The AI Landscape: US Companies, Global Competition, Inference, Hardware & Policy
Question — of 50
Assessment

The AI Landscape: US Companies, Global Competition, Inference, Hardware & Policy

Answer all 50 questions to test your understanding. Your results will be shown at the end.

50
Questions
5
Objectives
80%
Pass mark
1
Select All That Apply

A team compares a US foundation model company with a non-US cloud platform and gives both the same overall score because both 'have strong models.' Which comparison dimensions are being underweighted? Select all that apply.

  • How each company captures value, such as API revenue versus cloud pull-through or ads
  • How exposed each company is to policy constraints tied to geography and export controls
  • Whether each company can reliably secure compute or relies on external hardware partners
  • Whether both companies publish benchmark leaderboards using the same format
✓ Answer recorded
2
Single Choice

A PM says, 'Company NovaMind must be the stronger competitor because its flagship model tops the latest coding benchmark.' What is the best critique of that reasoning?

  • ABenchmark wins matter only for research companies, not commercial firms
  • BA single benchmark can stand in for model strategy, but not for market reach
  • CCompetitive strength should be judged across multiple criteria, not model scores alone
  • DCoding benchmarks are unreliable because they can never predict enterprise adoption
✓ Answer recorded
3
Match the Pairs

Match each comparison criterion to the question it is mainly intended to answer.

Concept
Business model
Model strategy
Hardware access
Market reach
Policy exposure
Definition
Can the company obtain enough compute at the right performance and cost?
How does the company turn AI capability into revenue or strategic leverage?
How vulnerable is the company to regulation, export controls, and national-security rules?
How effectively can the company distribute products across regions and customer segments?
Is the company emphasizing frontier models, open models, vertical tuning, or platform integration?
0 of 5 pairs matched
✓ Answer recorded
4
Assertion–Reason

Assertion: A company with weaker base models can still be competitively strong in AI. Reason: Distribution, installed customer base, and cloud bundling can outweigh small model-quality gaps in many markets.

  • ABoth A and R are true and R is the correct explanation of A
  • BBoth A and R are true but R is NOT the correct explanation of A
  • CA is true but R is false
  • DA is false but R is true
✓ Answer recorded
5
Single Choice

A comparison matrix gives a state-backed non-US firm and a venture-backed US startup equal scores for 'policy exposure' because both sell software globally. Which factor most likely makes that scoring too simplistic?

  • ABoth companies will eventually face the same open-source competition
  • BPolicy exposure depends mainly on whether their websites are translated well
  • CState ties, export rules, and procurement preferences can create very different policy risk profiles
  • DPolicy exposure matters only after a company reaches profitability
✓ Answer recorded
6
Select All That Apply

You are scoring two competitors: AtlasAI (US, sells model APIs directly) and Meridian Cloud (non-US, embeds AI into a large enterprise software suite). Which observations would justify giving Meridian Cloud a higher market-reach score even if AtlasAI's model performs slightly better? Select all that apply.

  • Meridian already has procurement relationships with large enterprise buyers in multiple regions
  • Meridian can cross-sell AI features into an installed software base without winning a greenfield model decision each time
  • AtlasAI publishes more technical papers, which means its sales motion will be weaker
  • Meridian's distribution channels can reduce customer acquisition friction despite modest model-quality gaps
✓ Answer recorded
7
Fill in the Blank

If a comparison matrix rates companies only on model performance and ignores how they actually make money, it is missing the ___ criterion.

If a comparison matrix rates companies only on model performance and ignores how they actually make money, it is missing the   criterion.
✓ Answer recorded
8
Single Choice

Based on the chart, which company is most likely to deserve the highest overall score if your weighting prioritizes hardware access and market reach over raw benchmark rank?

AI Company Comparison Matrix (Selected Firms) Scores: 1 = low, 5 = high Company Benchmark Rank* Hardware Access Market Reach Policy Exposure Model / Business Strategy Fit Microsoft 4 5 5 3 5 Google 5 5 5 4 4 Amazon 3 5 5 3 4 OpenAI 5 4 4 4 5 Alibaba 3 4 4 5 4 DeepMind UK 5 4 3 4 4 Mistral EU 4 3 2 3 4 *Lower benchmark rank is not shown; score is normalized so higher is better across all columns.
  • ARedwood AI
  • BBluePeak Cloud
  • CEuroStack Models
  • DPacific Logic
✓ Answer recorded
9
Select All That Apply

A scored comparison matrix shows FrontierFox and GlobalWeave tied. Which missing evidence would most likely break the tie in a defensible way? Select all that apply.

