5.2 trillion investment by 2030. As China's daily token consumption grows toward quadrillions, even marginal efficiency gains scale massively. With a conservative 4x throughput improvement, China could avoid building tens of thousands of AI data centers equivalent to ~7 trillion RMB ($1 trillion) in saved investment. Critically, this strategy shifts dependency from scarce, expensive GPU/HBM—where China lags—toward more accessible storage, caching, and systems engineering where domestic suppliers like CXMT are gaining strength. Rather than "replacing Nvidia," DeepSeek rebalances AI's value chain away from monolithic hardware dependency. Ultimately, DeepSeek's technical breakthroughs could lower the barrier to AI adoption across Chinese industries by making advanced capabilities affordable at scale—transforming who can access next-generation AI.

marsbit47m ago

Can DeepSeek Save China One Trillion Dollars?

marsbit47m ago

Overturning the Mainstream Approach to Hallucinations: Metacognition is the New Solution for Large Models to Break the Hallucination Barrier

This paper, "Hallucinations Undermine Trust; Metacognition is a Way Forward," proposes a paradigm shift in combating AI hallucination. It argues that the current mainstream approaches—striving for omniscience by scaling data/models or having AI abstain from uncertain answers—are fundamentally flawed. The former has inevitable knowledge gaps, while the latter imposes a crippling "utility tax," requiring the rejection of many correct answers to achieve high accuracy, due to models' poor "discrimination" (the ability to distinguish correct from incorrect answers internally). The core contribution is redefining hallucination not as "being wrong," but as "expressing false information with unwarranted certainty." The proposed solution is **Faithful Uncertainty** or **Metacognition**: enabling AI to accurately perceive its internal uncertainty and honestly express it in its language (e.g., using hedging phrases when unsure). This creates a more reliable assistant that provides useful information while signaling its confidence, minimizing harm from errors. The paper emphasizes that metacognition is critical for the era of AI Agents. Without it, Agents cannot intelligently decide when to use tools like search engines, leading to inefficiency and misuse. Key implementation challenges are highlighted: the "bootstrapping paradox" of training with static uncertainty data, the "alignment distortion signal" where human preference training suppresses internal uncertainty cues, and the difficulty of causally evaluating true metacognition vs. its superficial imitation. The paper concludes that the goal should not be an infallible AI, but one that is honest about the limits of its knowledge, thereby building user trust through transparent communication of its certainty.

marsbit52m ago

Overturning the Mainstream Approach to Hallucinations: Metacognition is the New Solution for Large Models to Break the Hallucination Barrier

marsbit52m ago

Hedge by Buying Gold and Oil, Chase Soaring Returns with AI. ‘Dated’ Bitcoin Enters a Bear Market

Bitcoin has recently declined, hitting a two-month low near $66,123, while Ethereum fell to a three-month low around $1,837. Analysts suggest the drop is not merely due to factors like ETF outflows or MicroStrategy's selling but reflects a deeper issue: Bitcoin is losing a broader asset competition. In a near-zero interest rate environment, Bitcoin previously thrived as an outlet for investor dissatisfaction with inflation and limited options. However, the market landscape has shifted. Bitcoin now occupies an "awkward middle ground," facing competition on three fronts. For inflation hedging, investors prefer gold, energy stocks, and commodity producers—assets with tangible backing and clearer pricing power. For growth exposure, AI-related companies with actual revenues and profits are more attractive. Even within crypto, investors can choose stablecoins, exchanges, or infrastructure firms tied directly to adoption, offering clearer business models and leverage. Thus, Bitcoin is no longer the top choice for hedging, growth, or crypto exposure. This shift is evident in market reactions: despite recent warnings about persistent inflation from a Fed official, Bitcoin did not rally as it might have in the past. Instead, capital flowed to assets with direct commodity or energy exposure. The recent ETF outflows and MicroStrategy sales are symptoms, not causes, of this new reality. Investors are becoming more selective, demanding clearer value propositions beyond mere scarcity. The emerging bear case for Bitcoin is not about it being a bubble or failed technology, but that scarcity alone is no longer sufficient.

华尔街日报55m ago

Hedge by Buying Gold and Oil, Chase Soaring Returns with AI. ‘Dated’ Bitcoin Enters a Bear Market

华尔街日报55m ago

Trading

Spot
Futures
活动图片