The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →AI looked like 2024’s biggest technology opportunity because it could trigger spending across the entire enterprise stack—not just on chatbots or model licenses. The executives featured in CRN’s 2024 CEO Outlook were effectively betting on demand for accelerators, servers, cloud platforms, data systems, security, AI-enabled PCs, consulting and data-center power. The source headline says 10 CEOs, but its article names 11: Lisa Su, Chuck Robbins, Michael Dell, Thomas Kurian, Enrique Lores, Antonio Neri, Sheila Rohra, Arvind Krishna, Pat Gelsinger, Yuanqing Yang and Giordano Albertazzi.
That distinction matters. These were technology-channel executives selected for CRN’s outlook project, not an objective ranking of the world’s largest technology companies. Their comments describe a large expected market opportunity in late 2023 and early 2024, not proof that every forecast became revenue or that every customer achieved a return.
The shared 2024 thesis: AI would expand technology budgets
Generative AI became unusually visible after ChatGPT and text-to-image tools reached mass audiences. In 2023, many companies experimented with demonstrations and proofs of concept. The executives told CRN they expected 2024 to be the year organizations began connecting those experiments to production systems and measurable business workflows.
The opportunity was broader than building a frontier model. A production deployment can require compute, storage, networking, data integration, identity controls, security, monitoring, employee training, workflow redesign, legal review and continuing optimization. A single customer might therefore buy cloud capacity, servers, GPUs, software, consulting and managed services at the same time.
#1 Best Overall
| Value-chain layer | What customers may need | Companies represented in the outlook |
|---|---|---|
| Semiconductor compute | CPUs, GPUs, NPUs and other accelerators for training and inference | AMD, Intel |
| Servers and data infrastructure | AI-ready systems, storage, data platforms and deployment capacity | Dell, HPE, Lenovo, Hitachi Vantara |
| Networking and security | High-speed data movement, connectivity, governance and protection | Cisco, HPE |
| Cloud and enterprise platforms | Model access, analytics, development tools and production operations | Google Cloud, IBM |
| Devices | Local inference, privacy, latency and new PC experiences | AMD, HP, Intel, Lenovo |
| Facilities | Power, cooling and high-density data-center capacity | Vertiv |
| Services channel | Architecture, integration, governance, training and managed operations | All of the above |
For technology vendors, that cross-layer effect made AI attractive. For buyers, it created a harder question: which spending actually improves a defined business process, and which merely funds an impressive pilot?
What each executive saw in the opportunity
Lisa Su, AMD: compute for data centers and PCs
Su described demand at both ends of the market. AMD promoted Instinct MI300 accelerators for data-center workloads and neural-processing capabilities in Ryzen processors for AI-capable PCs. The strategic bet was to sell the compute used for both training and inference, rather than depend on software revenue alone. AMD’s performance and leadership descriptions are company claims, not independent benchmark conclusions.
Chuck Robbins, Cisco: readiness, networking and security
Cisco argued that organizations could have an AI strategy without being operationally prepared. Its own AI Readiness Index reported that 95% of organizations had an AI strategy in place or under development, while 14% said they were ready to fully integrate AI. Those are Cisco survey results, not a universal measurement of the market.
Cisco’s commercial opportunity therefore included high-performance networking, data-center connectivity, security, data governance, validated designs and consulting. Robbins also highlighted privacy, confidentiality, intellectual property, bias and human-rights issues—constraints that can determine whether a deployment is allowed to proceed.
Michael Dell, Dell Technologies: turning pilots into systems
Dell framed generative AI as a systems and services opportunity. Customers would need infrastructure planning, procurement, servers, storage, integration, deployment and optimization. Partners could earn revenue throughout that lifecycle. The practical interpretation is that AI gave organizations a reason to expand or refresh infrastructure, but the investment would only make sense if utilization and business value justified it.
Rank #2
Thomas Kurian, Google Cloud: production, not demonstrations
Kurian identified the key transition as moving from pilots and proofs of concept to implementations that solve specific business problems. Google Cloud’s opportunity included cloud infrastructure, data analytics, development platforms, agent-based applications, implementation partners, training and certification.
Google Cloud said its certified-partner count had increased 15-fold since 2018 and that consulting and systems-integrator partners had committed to training more than 150,000 people for Google Cloud AI. These are Google’s stated ecosystem figures. They also illustrate the expected bottleneck: people who can connect models to business data, workflows, permissions and production controls.
Enrique Lores, HP: the AI PC
HP presented AI PCs as a new category. Local processing could reduce latency, improve privacy and security, and potentially lower cloud usage for suitable tasks. HP forecast that AI PCs could double the category’s growth rate over three years; that is a company forecast, not an established market result.
The limitation is equally important. CRN later reported Lores describing adoption as gradual because applications needed to mature. A capable neural processor does not create demand by itself; software must use it, and buyers must value the resulting experience.
Antonio Neri, Hewlett Packard Enterprise: hybrid infrastructure
Neri argued that AI would require a hybrid approach spanning compute, storage, high-performance networking, edge systems, private cloud, security, software and machine-learning platforms. HPE’s GreenLake consumption model was part of that thesis.
Hybrid deployment can address data sovereignty, latency, security, existing systems and cost controls. It also increases operational complexity: organizations must manage multiple environments, data paths, security policies and capacity plans.
Sheila Rohra, Hitachi Vantara: context makes models useful
Rohra emphasized that a general-purpose model is not the same as business value. Useful systems need high-quality enterprise data, domain expertise, business context, interoperability, retrieval and permissions. Data integration and governance can be more difficult than selecting a model.
Free tools Windows power users keep installed
One-click scans. No signup required.
Arvind Krishna, IBM: measurable enterprise use cases
Krishna cited a projection that AI could unlock $16 trillion in value by 2030. That is a forecast attributed through IBM, not realized economic output.