  • Whether one firm depends on a single cloud provider for training and inference capacity
  • Whether one firm reaches customers mainly through direct API sales while the other sells through existing enterprise channels
  • Whether one CEO gives more interviews about AGI timelines
  • Whether one firm faces materially different export-control exposure due to geography and supply chain
✓ Answer recorded
10
Sequencing

Arrange the steps in the best order for building a comparison matrix that avoids overvaluing model hype.

⠿Score each company against the same explicit criteria
⠿Collect evidence on each company for every criterion
⠿Write a short synthesis explaining why the scores differ
⠿Choose the comparison criteria and weighting
✓ Answer recorded
11
Single Choice

A product team says, 'Our chatbot is too slow, so we need more training data.' Which diagnosis is most likely correct?

  • ASlow responses are usually an inference issue such as serving latency, concurrency, or model size
  • BMore training data is the standard fix for response-time problems in production
  • CLatency mostly comes from the optimizer used during pretraining
  • DAny user-visible issue should first be treated as a training bottleneck
✓ Answer recorded
12
Select All That Apply

Your team is investigating why production costs spike when daily active users double. Which factors are primarily inference-related rather than training-related? Select all that apply.

  • Average output length per request increases after a new feature launch
  • The company wants to refresh pretraining data every quarter
  • Peak-time concurrency exceeds what the serving cluster can handle
  • The tokenizer was chosen before the original base model was trained
✓ Answer recorded
13
Match the Pairs

Match each factor to the side it most directly affects first: training or inference.

Concept
Time to serve 500 concurrent user requests
Cost of a multi-week run to create a new base model
Tokens generated per second for end users
Data curation for a fresh model pretraining corpus
Autoscaling policy for production traffic spikes
Definition
Inference
Inference
Training
Inference
Training
0 of 5 pairs matched
✓ Answer recorded
14
Assertion–Reason

Assertion: Increasing context length can raise inference cost even if the model weights never change. Reason: During inference, the system must process more tokens per request, which affects memory use and latency.

  • ABoth A and R are true and R is the correct explanation of A
  • BBoth A and R are true but R is NOT the correct explanation of A
  • CA is true but R is false
  • DA is false but R is true
✓ Answer recorded
15
Fill in the Blank

When a deployed model gets expensive mainly because many users are querying it at once, the dominant pressure is usually on ___ infrastructure.

When a deployed model gets expensive mainly because many users are querying it at once, the dominant pressure is usually on   infrastructure.
✓ Answer recorded
16
Single Choice

Based on the chart, which change most strongly suggests an inference bottleneck rather than a training bottleneck?

AI Service Metrics Before vs After Product Update Quarter A Quarter B Weekly model retrains: 2 Weekly model retrains: 2 Training job queue: 3 hrs Training job queue: 3.2 hrs Avg user latency: 240 ms Avg user latency: 620 ms Daily user requests: 1.2M Daily user requests: 3.8M GPU use at peak: 54% GPU use at peak: 91% Earlier operating period Later operating period Metric labels compare training cadence, training queue, latency, demand, and peak serving utilization
  • ATraining run duration increased from 9 to 14 days after adding more data
  • BUser response latency jumped from 700 ms to 2,400 ms during traffic peaks while model weights stayed unchanged
  • CThe fine-tuning dataset expanded from 2 million to 5 million examples
  • DExperiment costs rose after increasing epochs in a retraining pipeline
✓ Answer recorded
17
Select All That Apply

A team is debating whether to solve a cost problem by shrinking the model or by changing serving architecture. Which observations point more toward an inference-side fix? Select all that apply.

  • The same model is cheap overnight but expensive during midday demand spikes
  • Longer generated answers correlate with higher per-request cost
  • Pretraining failed to converge at the planned learning rate
  • Most complaints involve queueing delay rather than incorrect answers
✓ Answer recorded
18
Single Choice

A startup wants to launch in three regions next month. Which concern is most clearly an inference/deployment constraint rather than a training constraint?

  • AWhether the pretraining corpus was diverse enough across languages
  • BWhether each region has enough serving capacity close to users to meet latency targets
  • CWhether the optimizer used for fine-tuning can improve multilingual accuracy
  • DWhether the RLHF preference dataset should be expanded before release
✓ Answer recorded
19
Fill in the Blank

If p95 response time worsens only when many requests arrive at once, the likely issue is request ___ rather than model retraining.

If p95 response time worsens only when many requests arrive at once, the likely issue is request   rather than model retraining.
✓ Answer recorded
20
Select All That Apply

Based on the chart, which statements are supported and help distinguish training from inference constraints? Select all that apply.

AI Scenario Factors: Training vs Inference Constraints Example deployment for a global AI product; scores reflect relative pressure (1=low, 10=high) Training Inference Factor Constraint score by scenario side Compute cost Latency sensitivity Throughput at scale Memory / model size User traffic variability Cross-region deployment Data policy / governance 10 9 8 7 6 5 4 3 2 1 1 2 3 4 5 6 7 8 9 10 8 3 5 7 2 4 6 4 8 7 5 8 7 6
  • DriftAI's biggest current pain is likely inference, because traffic growth sharply increases monthly spend without any retraining event
  • LumaForge's main problem is definitely training capacity, since its major cost event is a long training run rather than user-driven serving growth
  • NorthRiver may have a training-side issue, because capability improves only after a costly new run rather than after serving changes
  • SableChat's latency problem can be assumed to be training-related because larger models are always slower to train
✓ Answer recorded
21
Single Choice

A team assumes that buying the fastest GPUs is always the best deployment choice. Which scenario most strongly challenges that assumption?

  • AA stable high-volume inference workload where a custom accelerator meets latency targets at much lower cost per query
  • BA research lab experimenting with new model architectures every week
  • CA startup that has no production users yet but wants optionality
  • DA prototyping project where engineers need broad framework compatibility
✓ Answer recorded
22
Select All That Apply

Which considerations legitimately favor cloud-based AI infrastructure over owning hardware for an early-stage deployment? Select all that apply.

  • Demand is uncertain, so paying for elasticity is worth more than maximizing utilization
  • The workload changes often, so avoiding long hardware commitments reduces architectural lock-in
  • Cloud is always lower latency than on-prem because providers own the network
  • The team wants to get to production before it has procurement and datacenter operations in place
✓ Answer recorded
23
Match the Pairs

Match each deployment situation to the hardware choice it most naturally favors first.

Concept
Rapid model experimentation across changing frameworks
Steady high-volume inference with a narrow model family
Launching quickly with uncertain traffic and no datacenter team
Needing mature tooling and the broadest ecosystem support
Avoiding upfront capex while preserving burst capacity
Definition
Cloud-based AI infrastructure
Cloud-based AI infrastructure
GPUs
GPUs
Custom accelerators
0 of 5 pairs matched
✓ Answer recorded
24
Assertion–Reason

Assertion: A custom accelerator can be the best choice for inference even if its software ecosystem is less flexible than GPUs. Reason: If the workload is stable, efficiency gains can outweigh reduced generality.

  • ABoth A and R are true and R is the correct explanation of A
  • BBoth A and R are true but R is NOT the correct explanation of A
  • CA is true but R is false
  • DA is false but R is true
✓ Answer recorded
25
Single Choice

Based on the chart, which option best meets the deployment target of under 120 ms latency, at least 2,500 requests/sec, and under $300k monthly cost?

AI Inference Deployment Options Option Latency Throughput Monthly Cost Region GPU Cluster A 95 ms 2,300 req/s $285k US-East GPU Cluster B 118 ms 2,700 req/s $312k US-West Custom ASIC 105 ms 2,900 req/s $276k US-Central Cloud TPU Pool 122 ms 3,400 req/s $268k EU-West Hybrid Burst 110 ms 2,550 req/s $298k Multi-region Edge Pods 88 ms 2,480 req/s $241k APAC Deployment target: latency < 120 ms, throughput ≥ 2,500 req/s, monthly cost < $300k
  • AAster GPU cluster
  • BBrio custom accelerator
  • CCirrus cloud burst setup
  • DDune legacy GPU pool
✓ Answer recorded
26
Fill in the Blank

When usage varies widely day to day, paying for cloud ___ can be more rational than owning enough hardware for peak load.