IBM highlighted three early enterprise areas:
- Code modernization: updating legacy applications and documentation.
- Customer service: assisting agents and answering routine questions.
- Digital labor: supporting HR, IT, procurement and recruiting workflows.
IBM also cited a 20% increase in customer loyalty for NatWest’s AI mortgage tool and an expected 75% reduction in recruiting time for a Silver Egg Technology proof of concept. These are individual company-reported results. They do not establish a general benchmark, and the recruiting figure was an expected proof-of-concept outcome rather than a confirmed production result.
Pat Gelsinger, Intel: AI everywhere, including the PC
Gelsinger described real-time transcription, translation, contextual assistance, local personal assistants and small language models running on devices. Cloud AI can support larger models and centralized governance, but it carries ongoing compute and network costs. On-device AI can reduce latency, work offline and keep some data local; memory, power, thermals and model size limit what it can do.
Yuanqing Yang, Lenovo: pocket-to-cloud solutions
Yang described an AI opportunity spanning generated content, large language models, devices, infrastructure, services and industry-specific transformation. Lenovo’s more useful warning was directed at partners: do not sell capabilities you cannot securely deliver and support. AI services require skills in data, security, evaluation, integration and operations—not just a sales relationship with a model provider.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesGiordano Albertazzi, Vertiv: power and cooling
Vertiv showed why AI is not solely a software story. High-density accelerators require more power, thermal management, capacity planning and facility work. Even a successful application can be constrained by rack capacity, electricity availability, cooling design or construction timelines.
Where customers must choose between competing deployment paths
Cloud, on-premises or hybrid
| Approach | Usually stronger when | Main trade-off |
|---|---|---|
| Cloud | Workloads are experimental, demand is uncertain, managed services are valuable, or rapid model access matters | Lower upfront commitment but potentially volatile usage, storage and data-transfer costs |
| On-premises | Data must remain local, utilization is predictable, latency is critical or control is paramount | Requires capital, facilities, operations staff, refreshes and capacity planning |
| Hybrid | Different data, latency, sovereignty or cost requirements apply to different workloads | More integration, monitoring and policy complexity |
Google Cloud describes its services as pay-as-you-go, with product-specific pricing and $300 in credits for new customers; current details are on its pricing page. A credit or hourly rate does not establish total cost. Buyers must include GPUs, instances, storage, networking, data movement, staff, security and support.
Frontier models or smaller specialists
Larger models can handle broader and more ambiguous tasks. Smaller or specialized models can lower inference cost and latency, run locally more easily and offer more predictable behavior for narrow workflows. The largest available model is therefore not automatically the best commercial choice.
AI PC or cloud AI
Local AI is most compelling when privacy, offline operation or latency matters and the model fits the device. Cloud AI remains stronger when users need large models, centralized controls, shared enterprise data or rapidly changing capabilities. HP’s gradual-adoption warning shows why hardware availability should not be mistaken for immediate demand.
Best Value
Why partners were central to the forecast
Channel partners were positioned to do the work vendors cannot productize completely:
- Assess data, security and infrastructure readiness.
- Choose models, accelerators and deployment locations.
- Integrate AI with applications and identity systems.
- Design evaluation, human-review and compliance controls.
- Train employees and redesign workflows.
- Monitor quality, usage, drift and cost in production.
- Optimize or migrate systems when economics change.
This is why the opportunity was “sell the operational capability to make AI useful,” not simply “sell AI.” A partner that cannot support security, reliability and governance can turn a promising deployment into a liability.
Risks that can erase the expected return
- Poor data: stale, duplicated or incomplete records produce confident but unreliable outputs.
- Pilot purgatory: demonstrations multiply without an accountable business owner or production integration.
- Security and privacy exposure: prompts, retrieval systems, logs and misconfigured permissions can disclose sensitive information.
- Intellectual-property uncertainty: training data, generated code, model outputs and customer content may raise licensing or ownership questions.
- Hallucinations and bias: plausible errors require evaluation and human review, especially in high-impact decisions.
- Hidden infrastructure costs: power, cooling, storage, networking, egress, monitoring and specialized staff may be absent from an initial business case.
- Underused hardware: an expensive AI server is uneconomical when demand is intermittent or models are poorly optimized.
- Vendor lock-in: proprietary APIs, data formats and accelerators can make migration costly.
- Overstated productivity: faster drafting or code generation may be offset by verification and rework.
- Regulatory exposure: sensitive uses may require auditability, geographic controls, documentation and human oversight.
How to test an AI opportunity before spending
- Define one workflow and baseline it. Record current cost, cycle time, error rate, service level and human effort.
- Classify the data. Identify confidential information, retention rules, geographic restrictions and access permissions.
- Compare deployment options. Price cloud, on-premises and hybrid designs using expected utilization—not peak marketing capacity.
- Choose the smallest adequate model. Test quality, latency and cost against a larger alternative.
- Design controls before launch. Specify evaluation sets, human review, logging, incident response and ownership.
- Measure production economics. Include model calls, storage, data transfer, infrastructure, integration, support and rework.
- Set a stop rule. End or redesign the project if it does not improve the baseline at an acceptable risk and cost.
What the 2024 outlook really meant
The executives were not making the same bet. AMD and Intel emphasized compute, HP and Lenovo devices, Dell and HPE systems, Cisco networking and security, Google Cloud and IBM production platforms, Hitachi Vantara data context, and Vertiv facilities. Their common prediction was that AI would force organizations to buy capabilities across many layers at once.
That created a genuine opportunity for vendors and partners, but it did not guarantee customer value. The durable test was whether a deployment solved a defined problem after accounting for data preparation, infrastructure, governance, human oversight and ongoing operating cost.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