When usage varies widely day to day, paying for cloud   can be more rational than owning enough hardware for peak load.
✓ Answer recorded
27
Select All That Apply

A CFO wants to move a mature recommendation model off premium GPUs. Which facts would make a custom accelerator recommendation more defensible? Select all that apply.

  • Traffic is high and predictable across most days
  • The model architecture changes every sprint as researchers test novel layers
  • Per-query margins are thin, so small efficiency gains matter at scale
  • The serving software already depends on highly GPU-specific kernels that cannot be adapted in time
✓ Answer recorded
28
Single Choice

A team chooses an on-prem GPU cluster because the sticker price per accelerator is lower than cloud. What important comparison are they most likely missing?

  • AWhether open-source models are improving quickly enough
  • BTotal cost including utilization, power, networking, operations, and overprovisioning
  • CWhether the GPUs can also be used for employee laptops later
  • DThe exact number of benchmark papers published by the hardware vendor
✓ Answer recorded
29
Fill in the Blank

If a hardware choice keeps average cost low only because machines sit busy almost all the time, the decision depends heavily on high ___ rates.

If a hardware choice keeps average cost low only because machines sit busy almost all the time, the decision depends heavily on high   rates.
✓ Answer recorded
30
Select All That Apply

Based on the chart, which conclusions best support a recommendation to stay on GPUs for now? Select all that apply.

AI Deployment Hardware Comparison Illustrative options for a medium-scale inference deployment Option Monthly Cost Latency Throughput Deployment Time GPUs $138k 48 ms 8,600 req/hr 2 weeks Custom ASIC $118k 36 ms 10,400 req/hr 14 weeks FPGA $126k 42 ms 7,900 req/hr 9 weeks Cloud TPU $132k 40 ms 9,800 req/hr 6 weeks CPU Cluster $110k 74 ms 5,400 req/hr 1 week Axis label (table columns): lower cost and latency are better; higher throughput is better. Deployment time reflects estimated time to put the option into production at the same scale.
  • Cobalt GPU has the best framework coverage and meets latency, though it is not the cheapest
  • Nimbus ASIC is cheapest per 1M tokens, but misses the team's latency target for interactive use
  • Orbit Cloud Burst should always be preferred because managed services eliminate all lock-in
  • The workload's architecture is changing monthly, which raises the value of software flexibility
✓ Answer recorded
31
Single Choice

An analyst says, 'Policy is a background issue; the best model will win anyway.' Which response is strongest?

  • AThat is mostly true because policy affects only government buyers
  • BPolicy can shape compute access, market access, and cost structure, so it can materially alter who can compete
  • CPolicy matters only for semiconductor manufacturers, not AI companies
  • DIf a company is global, policy exposure averages out across regions
✓ Answer recorded
32
Select All That Apply

Which policy factors could directly weaken an AI company's competitive position even if customer demand remains strong? Select all that apply.

  • Export controls that limit access to advanced chips or manufacturing equipment
  • Data-localization rules that raise deployment complexity across regions
  • A founder's preference for publishing longer blog posts about alignment
  • Public funding programs that shift talent, compute, or procurement toward rivals
✓ Answer recorded
33
Match the Pairs

Match each policy factor to the most plausible first-order competitive effect.

Concept
Export controls on advanced chips
Strict data-localization requirements
Public funding for domestic AI infrastructure
Government procurement preferences
Model liability or safety reporting rules
Definition
Improved compute access for favored local firms
Reduced access to high-end compute and slower scaling
Distribution advantage for eligible suppliers
Higher cost and complexity for cross-border product deployment
Higher compliance burden and slower release cycles
0 of 5 pairs matched
✓ Answer recorded
34
Assertion–Reason

Assertion: Public funding can strengthen a company's AI position even when the company itself is not directly subsidized. Reason: Government spending on shared compute, universities, and semiconductor capacity can improve the surrounding ecosystem the company depends on.

  • ABoth A and R are true and R is the correct explanation of A
  • BBoth A and R are true but R is NOT the correct explanation of A
  • CA is true but R is false
  • DA is false but R is true
✓ Answer recorded
35
Single Choice

Based on the chart, which policy factor appears most likely to have the highest immediate impact on Quartz AI's competitive position?

Quartz AI: Estimated Policy Impact by Factor 0 10 20 30 40 Estimated immediate impact score 36 28 23 17 31 Export controls Public funding Data regulation Energy-use rules Gov't procurement Policy factor Scale: 0–40
  • AA future tax credit for university research partnerships
  • BNew export controls on the exact accelerator class Quartz depends on for scaling
  • CA proposed transparency rule for labeling AI-generated images
  • DA municipal grant for local startup coworking space
✓ Answer recorded
36
Fill in the Blank

When governments restrict shipment of advanced chips to certain markets, the company faces a form of policy risk known as export ___.

When governments restrict shipment of advanced chips to certain markets, the company faces a form of policy risk known as export  .
✓ Answer recorded
37
Select All That Apply

An assessment ranks policy risks but uses only one news article as evidence. What would improve the ranking most? Select all that apply.

  • Triangulating with at least two credible sources such as official policy text, earnings calls, or industry analysis
  • Estimating which company functions are exposed, such as compute supply, market access, or compliance cost
  • Replacing evidence with the analyst's intuition, since policy moves too fast for careful sourcing
  • Separating immediate operational impact from longer-term strategic impact
✓ Answer recorded
38
Single Choice

A non-US AI company can still look stronger than a US rival on paper until one factor is considered. Which factor is most likely to change the ranking if the company relies on imported advanced accelerators from a restricted supply chain?

  • AEmployee satisfaction survey results
  • BExport-control exposure affecting future compute availability
  • CWhether its models are multilingual by default
  • DHow often it attends major AI conferences
✓ Answer recorded
39
Fill in the Blank

Rules requiring user data to remain within national borders create a ___ burden for firms operating one AI product across many countries.

Rules requiring user data to remain within national borders create a   burden for firms operating one AI product across many countries.
✓ Answer recorded
40
Select All That Apply

Based on the chart, which statements best justify ranking policy risks for Harbor Models? Select all that apply.

Harbor Models: Estimated Exposure to Policy Factors Policy factor Harbor revenue linked to factor Key input cost linked to factor Competitor benefit Harbor preparedness Export controls 18% 46% High Low Data regulation 37% 21% Medium Medium Public funding 9% 14% High High Safety licensing 24% 11% Low Medium Energy permits 12% 28% Medium Low Model liability 31% 8% Low Low Axis label: percentages show estimated share of Harbor's revenue or input cost exposed to each policy factor.
  • Export controls deserve a high ranking because Harbor sources most advanced chips from a now-restricted channel
  • Data localization deserves a medium-to-high ranking because 52% of Harbor revenue comes from cross-border deployments
  • Public funding should rank lowest because Harbor has limited direct exposure to subsidy programs relative to other policy risks
  • Procurement preferences deserve attention because Harbor is trying to enter regulated government-adjacent sectors
✓ Answer recorded
41
Single Choice

Which opening makes the strongest first sentence for a strategic briefing?

  • AUS firms should build AGI first because leadership on benchmarks guarantees market dominance
  • BThis briefing recommends how a US enterprise-AI provider can improve margin and resilience over 24 months despite global model competition and hardware-policy constraints
  • CAI is changing everything, so companies should move fast and be innovative
  • DGlobal competition is intense, and many firms are trying different strategies right now
✓ Answer recorded
42
Select All That Apply

A draft briefing says US firms should 'just train bigger frontier models than foreign rivals.' Which additions would make the strategy more credible and complete? Select all that apply.

  • Explain how the recommendation supports a specific business objective such as margin, adoption, or resilience
  • Include inference economics, such as serving cost and latency, not only model-training ambition
  • Assess hardware supply concentration and policy exposure that could block execution
  • Replace evidence with a stronger claim that scale always wins in platform markets
✓ Answer recorded
43
Match the Pairs

Match each briefing element to the role it should play in a strong recommendation.

Concept
Business objective
Company strategy
Inference economics
Hardware supply
Policy risk
Definition
Identifies external rules that could change access, cost, or market reach
Defines what success looks like, such as margin improvement or market expansion
Tests whether the offering can be delivered profitably at production scale
Explains where the firm will compete and how it will differentiate
Assesses whether the compute plan is feasible and resilient
0 of 5 pairs matched
✓ Answer recorded
44
Assertion–Reason

Assertion: A strategic briefing can recommend a smaller model portfolio over a larger one. Reason: If the firm's objective is profitable enterprise adoption, lower serving cost and easier deployment may matter more than chasing every frontier benchmark.

  • ABoth A and R are true and R is the correct explanation of A
  • BBoth A and R are true but R is NOT the correct explanation of A
  • CA is true but R is false
  • DA is false but R is true
✓ Answer recorded
45
Single Choice

Based on the chart, which recommendation best fits a briefing whose stated objective is to improve operating margin within 12 months while staying globally available?

AI Strategy Options: 12-Month Margin vs Global Availability Strategy option 12-mo margin Global reach HW supply risk Policy risk A. Premium training push -3 pts High High Med B. Inference efficiency +5 pts High Low Low C. Region lock-in +4 pts Low Med Low D. GPU stockpile -1 pt High Med Med E. Compliance-first rollout +1 pt Med Low Low Axis label: margin impact measured as estimated operating-margin change within 12 months. Global reach reflects ability to remain available across major markets; HW = hardware. Risk ratings are relative and fictional. Best choice depends on the stated objective, not any single column.
  • AIncrease training spend aggressively to chase the top research benchmark
  • BPrioritize model-routing and smaller serving models, diversify compute suppliers, and focus expansion on regions with lower compliance friction
  • CDelay all international launches until policy uncertainty disappears
  • DAcquire more premium GPUs first, then define the product strategy later
✓ Answer recorded
46
Fill in the Blank

If recommendations in a briefing do not clearly support the stated business objective, the briefing lacks strategic ___.

If recommendations in a briefing do not clearly support the stated business objective, the briefing lacks strategic  .
✓ Answer recorded
47
Select All That Apply

Which flaws would most weaken a 1-page strategic briefing on US AI competitiveness? Select all that apply.

  • It recommends international expansion but ignores data-localization and procurement barriers
  • It focuses on training larger models without discussing whether inference economics support the business
  • It uses a simple chart to summarize trade-offs before making a recommendation
  • It proposes a hardware-intensive path without examining supplier concentration or access risk
✓ Answer recorded
48
Single Choice

A US firm wants to respond to global competition in healthcare AI. Which recommendation is most strategically coherent?

  • ACompete head-on in every geography with one frontier model, regardless of local data rules or reimbursement differences
  • BFocus on regulated enterprise niches, optimize inference cost for recurring workloads, build hardware redundancy, and target markets with manageable compliance pathways
  • CPause product work and wait for policy clarity before making any commercial commitments
  • DOpen-source all core models immediately because openness always maximizes market reach
✓ Answer recorded
49
Fill in the Blank

A recommendation that mixes margin goals, premium-latency targets, and low-cost geographic expansion without prioritizing among them lacks a clear strategic ___.

A recommendation that mixes margin goals, premium-latency targets, and low-cost geographic expansion without prioritizing among them lacks a clear strategic  .
✓ Answer recorded
50
Select All That Apply

Based on the chart, which points should appear in the executive recommendation section of the briefing? Select all that apply.

AI Briefing Priorities by Strategic Objective Business objective Primary action Benefit Policy risk Capex Protect margins Lower inference cost High Low Med Expand abroad Multi-region compliance Med High Low Launch AI product Secure chip supply High Med High Defend market share Faster model refresh Med Med Med Win enterprise deals Audit / safety controls Med Low Low Scale usage Optimize serving stack High Low Med Use recommendations that match the stated objective while balancing inference economics, hardware supply, and policy exposure.
  • Shift from a single frontier-only offer to a tiered product strategy because the mid-tier segment is growing faster and has better margins
  • Ignore serving efficiency because benchmark rank is already competitive enough to guarantee adoption
  • Reduce single-vendor hardware dependence because supply concentration threatens execution of the growth plan
  • Sequence international expansion toward regions with lower compliance burden before entering the highest-friction market
✓ Answer recorded
0% SCORE
Assessment Complete
0
Correct
0
Incorrect
50
Total